Laryngeals are sounds that occurred in the Proto-Indo-European
language, caused changes to neighbouring sounds, and disappeared. They were postulated to explain anomalies
in the verb system, and they proved useful for explaining other phenomena. It was many decades before confirmation of their existence was found, in the newly deciphered Hittite
language. Linguists are still not agreed on how many there were or how they were pronounced.
'Laryngeal' is also an ordinary phonetic term meaning made in the larynx; but it is preferable to call this glottal or laryngal, to avoid confusion with the still unknown values of the Indo-European consonants.
The verbs were central to the reconstruction
of the ancestor of Indo-European
languages, which was going well in the 1870s, and explanations had been found for the vowels i
, and unaccented o
. This postulated an earlier stage in which the only vowel was e
. Most verbs had a stem of the form CeC-
, the base vowel e
surrounded by consonants. Other vowels were derived from them: see ablaut
The anomalous verb stems were those that had no consonant preceding, such as *ed-, 'eat', or a long vowel and no following consonant, such as *dhee- 'put', and those that had the vowels a and o. Those with other vowels also lacked consonants, and had the same pattern as the e-stems: *ag- 'drive', *okw- 'look', *staa- 'stand', and *doo- 'give'.
If a and o had been normal vowels, why were there no normal-shaped stems CaC- and CoC-? Why were these two vowels also associated with lack of consonant? If long vowels were normal, why were there no stems of the shape eeC- or CeeC-?
The proposed solution
Ferdinand de Saussure
proposed the existence of elements he called coefficients sonantiques
in 1879, an abstract term not committing himself to any definite phonetic
form. These stood in the place normally occupied by consonant
s, and caused one or both of two changes: shifting the neighbouring vowel; and lengthening it if they followed. Then they disappeared, leaving their mark on the altered vowels.
Later the same year Hermann Möller, seeking to make a connexion between the Indo-European and Semitic families, introduced the name laryngeal, suggesting that the hypothetical coefficients were pronounced like the Semitic laryngeals. The term strictly means pronounced within the larynx, and is synonymous with glottal: Semitic languages contain h and the glottal stop. But in Indo-European the "laryngeals" might have been those, or might have been other guttural sounds such as pharyngeals or velars. We still have no clear evidence.
At least three laryngeals are usually postulated. A more abstract notation is to use schwa with a subscript number. An alternative, much more common these days, is to use some kind of H with subscript. H1 is e-coloured, H2 is a-coloured, and H3 is o-coloured. So the six roots given above come from earlier forms with laryngeals:
*ed- from H1ed-
*dhee- from dheH1-
*ag- from H2eg-
*staa- from steH2-
*okw- from H3ekw-
*doo- from deH3-
Sometimes evidence shows a laryngeal, but the vowel colouring has been lost so we can't tell which one it was: this is annotated without subscript
, as H
Proof in Hittite
of the ablaut
grades, described above, was the initial motivation for postulating laryngeals, but many other niggling exception
s in other parts of the grammar and vocabulary became simpler to explain if laryngeals were invoked. Each phenomenon could however have a different explanation. There was no direct evidence for fifty years. But in 1927 Jerzy Kuryłowicz
announced that Hittite
contained consonants in just those positions where laryngeals were predicted. This had been overlooked when Hittite was first deciphered in 1915.
Hittite was written in a cuneiform script, borrowed from the Semitic language Akkadian. By comparison with surviving Semitic languages, it is clear that Akkadian and therefore Hittite had some kind of guttural sounds, close enough to those predicted for the Indo-European laryngeals. The exact value of the Hittite sounds is unclear. There was only one letter, now transcribed h or h, but it was sometimes written doubled, as in pahhur, corresponding to Greek pyr and English fire. In some Hittite consonants, the use of doubling indicated a voicing contrast, such as p versus b, but it is not known whether this is true of h. That is, though the laryngeals were found, we're not sure how many of them Hittite preserved as distinct sounds, or what exactly they were.
It also appears to imply a fourth laryngeal, because in some cases where an a-coloured laryngeal is postulated, there is no h in Hittite, though the vowel did become a. Other instances of H2e do however show up as ha.
The Hittite verb system is rather different from those of the other branches. Verbs belong to either the hi-class or the mi-class, these being the first person singular endings of the present tense. In other IE languages, such as Greek, the mi-class is very reduced, and almost all verbs have present tenses ending in long -o (or its later developments). But in the perfect tense of Greek, we find the ending -a. This comes from a laryngeal, as does the Hittite -hi. So either the laryngeal h was originally present in both tenses, and was later displaced by new formations, or it spread in Hittite, replacing older endings.
the Proto-Indo-European (PIE) vowels e
both became a
, wiping out the basic evidence for vowel-colouring; but laryngeals are convincing as explanation of at least three disparate phenomena restricted to the Sanskrit branch. They are the difference between set
stems; the exceptions to Brugmann's Law
; and the existence of voiceless aspirated
Set and anit are ancient Indian grammatical terms, from sa-i 'with i' and an-i 'without i', referring to the infinitive. Some are like bhavitum 'to become', from the root bhu, and others are like kartum 'to do', from kr. Those with i come from roots extended by a laryngeal: bheweH-. When the unaccented vowel was reduced to zero grade between consonants, giving *bhéwHtum, the laryngeal found itself between consonants, so turned into a vowel, in order to remain pronounceable. In most IE languages this so-called schwa indogermanicum became a, but it is characteristic of Sanskrit that it became i.
