Copyright ©1999-2000 Piotr Gąsiorowski,,

Introduction to Proto-Indo-European Stress

This essay is slightly more technical than most of the stuff on this site. The transcription system employed here is the standard one used in IE studies. There are a few phonetic transcriptions using the International Phonetic Alphabet. To view them, you must download and install the Lucida Unicode font (you can download it from essay is encoded using Unicode (UTF-8).

1. Fundamental distinctions: Stress, pitch accent, and tone

English is a typical stress language, which means that in an English word there is a pattern of strong and weak syllables. If a word is pronounced in isolation, one of its syllables is more prominent than any of the others; it is said to carry primary stress (or the main stress of the word). Other things being equal, such a syllable stands apart as being somewhat longer and louder; it is pronounced with a characteristic high or changing pitch, and its vowel has a full, distinct quality. Longer words may also contain strong syllables with less than the primary degree of stress (typically, these are not marked by pitch movement). Such syllables are required in order to achieve a balanced rhythm or to make the morhological structure of a word clearer; in such cases we speak of secondary stresses. Weak syllables are reduced phonetically, their vowels being often pronounced as ‘schwa’ (the English weak vowel [ə]). They sometimes disappear in fast speech, as when family becomes fam’ly. In many stress languages (Spanish, Polish) stress is mostly a matter of increased pitch or intensity, with little or no vowel reduction.

Note: Primary stress may be fixed – most typically on the first syllable of a word (Czech, Hungarian, Gothic, Old English), or the penultimate (last but one) syllable (Polish, Welsh). But its location may also be determined by syllable weight (Latin, Hindi) or vary according to various lexical and morphological factors (as in Russian and partly in English). Such free stress may serve to distinguish words, as in English import (noun) vs. import (verb). Stress is sometimes marked with an ‘acute’ accent (á), but phoneticians prefer a stress mark (') placed before the stressed syllable ('import vs. im'port), as stress is a feature of whole syllables rather than vowels.

Pitch accent languages are similar to stress languages in that there is (typically) a single occurrence of high pitch (H) per word. In such languages a heavy syllable (usually one containing a long pure vowel or a diphthong) may be analysed as consisting of two smaller units (called moras), and the high pitch may be linked to either half of such a syllable, producing a high-low (HL) or low-high (LH) tune. HH is in principle prohibited, since a word may only contain a single high pitch (though some languages have rules of ‘pitch spreading’ which allow an H attached to the ‘accented’ mora to span a number of moras to the right or to the left, as in Japanese). In languages that have this HL vs. LH contrast we usually call the falling tune the circumflex accent (â), and the rising tune the acute accent (á).

Warning 1: In Lithuanian, confusingly, it is the falling accent that is called ‘acute’ for historical reasons. People who wish to use Lithuanian data should make sure that they understand the accent diacritics.

Warning 2: The French ‘accents’ are orthographic devices that have nothing to do with phonetic pitch accent. Modern Greek retains the traditional accent marks but is nevertheless a stress language.

Warning 3: In Greek the accent is marked over the second element of diphthongs, but the tune it represents is that of the diphthong as a whole, e.g. in boûs the diphthong [ou] has a HL tune, and in Zeús the tune on the diphthong [eu] is LH.

Warning 4: In Classical Greek and Lithuanian a short vowel cannot support a complex tune (HL or LH). If accented, it is simply high-pitched (H). Such a vowel is nevertheless marked with an ‘acute’ (á) in Greek. Lithuanian orthography uses a ‘grave’ mark (à) for precisely the same purpose.

Note: You may have heard that Classical Greek had a third accent, known as the ‘grave’ (à). In fact, this diacritic does not represent a separate kind of tune but marks the location of a suppressed word-final acute – an H which would be there if the word were pronounced in isolation, but which is deleted in the context of a phrase in order to satisfy certain constraints on phrasal accentuation. Compare this with the rhythmic suppression of primary stress in English phrases, as on the second syllable of thirteen in phrases like thirteen men.

