4.2. Consonants of Proto-Indo-European, part 1: phonation

The consonant phoneme inventory of Proto-Indo-European, as usually reconstructed:

The IPA transcriptions in the table above are based upon their manifestations in the surving daughter languages, as exemplified in SoundCorrespondences.html and discussed in detail below.

Stops: Voicing and aspiration

The system of stops reconstructed for Proto-Indo-European is typologically unusual, as it exhibits a three-way system of phonatory contrasts between voiceless (e.g. *t), voiced (e.g. *d) and voiced aspirates (e.g. *dʰ). For example:

d vs dʰ

d vs t

gʰ vs g vs k

*leug‑ > English lock

*leugʰ‑ > English (tell a) lie

While a contrast between voiced and voiceless consonants is extremely common, in systems of three-way phonation contrasts the third term is usually voiceless aspirated stops, as in e.g. Ancient Greek δ/τ/θ (i.e. [d] vs. [tˉ] vs. [tʰ]) or Thai [d] vs. [tˉ] vs. [tʰ]. Lisker and Abramson (1970) characterise such a three-way contrast in terms of regions along a continuum of voice onset time (VOT), that is, the time interval between the onset of voicing with respect to the release of the stop closure.

Figure 1. Vertical bars: Distribution of voice onset time in alveolar plosives as measured in a database of sound recordings of Thai. Connected lines (see scale of y axis): percentage of identification judgements by listeners in a forced-choice auditory identification task. The x axis is voice onset time, in 10 ms increments, with respect to release of the stop closure which is set at 0 ms.  From Lisker and Abramson (1970: 567, figure 3).

In fully voiced stops, voicing commences before the release of closure, so that VOT is negative (roughly, anything in the range from −150 ms to −30 ms); this is also known as prevoicing. The dotted vertical bars towards the left of figure 1 in the range from −110 ms to −50 ms plot tokens of this kind, from Thai. The continuous dotted line in the upper left of the figure indicates that stops with prevoicing up to c. −30 ms are mostly classified by listeners as being voiced stops [d], on over 80% of trials. Stops with prevoicing from −30 ms to −10 ms are more uncertainly or ambiguously identified by listeners as [d] or [tˉ] (transcribed as just [t] in the legend to Figure 1), with a 50%−50% breakdown in identification for tokens with VOT in the region from −10 ms to 0 ms.

In voiceless unaspirated stops, voicing commences more or less around the time at which the stop is released, or slightly later. In figure 1, the VOT of voiceless unaspirated [t] is plotted as solid vertical bars from 0 to about 20 ms. The highest such bar is at 0 ms, and the second highest at 10 ms, which shows that the onset of voicing in such voiceless unaspirated stops in Thai is synchronized quite tightly with the release of stop closure. The vertical dashed bars to the right of the figure, from around +30 ms to +150 ms, are VOT measurements in tokens of voiceless aspirated [tʰ].

The prominent peak in the continuous solid line shows that stops with VOT in the range from 0 to about 40 ms are most likely to be identifed as voiceless unaspirated [tˉ]; from over 80% of tokens with VOT in the range 10 ms to 30 ms were so identified by Lisker and Abramson's experiment subjects. The perceptual boundary or “cross-over” between [tˉ] and [tʰ] responses is around 45 ms.

Voice onset time and aspiration in Sanskrit, Ancient Greek, and Proto-Indo-European

Voiced aspirated stops were not included in Lisker and Abramson's comparative study, so in order to use such results to try to understand the Proto-Indo-European stop system, further data is needed. Although I don't have perceptual data concerning voiced aspirates to present here, we do have audio recordings of modern spoken Sanskrit and other modern languages with voiced aspirates to examine acoustically.

In Lisker and Abramson's sample, and indeed in most languages whose phonology has been studied so far, voice onset time and aspiration are not independent variables or features; a few different patterns of interdependence are common.

A) Languages with a two-way contrast between early vs. later VOT.

[d] vs. [tˉ]; i.e. prevoiced vs. voiceless unaspirated as in, for example Spanish, French, or Italian.

