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

11. The Indo-Iranian Voiceless Aspirates

11.1. Evidence for the origin of the voiceless aspirates in voiceless stop plus laryngeal

One of the most generally accepted supports for the laryngeal theory is the analysis of the Ind.-Ir. voiceless aspirates, ph, th, kh, as reflexes of voiceless stop plus laryngeal. This analysis was first made by Saussure. In a paper read in 1891 he suggested that the Ind.-Ir. voiceless aspirates, ph, th, kh, developed from PIE voiceless non-aspirates plus laryngeals, e.g. Skt. pr̥thú ‘large’ from /pltXws/ [pl̥tXus]. The evidence has been reexamined and enlarged by Pedersen, Kurylowicz, Messing, and Sturtevant.1 Especially Kurylowicz has provided phonological and morphological evidence in favor of this origin of the Ind.-Ir. voiceless aspirates.

The phonological evidence is found in words with ph, th, kh that have cognates in which the ph, th, kh is followed by an ‘original long vowel.’ Thus Gk. ἔστην is cognate with Skt. tíṣṭhati; both are from a PIE root /steA/, cf. Skt. Aor. 3d sg. ásthāt and pret. ptc. sthitá. The th presumably spread throughout all forms of the verb from those forms in which the root vowel was lost and the laryngeal stood directly after the stop, e.g. the 3d sg. /te-(s)tA-e-ty/.

As morphological evidence in favor of this explanation Kurylowicz pointed out that Skt. verbs of the ninth class whose root ends in a voiceless stop have an aspirated voiceless stop, never an unaspirated stop. The PIE base of verbs in this class ended in a laryngeal, e.g. Skt. puṇā́ti < /pw-n-eX-ty/. Skt. ninth class verbs with voiceless stop are : mathnā́ti ‘shakes’, grathnā́ti ‘ties’, and śrathnīté ‘slackens’. Moreover suffixes with aspirated stop are found alongside suffixes with non-aspirated stop and long vowel, e.g. Skt. -tā-, -tha-; see further EI 46-50.

In support of the theory that voiceless aspirates developed from clusters of stop plus a consonant which has disappeared it has been observed that voiceless aspirates alternate with voiceless stops. Thus, beside Gk. πλάθανον ‘dish’, Skt. pr̥thú ‘broad’, we find a voiceless stop τ rather than θ in Gk. πλατύς ‘broad’. If voiceless aspirates had developed from one PIE phoneme, the irregularities like Gk. πλατύς are difficult to explain; if they developed from a cluster, which in some ablauting forms was separated by a vowel, e.g. /plteXw-/ : /pltXew-/, the reason for the interchange of stop and aspirate is apparent. Another such contrast is found in Av. nom. sg. pantā ‘way’, gen. sg. paθō, with t in the strong cases, θ in the weak. Aspirates would have developed only in the inflected forms in which the laryngeal was contiguous with the stop; they were generalized in one set of forms, lost in another.

Another reason for assuming the secondary development of ph, th, kh is their limited occurrence. ph, th, kh are rare in Ind.-Ir.; moreover they are restricted to certain phonetic environments; none is found before r. Consequently there is strong evidence that ph, th, kh originated in p, t, k plus laryngeal.

11.2. The problem of the time of origin

The time of origin of the voiceless aspirates, however, is disputed. Kurylowicz considered them Ind.-Ir. developments. Sturtevant, like Brugmann, Hirt, and most other Indo-Europeanists since Grassmann, ascribes them to PIE and assumes a four-fold series of PIE stops: p ph b bh, etc., of which ph, th, kh (kʷh), have no distinct reflexes in any dialect but Ind.-Ir.