In tone languages pitch is divorced from stress and prominence, which means that various combination of H and L (and sometimes also M = mid) tones may occur in a single word. There are various kinds of tonality, but they fall, by and large, into two broad types. Type A (the register system, common in Africa) has polysyllabic words with usually one tone per syllable. For example, the tonal structure of a word of four syllables may be LLLL, or HHHL, or LHLH, etc. Type B (the contour system, common in the Far East) prefers monosyllabic words but allows different tonal combinations to be linked to one and the same vowel. The result is a ‘contour tone’ (a tune which falls, rises or undulates within a single syllable). Contour tones may be fairly complex, and they contrast with one another as well as with ‘level’ tones (constant high, low or mid pitch). For example Mandarin Chinese has four contrastive tones: high, falling (from high to low pitch), rising (from mid to high) and falling-rising (from mid to low and back to mid). Under a sufficiently abstract analysis these tunes may be interpreted as H, HL, LH and L. Some languages of southeast Asia introduce further complications, allowing tone to combine with various glottal features (creaky or breathy voice, or the glottal stop [ʔ]).

2. Stress in Proto-Indo-European

2.1. Static paradigms

Proto-Indo-European, as reconstructed with the help of the comparative method (I mean the most recent common ancestor of all the IE languages including Anatolian), was a stress language in which syllable strength was chiefly a matter of pitch differences and, presumably, of intensity (loudness). In this respect it was similar to Spanish or Polish, but not to English with its emphatic ‘expiratory’ stress (see Part 1). It thus stood close to the borderline between stress systems and pitch accent systems. Indeed, some linguists have attributed pitch accent contrasts to PIE on the strength of accentual correspondences between Balto-Slavic and Greek. However, scholars such as Jerzy Kuryłowicz and – more recently – Paul Kiparsky have convincingly argued that such contrasts arose independently in the branches in question.

Note 1: The best evidence for the original location of stress in PIE comes from Vedic (Classical Sanskrit developed its own stress system, similar to that of Latin). The location of pitch accent in Classical Greek (especially in Greek noun paradigms) also reflects the PIE stress pattern. There are, to be sure, some specifically Greek constraints on the distribution of pitch accents, but in the environments where such restrictions do not apply, Greek usually agrees with Vedic. In the Germanic languages the original location of stress is sometimes reconstructible thanks to the phonetic ‘fingerprints’ of Verner’s Law. Germanic spectacularly bears out the testimony of Vedic and Classical Greek. Finally, the evolution of pitch-accent systems in Balto-Slavic makes most sense if we adopt the stress system reconstructed on the basis of Vedic, Greek and Germanic as its starting-point.

PIE stress was free – not in the sense that nobody cared where it fell, but because it was determined neither by phonological factors, nor by counting syllables from the beginning or the end of a word. Its location depended on the inflectional type to which a given word belonged. PIE paradigms can be classed as static or mobile. In a static paradigm the stress of each inflected form was fixed on the same syllable of the stem, while in a mobile paradigm the stress fell on the stem in some forms, and on the inflectional ending in others.

Warning: It seems that in the vocative case of PIE nouns the main stress was invariably word-initial even in those static declensions that had paradigmatically enforced stress on some other syllable. This seems to mean that PIE vocatives had a characteristic falling intonation.

Note 2: In case you don’t remember what a stem is – it’s the form of a word that remains if you remove all inflectional affixes. As PIE had no prefixes, inflections were always added at the end of a stem. One very important class of PIE stems included those that ended in *e or *o (these vowels could alternate in various forms of the same stem), e.g. the noun stem *wlkwo- ‘wolf’ or the verb stem *bhere- ‘bear, carry’. Declensions and conjugations involving such forms are called thematic. A stem could be morphologically complex, consisting of a central root morpheme followed by one or more suffixes.

Note 3: In the examples below stress will be marked in the traditional way (á) for the sake of simplicity. Do not confuse this with the acute pitch accent (though the phonetic realisation may be the same) and bear in mind that stress is a matter of relative syllable strength.

Thematic nouns and adjectives were always of the static type; so were some common athematic paradigms, e.g. so-called es-stem neuters like *nébhes- ‘cloud’. In the table below only a didactically useful selection of PIE case forms is shown (and the dual number is ignored):



‘new’ (animate)


Nom. sg.





Acc. sg.





Gen. sg.





Loc. sg.





Voc. sg.



no vocative


Nom. pl.





Gen. pl.





Note 4: In *wĺkwo- the stressed syllable does not contain a vowel; the consonant [l] is the core of that syllable. Although syllabic [l] is common in English (cf. bottle), it doesn’t occur in stressed syllables (except, perhaps, as a variant of [ʊl], as in bull). But there are languages in which stressed syllabic consonants are commonplace. The Sanskrit word for ‘wolf’ is vŕkas (with stressed syllabic [r]), and Czech actually has vlk with stressed syllabic [l], just as in PIE!