[d] vs. [tʰ]; this is the pattern of contrast in word-medial position in English (e.g. bidder vs. bitter). In this kind of contrast, voicing and aspiration are completely correlated: voiced implies unaspirated, and vice-versa, and voiceless implies aspirated, and vice-versa. There is therefore a (useful) redundancy in the signal, and it is not easy to determine whether voicing or aspiration is the “primary” phonetic cue to the contrast, nor even whether one of these features is primary and the other secondary.

B) Languages with a three-way contrast, e.g. Ancient Greek δ  vs. τ vs. θ or Thai [d] vs. [tˉ] vs. [tʰ]. In these languages, both aspiration and voicing are distinctive, but they are not completely independent or uncorrelated, because aspirated stops [tʰ] are also (predictably) voiceless.

Sanskrit and various modern Indic languages exhibit the fourth possible combination of voicing with aspiration. Since voicing commences before the release of the stop, and continues into a following vowel, the aspiration interval is simultaneous with voicing; such aspiration is typically breathy voicing [ʱ].

In terms of this kind of phonological tabulation, 3-way contrasts can be understood as a system in which one of the possible combinations of voicing and aspiration is not exhibited; commonly, it is the voiced aspirates that are absent from the system, as in e.g. Ancient Greek, Thai. Note that although the Greek letters φ θ χ are nowadays typically read as voiceless fricatives [f], [θ] and [x], that reflects a later (Byzantine and Modern Greek) pronunciation. In Ancient Greek they were voiceless aspirated plosives, not fricatives.

According to the above pattern, voiced aspirates are only expected in 4-way (2 × 2) systems in which all 4 combinations of voicing and aspiration occur.

The phonological system that is usually reconstructed for Proto-Indo-European pattern is atypical in that it is believed to have had voiced aspirates in a 3-way system of contrasts, i.e. it apparently lacked voiceless aspirated stops, even though they are cross-linguistically more common than voiced aspirated stops. For example:

Motivation for reconstructing voiced aspirates in Proto-Indo-European

Given the atypicality of voiced aspirates in a 3-way system of phonatory contrasts, I shall try to explain why the conventional reconstruction of [tˉ] vs. [d] vs. [dʱ] is proposed. Then, in later sections, I shall consider the role of phonological neutralization of opposition as an aid to understanding the system of contrasts. Then I shall present an enriched and more universal phonetic framework for phonation contrasts, in which to situate data from a wide range of modern Indo-European languages. This will lead us to a more complete picture of the range of fine-grained pronunciation differences observed across the language family, including some previously unrecognised and undocumented combinatorial possibilities.

Aspirated stops are found in various cognate words in Ancient Greek and Sanskrit. Although Sanskrit does have voiceless aspirated stops, the Sanskrit cognates of Ancient Greek voiceless aspirated stops are usually voiced:

Since such words have aspirated stops in both Ancient Greek and Sanskrit, it is simplest to infer that their Proto-Indo-European ancestors also had aspirated stops. But were they voiced aspirates or voiceless aspirates? If they were voiced aspirates, why or how did they become voiceless in e.g. Ancient Greek? If they were voiceless aspirates, why or how did they become voiced aspirates in Sanskrit?

Note that in the table above, the English (Germanic) words also have voiced consonants, and voiced consonants are also found in cognates in languages in many or most other branches, e.g. Lithuanian (Baltic), Bosnian (Slavic), Armenian, Albanian. Therefore, the PIE ancestors of these words were also probably voiced. Combined with the evidence of aspiration in Ancient Greek and Sanskrit, this implies that the PIE consonants in these words were voiced aspirates; they remained voiced aspirates in Sanskrit, but lost their voicing in Ancient Greek and lost their aspiration in other languages.

In Lisker and Abramson's cross-linguistic study of voice onset time, voicing and aspiration are not totally independent features: negative VOT yields voiced unaspirated stops, VOT around 0 yields voiceless unaspirated stops, and positive VOT yields voiceless aspirated stops. Voiced aspirated stops don't fit into that scheme very well. But if we consider VOT contrasts and aspiration as separate, independent features, 5 possible categories arise (not 6, because there is no categorical difference between voiceless unaspirated stops with VOT c. 0 ms and those with later VOT's: they are all voiceless and unaspirated.