11.2a. Kurylowicz' chief arguments are methodological. He concluded from assimilation phenomena that in PIE there was no contrast of voicing between bh dh gh, and p t k, that is, that from the point of view of voicing bh dh gh were as neutral as the resonants. Thus, there are no PIE roots with /bh dh gh gʷh/ and /p t k kʷ/ in either sequence; roots like bhet, tebh are not found. Furthermore, noting that PIE bh dh gh plus voiceless stop, e.g. t became bdh, ddh, gdh in PIE or early Ind.-Ir. (Bartholomae's Law), Kurylowicz argues against the assumption of a PIE four-stop system; if PIE had had a four-stop system Kurylowicz would expect Ind.-Ir. pth or pht, etc. At the time of the Ind.-Ir. dissimilation of aspirates, however, the contrast in voicing was established, as indicated by the dissimilation of PIE /bheudh-/ to Skt. bódh-ati. Kurylowicz concludes that only Ind.-Ir., not PIE, had a four-stop system. This argument is not wholly convincing. For we cannot formulate the allophones of phonemes solely from combinatory or assimilatory changes. Bartholomae's Law is a description of an assimilatory change, progressive in both voicing and aspiration. Nothing in the structure of the PIE phonemic system would lead us to expect a regressive assimilation, such as pt or some other cluster, possibly pth from bh plus t. In a similar morphological structure, /wyd-/ plus /-to-/, we find regressive assimilation, with a distinctive PIE cluster (/tt/ [tst]), see 2.1d. From this cluster we can draw no conclusions about the allophones of PIE /d/ and /t/, except those in this particular cluster. Bartholomae's Law likewise gives us only an indication of the allophones of PIE /bh/ etc. and /t/ in a particular environment.

Kurylowicz is more convincing when he advances a reason for the development of PIE /pA tA kA/ to separate phonemes only in Indo-Iranian. Only in Ind.-Ir. did an allophone of one of the other obstruents develop with which /pA tA kA/ fell together. Siebs, KZ 37.293, suggested that in Ind.-Ir. /bh dh gh/ had an unvoiced allophone after /s/, [spʽ  stʽ  skʽ ]; in Greek, /bh dh gh/ on the other hand became φ θ χ after /s/ as well as in other phonetic environments. The Ind.-Ir. phonemes /ph th kh/ resulted from the coalescence of this allophone of /bh dh gh/ and the reflex of the cluster /pA tA kA/.

11.2b. Sturtevant has supported his contention that PIE had the phonemes ph, th, kh by suggesting reflexes of such phonemes in other IE dialects, especially Latin, which differ from those of bh, dh, gh or p, t, k. In some dialects no such differences can be found; thus in Gmc. our evidence points to the same development for /ph th kh/ and /p t k/; in Gk. /ph th kh/ fell together with /bh dh gh/. Sturtevant has published the developments of the hypothetical IE ph th kh in the various dialects, and has added formulae for their development in Latin. In support of these formulae he has proposed new etymologies, connecting: Lat. hāmus ‘fish-hook’ with Gk. χαμός; Lat. congius with Skt. śan̄khá ‘shell’; Lat. radius ‘spoke’ with Lat. rota ‘wheel’, Skt. rátha ‘chariot’; Lat. -idus with Skt. -átha; Lat. vibrāre ‘vibrate’ with Skt. vyáthate ‘trembles’; Lat. folium ‘leaf’ with Skt. phála ‘fruit’; Lat. lambō ‘lick’ with Gk. λαφύσσω ‘devour’. Other etymologies, though possibly weak, had been proposed for these Latin words.2 But even if those proposed by Sturtevant were as convincing as the etymologies in older works, it is difficult to establish phonetic formulae for such a small group of words as those in Latin with reflexes of ‘voiceless aspirates.’ Furthermore it need not follow that PIE had the phonemes, ph th kh, even though Sturtevant's etymologies are accepted. For the Latin and Armenian developments, and Slavic x, may as plausibly be derived from PIE clusters of voiceless stop /p t k/ plus laryngeal as from PIE voiceless aspirated stops, ph th kh.

Since we find unambiguous development of stop plus laryngeal only in Ind.-Ir., it is there that we will most probably find our best evidence for assuming the time of origin of ph, th, kh, whether IE or Ind.-Ir.