Note 5: I don’t know if any speaker of PIE ever said ‘O yoke’ to a yoke. I suppose the potential vocative would have received initial stress if it had occurred to anyone to use it.

Note 6: Forms cited as plurals of neuter nouns were in fact collectives derived from them rather than actual plurals. The collective suffix was originally a stem-forming element, not an inflection (though of course it came to be analysed as a Nom. pl. ending in the daughter languages.

Note 7: If a stem-final *o was followed by a suffix beginning with a vowel, the two vowels coalesced yielding a long pure vowel or a diphthong, so that e.g. *wĺkwo-es was pronounced as *wĺkwo:s, and *jugó-i as *jugói (two syllables).

The stress of thematic verbs was also static. Below is the present tense of *bhére- ‘bear, carry’ (here, too, the dual number is ignored):












2.2. Mobile paradigms

2.2.1 Nouns

Mobile stress was common among nouns belonging to athematic (that is, non-thematic) classes, especially when the stem ended in a consonant or was simply identical with the root (nouns which form stems without any derivational suffixes or thematic vowels are known as root nouns). Many consonant-final suffixes (such as *-er, *-ter, *-on, *-ont) favoured stress mobility.

Informally speaking, mobile stems had two ‘states’, one of them with stress on the root syllable or a stem-forming suffix (in either case within the stem), and the other with stress on an inflectional ending. ‘State 1’ was typical of the nominative and the accusative, whereas ‘State 2’ occurred in the genitive/ablative, dative, locative and instrumental cases (the vocative, as usual, stood apart from the other cases).

In rare instances stress seems to have had two alternative locations within a single stem (see the ‘wood’ and ‘water’ words in the table), as well as shifting to the inflectonal ending in some forms.






Nom. sg.






Acc. sg.




= Nom. sg.

= Nom. sg.

Gen. sg.






Loc. sg.






Nom. pl.\coll.






Gen. pl.






Loc. pl.






Note 1: The last two words are highly irregular neuters. The ‘water’ word is a so-called heteroclitic noun: it has a stem ending in *-r in the nominative/accusative and in *-(e)n in the other cases. For the sake of simplicity the table gives only one form of each case, but sets of correspondences found in the IE languages suggest that some cases had more than one form, e.g. the Gen. sg. of *doru could also be *dorwós, the Gen. sg. of *wódr could be *wednós, *udnós, etc. The collective *wedó:r ‘waters, a large amount of water’ could also be understood as synonymous with *wódr.

It is worth observing that in mobile paradigms the location of stress has an effect on vowel quality. Unstressed syllables may have so-called zero vocalism (where, in the absence of *e or *o glides, liquids or nasals take over their function as syllabic segments). A shift of stress may also lead to syncope, so that a vowel disappears altogether (as in *drous, leaving a consonant cluster. When neither syncope nor consonant syllabification seems possible, an unstressed vowel remains but *e is preferred to *o in this position (*pedós vs. *pódm). The result is a system involving complex alternations (so that the stem *kwon- may also be realised as *kuwo:n, *kun- with a vocalised glide or *kwn- with syllabic [n]). As complex systems are difficult to learn, many of the irregularities visible in the table tended to be levelled out already in PIE times. Static paradigms, with their immobilised stress and completely predictable forms, expanded at the cost of mobile declensions. This process was similar to the spread in English of regular plurals in -(e)s: horses (Old English hors) , hares (OE haran), cows (OE cy), sons (OE suna) etc.; only a handful of odd archaic plurals remain (oxen, men, geese, deer, sheep) to show that English once had a number of declensions.

2.2.2 Verbs

Stress mobility occurred first of all in certain types of athematic verbs (stems which did not end in *e or *o). The root *gwhen- ‘strike, slay’ offers a good example of a mobile paradigm. In the present tense the stress was on the root in all the three persons singular; it shifted to the personal endings in the plural. Significantly, the root had a different phonetic form when it carried no stress: *gwhen- was reduced to *gwhn- (the nasal was syllabic if it stood between consonants):













In the mobile class we find some of the most frequently used PIE verbs, such as *hes- ‘be’, *hed- ‘eat’, *hei- ‘go, walk’. etc. Even before the fragmentation of PIE they were already becoming slightly irregular because of the vowel alternations accompanying mobile stress:













Note 2: I reconstruct a ‘laryngeal’ (or rather an aspirate just like English [h]) in this verb. If you don’t trust this reconstruction, just drop the aitches.