The presumed laryngeal also triggered vowel lengthening. With the past participle suffix -tós the accent is removed from the stem, and both stem vowels are reduced to zero, giving *bhwHtós. In this position the w then turns into a vowel, *bhuHtós, and then lengthening and loss of the laryngeal (as in the original ablaut pattern laryngeals were invented for) give the actual Sanskrit bhuutás.
In later (Classical) Sanskrit the range of set roots was extended by analogy, so not all set roots mark laryngeals. All languages undergo extensive ploughing-over and early patterns are often obscured.
Normally PIE o became Sanskrit a. Brugmann's Law says that when this happened in an open syllable before a resonant (a liquid or nasal), it was also lengthened. So *kekore 's/he made' regularly becomes cakaara. Comparison with Greek shows that the first person 'I made' should have been *kekora, which should have also become cakaara, but the actual form is cakara. Something prevented Brugmann's Law operating. It might have been an analogy with some other 'I' form, in order to keep the two persons distinct; but the unusual thing about the perfect tense ending -a 'I' is that, while the vowel a is common in PIE stems, it's very rare in inflections. This suggests it only occurred where created by a laryngeal. Then the PIE form *kekorH2e would contain a closed syllable -kor-, not open -ko-, so Brugmann doesn't apply.
The third major strand from Sanskrit evidence is the voiceless aspirates ph th th ch kh. They were long regarded as part of the original PIE consonant set, but actually evidence for them outside the Indo-Iranian branch including Sanskrit is slim. Kuryłowicz proposed that they developed from a plain stop p t t c k that had come to be next to a laryngeal when the intervening vowel dropped out. So the Sanskrit root stha- 'stand' came from steH2- and the new consonant th was then generalized to all forms of the word.
Greek and Armenian evidence
is a solitary branch of the Indo-European
family, markedly different from all others, but in several respects it and Greek
share unusual features. Some words, particularly in front of n l r
, have an extra (prothetic
) vowel compared to other branches.
Greek ónoma, Armenian anun, cf. English name, Latin nomen, Sanskrit naaman.
Greek odoús, Armenian atamn, cf. English tooth, Latin dens, Sanskrit dantah.
Greek astêr, Armenian astł, English star, Latin stella.
Greek eleútheros, Latin liber 'free'.
From the vowel preserved in Greek we can see which laryngeal was originally there: H1leudh- 'free', H2ster- 'star', H3nom- 'name'.
Greek and Sanskrit share the augment, an initial vowel e- (becoming a- in Sanskrit) on some past tenses. Armenian also has this, though only on the third personal singular of monosyllables. They also have reduplication in the perfect tense. The pluperfect is formed with both augment then reduplication. Where there was a laryngeal, this sometimes causes lengthening. So H1leudh- 'free' gives eleeluutha 'I loosened', from e-H1le-H1leudh-H2e. In a few cases the vowel-colouring effect of the laryngeal is also preserved.
The lengthening effect seems to explain the two Greek words ikhthuus 'fish' and muus 'mouse'. We can suspect that there was a laryngeal here, because the Armenian words are jukn and mukn. This also shows that it's not strictly true that the laryngeals disappeared from all modern languages.
There are more direct survivals in Armenian:
Latin anus = Hittite hannas = Armenian han
Latin avus = Hittite huhhas = Armenian haw
Latin ventus = Hittite huwantsa = Armenian hogi
is another unique and distant branch of Indo-European
, and also preserves some actual descendants of the laryngeals, in words like hidhur
'empty', and herdhe
Some Latin perfect tenses are formed by reduplication, but others are formed by lengthening the vowel: these may be in places where laryngeals are predicted. So edō 'I eat', ēdī 'I ate', from originally reduplicated H1e-H1d-.
Etruscan is not an Indo-European language, but might be distantly related to the family as a whole. The word for 'before' is hant-, whereas Latin has ante (and English and and answer are related).
The attempt to link Indo-European to other families, called the Nostratic theory, has a problem with laryngeals. Although Nostratic would connect Indo-European to Semitic, the connexion is not very close even within Nostratic. All the other groups that would belong somewhere in the superfamily - such as Kartvelian, Uralic, Altaic, and Dravidian - show little or no evidence of ever having had laryngeals.
The Nostraticist Allen Bomhard
writes that we can now state with complete confidence the values of the laryngeals. This of course means we can't do anything of the kind. They could be almost anything.
One good idea is that H1, H2, and H3 were respectively palatal, velar, and rounded velar fricatives, that is [ç], [x], and [xw]. This fits the pattern of stops [kj k kw] that we already know about, and also makes them quite easy for us to pronounce. None of those awkward pharyngeals. It's also a good idea because I thought of it by myself, though quickly discovered that Gamkrelidze and Ivanov, authors of the glottalic theory, had got there before me.
But H1 was more easily lost, so perhaps it was something weaker, like a glottal stop. H4 if it existed could have been [h]. H2 and H3 might have been pharyngeal. H3 might have been voiced (because of the Hittite writing); as it caused o-colouring it was very likely rounded (labialized). Among these possibilities you are free to choose, to suit your taste.
I benefited enormously from discussions on the Indo-European internet mailing list, and am particularly grateful to Peter Gray and Miguel Carrasquer Vidal for their explanations.
Among books I have consulted are
L.R. Palmer, 1980, The Greek Language, Faber & Faber
Winfred Lehmann, 1993, Theoretical Bases of Indo-European Linguistics, Routledge