11.3. Origin of voiceless aspirates in Indo-Iranian

A test for the time of origin is the development of kh before Ind.-Ir. front vowels. In Ind.-Ir., velar stop phonemes were palatalized when they stood before reflexes of PIE /y ī e ē/, as illustrated in the examples cited below. In such environments there is, however, no evidence for palatalization of Skt. kh; Skt. kh is preserved both before Skt. y and before a from PIE /e/.

We find Skt. kh before y in khyāti ‘sees’ and other forms of the root khyā-; the desiderative, however, is cikhyāsita. Here the reduplicated syllable had lost aspiration and the k was palatalized to c, presumably as a later analogical development; for the perfect cakhyau there is likewise palatalization of k, but not of kh. Palatalization of g- from PIE /gʷ-/, however, is found in jyāsyati, the future of jyā- ‘overpower’.

Before a from PIE /e/ we find palatalization in járati ‘wastes away’ but not in skhalati ‘stumbles’; in Skt. dáhati, Av. dažaiti ‘burns’ from PIE /dhegʷh-/ but not in Skt. rikháti ‘scratches’; in the causative form of Skt. arc- ‘shine’ from PIE /erk-/, arcayati, but not in the causative of īn̄kh- ‘swing’, īn̄kháyati.

To be sure, we find palatalization in some Av. forms cognate with Skt. sákhi ‘friend’, if not in Skt. The Av. nom. sg. is haxa; x would be the regular development of Av. k before consonants, cf. xrūrəm ‘bloody’, Skt. kravís; I assume that x is thus the regular reflex of k before laryngeals. In the oblique cases the Av. forms are secondary, e.g. in the Inst. sg. haša; š is the regular development of palatalized k before y, i, cf. Av. vašyate: Skt. ucyáte; the laryngeal was apparently lost here and the k then palatalized.

I conclude from the absence of palatalization of /kh/ that /kh/ was not a Skt. phoneme at the time of secondary palatalization, and consequently not a phoneme in PIE. If /kh/ had been a PIE phoneme, it like PIE /gh/ should have developed to an Ind.-Ir. unit phoneme, and like /k g gh/ should have undergone secondary palatalization in Ind.-Ir. I assume that the other voiceless aspirated stops too became phonemes only after the time of secondary palatalization; since secondary palatalization is later than the time of the dissimilation described in Grassmann's Law, /ph th kh/ became phonemic relatively late in the Ind.-Ir. period.

The assumption that the laryngeals survived after stops in Ind.-Ir. may be supported by irregularities in the dissimilation of aspirates. By Bartholomae's Law the second of two contiguous stops is aspirated in composition of aspirated and unaspirated stop, e.g. viddhá from vidh-to-, buddhá from bhudh-to-, etc.; this change occurred very early, presumably in PIE. If the preceding syllable began with an aspirate, this aspiration is lost, as in buddhá; this change occurred in Plnd.-Ir. Notable exceptions are dhattá, the 2d pl. imper. pres. of dhā- ‘place’, and other forms of this root, see Whitney 160. I assume that the process described in Bartholomae's Law was here inhibited by the laryngeal of the root; dhatta is from PIE /dhe-dheX-te/; the laryngeal survived here after the change described in Bartholomae's Law, /dhe-dhX-te/ and upon its loss the cluster developed to tt. In roots without such laryngeal, e.g. /bheudh-/, the change described in Bartholomae's Law was not checked.

The palatalization of Skt. /k/ likewise was prevented before laryngeals. The laryngeals here functioned as in Hittite, where they had prevented palatalization of /t/, cf. Hitt. ti-it-ti-an-za (3.4).