Another category characterised by mobile stress was the PIE perfect; here are the perfect forms of *leikw- ‘leave, abandon’:













Note 3: The perfect stem is ‘reduplicated’ (the initial *le- represents a symbolic attempt to duplicate the root syllable). In the PIE perfect the *o of the singular alternates with zero (that is, vowel reduction or deletion) in the plural.

Finally, it is worth mentioning that the ‘zero grade’ played an important role in word formation. Verbal roots often produced large families of derivatives. Some suffixes were typically stressed and required the root to which they attached to have zero grade vocalism. For example, from *leikw- we can regularly derive a deverbal adjective in *-to (*likwtós ‘left over, forsaken’, hence Latin relictus). The root *hed- ‘eat’ yields a present participle in -ont (*hdonts ‘(one who is) eating’), which is the etymological source of the mobile noun stem *hdont- ‘tooth’:


Nom. sg.


Acc. sg.


Gen. sg.


Loc. sg.


Nom. pl.


Gen. pl.


Loc. pl.


3. Internal reconstruction within PIE

3.1 Chronological layers

What I’ve written so far may sound self-contradictory. On the one hand, it is claimed that the most recent common ancestor of the IE languages associated ‘musical’ values with syllable prominence. As the static paradigms demonstrate, the contrast between stressed and unstressed syllables manifested itself primarily through pitch differences which did not influence vowel quality. On the other hand, in the mobile paradigms stress does seem to determine vowel contrasts; a vowel may undergo reduction or even disappear in the absence of stress.

This apparent inconsistency disappears if we assume that the two types of behaviour attributed to PIE vowels in grammatical paradigms belong to different chronological layers within the protolanguage. A moment’s reflection tells us that the static paradigms (such as thematic declensions and conjugations) must be historically younger than the mobile ones, otherwise vowel reductions and other phenomena conditioned by the absence of stress would have affected them as well.

If, however, such paradigms did not exist at the time when vowel reductions were taking place, it is only natural that they shouldn’t display any effect of stress on the pronunciation of vowels. Although they undoubtedly flourished in PIE (as reconstructed by the comparative method, that is by bringing together whatever evidence survives in the descendent languages), they must have been relatively young then, or at least newly restructured. They owed their brilliant career in PIE to the advantages they offered: simple, regular and highly productive patterns of declension and conjugation.

Note: A productive paradigm is one that occurs frequently and reflects the ‘normal’ rules of the language and their predictable outcome. It allows speakers to form inflected words automatically. E.g. the past tense in English is productively formed with the suffix -ed. If a newly coined or borrowed verb (say, wug) is to be inflected, a native speaker will without a moment’s hesitation form a regular past simple or past participle (wugged). Productivity versus unproductivity is sometimes a matter of degree rather than of clearcut divisions, but in the case of PIE it is obvious that the thematic formations were successful newcomers spreading at the expense of numerous unproductive types.

The regular plural (in -s) of Modern English nouns has its roots in one of the Old English masculine declensions. It has served as a magnet for nouns originally forming their plurals in other ways. Apart from a handful of miscellaneous survivals (like men, teeth, sheep, children, oxen) all countable nouns in Modern English form their plurals by adding one and the same inflection. Members of the other Old English types did not necessarily die out; they simply joined the dominant declension. For example, the plural of oak (OE a:c) is now oaks rather than*each (the expected phonetic development of OE æ:č). Oaks, unlike *each, need not be learnt separately and is at first glance recognisable as a plural form.

Note that if we had no Old English texts we would never be in a position to discover that the Old English ‘oak’ word had a suffixless plural with a changed root vowel. But we would be able to infer the existence of such a paradigmatic type from its rare survivals (man : men, tooth : teeth, goose : geese, foot : feet, mouse : mice, louse : lice). Through a careful analysis of those half a dozen surviving members we could identify it as a distinct pattern of plural formation and speculate about its origin. One thing would be obvious – that pairs like tooth : teeth are living fossils produced by processes that have been ‘overwritten’ by more recent changes in English phonology and morphology.