As well as accommodating voiced aspirated stops such as [bʱ], this more fine-grained scheme also allows for the existence of stops with voiceless closure, with voicing commencing upon release of the stop and at the same time as aspiration, i.e. [pʱ]. Such consonants have been almost completely unnoticed until now, and phonetic terminology lacks a name for them; they might be called “breathy-aspirated voiceless stops”.

Detailed examples of phonation types in Indo-European

A wide range of variations in voicing and aspiration found in contemporary recordings of Indo-European languages is exemplified in more detail in the following waveforms and spectrograms.

[b] shows prevoicing, but no aspiration:

Anglo-Saxon bēon “to be”

Latvian būt “to be”

Balochi بو [bu]

Albanian botë [botʰ]

[b̥] (equivalently [pˉ]) has voiceless closure (i.e. no prevoicing) and no aspiration:

Irish bí

Modern English be. Although spelled with the letter b (which denotes a voiced stop in many languages) this plosive is not usually voiced in English, especially in initial position. As here, it is typically produced with a voiceless unaspirated stop.

[bʱ] has prevoicing and (breathy-)voiced aspiration:

Sanskrit भू  [bʱu:], with an interval of voicing+aspiration

Armenian բոյս ‎[bʱuĭs], with a brief interval of voicing+aspiration:

[pʱ] has voiceless closure, with aspiration beginning at the same time as voicing. This is very similar to the voiceless unaspirated plosives in English be and Irish bí, above, but is slightly different from those.

Eastern Armenian բերք [pʱer̥kʰ] “harvest” (cognate with “to bear” < PIE *bher-), with some voicing during the brief (13 ms) interval of aspiration. Contrast this example with Armenian բոյս ‎[bʱuĭs] immediately above, in which there is voicing during the stop closure.

Welsh bod [pʱotʰ] “be”:

Western Armenian բերք [pʰer̥kʰ] “harvest”, with brief voiceless aspiration for 15 ms:

[pʰ] and [ph] have no prevoicing, and a briefer or longer interval of voiceless aspiration, respectively.

English path, with a voiceless aspiration interval of c. 45 ms:

Ancient Greek φυσικός [pʰʉsikós], with a voiceless aspiration interval of c. 36 ms proceeding to a c. 26 ms interval of voicing+aspiration:

The contrast between unaspirated, brief and long aspiration intervals can be important, as in Ancient Greek /g/ [kˉ] γέρανος [kˉeranos] “crane” vs. /k/ [kʰ] καρπός ‎ [kʰarpós] “harvest” vs. /kʰ/ [kh] χόρτος  [khɔrtos] “garden”. Whether these pronunciations are accurate for Ancient Greek is of course unknown; scholarly opinion (e.g. Allen 1968) based on a variety of phonological clues holds that γ /g/ was pronounced as a voiced and likely unaspirated stop [g], κ /k/ was pronounced [kˉ], and χ /kʰ/ as [kh].

Unaspirated [k] in γέρανος [kˉeranos] “crane”. In this token, the slight interval of frication following release of the initial plosive is probably velar friction, not aspiration. The recording is of a male speaker from Spain who has contributed a large number of recordings of Ancient Greek to forvo.com.

Brief aspiration in [kʰ] καρπός [kʰarpós] “fruit, grain”. This word survives into Modern Greek with no change in spelling, and the recording is of a native speaker of Modern Greek. The aspiration interval is of c. 20 ms.

Long aspiration [kh] χόρτος [khɔrtos] “garden”. The aspiration duration is c. 100 ms, over twice as long as the 45 ms aspiration in English path, above, and much longer than any of the previous examples of aspiration. This might be a deliberately slow, “hypercorrect” pronunciation, as this pronunciation does not survive into Modern Greek, and therefore a strongly aspirated [kh] is rather unfamiliar to many contemporary learners of Ancient Greek. It is important to teachers and learners to distinguish χ [kh] from κ which, as seen above, may be pronounced [kʰ], with slight aspiration. This strongly aspirated ancient pronunciation developed into the voiceless, velar fricative [x], as in Modern Greek χόρτος [xɔrtʰʊs] “garden”.