In the other IE dialects voiceless stop plus laryngeal failed to produce distinct phonemes. Although there is great variation in development, in general we may note that : in dialects which maintained PIE /bh dh gh gʷh/ as aspirates, e.g. Greek, or as spirants, e.g. Latin, clusters of voiceless stop, /p t k/, plus laryngeal fell together with reflexes of PIE /bh dh gh gʷh/; in dialects in which the reflexes of PIE /bh dh gh gʷh/ are not spirants or aspirates but rather voiced stops, e.g. Baltic, clusters of /p t k kʷ/ plus laryngeal fell together with reflexes of PIE /p t k kʷ/. Even in the latter dialects there is evidence for a persistence of a spirant increment; in the Gmc. dialects the reflexes of PIE /p t k kʷ/, whether or, not followed by laryngeal, are spirants; in Slavic, the reflex of the voiceless velar stop followed by laryngeal fell together with a spirant, x. From these developments I conclude that /p t k kʷ/ plus reflexes of laryngeals survived into the various dialects as complex aspirated clusters.

11.4. Voiceless aspirates from voiceless stops plus PIE /h/

How many of the laryngeals combined with voiceless stops in such clusters has been disputed. Kurylowicz admitted such development only for /h/, Sturtevant for any of the voiceless laryngeals.3 Yet most of Sturtevant's examples with identifiable laryngeals have /h/; his only example for /x/ is a new etymology, Lat. mūtāre, Skt. methati. Sturtevant further suggests that the Gk. aspirated perfects owe their aspiration in part to the 1st sg. ending -xa, IHL 84. This suggestion has not been widely accepted, see Lang. 19.166, SSP 200-2; and in deciding that /x/ had not caused aspiration, EI 254, Kurylowicz had pointed out that before the Skt. ending -a from -xa stops are not aspirated. Sturtevant's evidence for assuming aspiration before /ʔ/ is similarly scanty. Unless more evidence is cited in favor of the development of voiceless aspirated stops from /p t k/ plus any voiceless laryngeal, I follow Kurylowicz' formulation that PIE /p t k/ combined only with /h/ to yield aspirated clusters which in Ind.-Ir. became separate phonemes.4

I assume then that the laryngeal /h/ survived into the dialects after the PIE voiceless stops. In all dialects but Ind.-Ir. the clusters of voiceless stops plus /h/ did not become phonemes; we have, however, in the patterns of development of these clusters in other dialects some evidence for PIE clusters with aspiration. These clusters became separate phonemes only in Ind.-Ir. presumably after merging with allophones of PIE /bh dh gh/. Phonological developments in Ind.-Ir., that is, the absence of palatalization of /kh/, and the variety of development of stop plus laryngeal clusters in other dialects support the conclusion that /ph th kh/ were phonemes only in Ind.-Ir., not PIE.


1 H. Pedersen, IF 5.49-51, 56, 64, KZ 40.178, has dealt with the Slavic reflexes of these clusters; he touches on the general subject briefly in La cinquième déclinaison latine 48. J. Kurylowicz, EI 46-55 and 254-5; Messing SSP 180-4; Sturtevant, IHL 83-5, Lang. 17.1-11 The Indo-European Voiceless Aspirates. See also J. Wackernagel, Altindische Grammatik 1.118-23 (1896).

Some Indo-Europeanists assume also a k̑h and kʷh, but the evidence for these is extremely meager. Apparently they were assumed in order to fill out the system. Since we have evidence for only one velar voiceless aspirate, I write only kh.

2 Messing discussed Sturtevant's etymologies at some length in SSP 180-4.

3 EI 254-5; Sturtevant, IHL 83-6, Lang. 17.1-11.

4 If we admit aspiration only by /h/ Couvreur's objection to Kurylowicz' analysis of the b in Skt. píbati ‘drinks’ is invalid. Kurylowicz, EI 54-5, ascribed the voicing of b to the presence of a following voiced laryngeal, /γ/. Couvreur, Hett. . 300, finding that Kurylowicz had suggested /γ/ as a voiced counterpart of /x/, expected also an aspirated voiced stop in píbati. If /x/ did not aspirate preceding voiceless stops, Couvreur's objection is of course invalid.