The technique by which we can catch glimpses of remote stages of a language using only the data surviving in that language (mainly the irregularities of its morphology) is known as internal reconstruction. On the whole it is less reliable and less precise than the comparative method because of the inevitable loss of most old forms in the language that is our sole witness. In a family of languages, a piece of information irrevocably lost in one branch will often be preserved in another, and comparison enables us to fit such pieces together. In the example above, if we were only allowed to use Modern English data and had no knowledge of the other Germanic and Indo-European languages, we would never learn that the tooth : teeth pattern represents PIE consonant-final root nouns, or that the noun tooth is etymologically related to the verb eat.

But internal reconstruction also has its advantages: it still works, to a degree, when the comparative method is helpless. The comparative method is a means of reconstructing the most recent ancestor of all the languages being compared. It cannot tell us anything of the remoter stages.

The first step in our internal reconstruction of pre-PIE stress is the observation that the mobile paradigms preserve clear traces of vowel reduction in unstressed syllables. Since this reduction can’t have been a synchronic phonetic process in PIE (even if it could be used as a morphological device), we can conclude that it belongs to a more ancient stage of the language – presumably several hundred years before the PIE unity began to split into the known branches.

In those remoter times PIE (or rather its ancestor) had a stress system in which the morphological mobility of stress was presumably much more important than in PIE proper. The nature of stress was also different: stressed syllables were first and foremost peaks of expiratory energy; vowels under stress were longer and had more distinct qualities than those in weak syllables. In the course of time phonetically reduced vowels in weak positions became ‘schwas’ or were completely lost if this kind of ultimate reduction was permitted by syllable structure. Certain consonants became syllabic as vowels standing next to them were deleted.

By the time these changes were over, the mobile paradigms had been rendered so irregular by stress-induced phonetic alternations that their learnability must have decreased. There was a demand for innovations that would make the grammatical system more rational by getting rid of some entirely redundant complications. Just then the stress system itself underwent a change, allowing full vowels to return to unstressed positions and making historically weak syllables stressable (both phenomena are visible in *wĺkwos ‘wolf’). The time was ripe for the emergence of regular static-stress pradigms.

3.2 Pre-PIE alternations

The gist of the scenario outlined in the previous section is that an original alternation of stress placement gave rise to complex vowel alternations in PIE. As the stress system of PIE continued to evolve, those alternations became separable from stress patterns while acquiring some morphological roles to play. The attested paradigms in which they occur are usually recessive (prone to remodelling after the fashion of newer and more transparent types of declension or conjugation) and often so obscured that in the grammars of the recorded IE languages they are classed as ‘exceptional’ or ‘irregular’.

Vowel reductions presuppose a stage at which the vowels of unstressed syllables were still full-timbred. We can identify such a stage in the known histories of languages which can be regarded as analogous to PIE with regard to vowel alternations. For example, the Polish alternation pies ‘dog’ (Nom. sg.) ~ psa (Gen. sg.) has developed from Proto-Slavic *pɪs-ʊ ~ *pɪs-a, where both forms were disyllabic and contained the same shape of the root (cf. Old Church Slavic pьsъ ~ pьsa). Another case in point is the French pattern represented by tiens [tjɛ̃]~ tenez [t(ə)ne] from Latin tenes ~ tenētis.

A precise reconstruction of pre-PIE is impossible for lack of comparative data that would enable us to filter out useful signals drowned in the usual noise that accumulates in the course of linguistic evolution. The inherent limitations of the method of internal reconstruction allow us to recover only a faint adumbration of what PIE may have looked like hundreds of years before it disintegrated into a number of languages.

Below is one possible historical interpretation of PIE alternations. It’s highly conjectural and I realise that a skilled Indo-Europeanist could easily invent a plausible alternative. Still, the range of permissible reconstructions is constrained by what we know about the universal properties of stress systems. In other words, while we can’t pick out a single proposal that could be considered the best, some scenarios may be eliminated as unlikely on typological grounds.

Note: In the table below no attempt is made to reconstruct pre-PIE vowel qualities with any precision (the letter a may represent any non-high vowel); nor do I wish to speculate about the origin of the e~o alternation in ‘comparative’ PIE.


‘Comparative’ PIE


pád-, pad-á-

po:ts, pedós


dajáw-, dajaw-á-

d(i)jé:us, diwós

Sky God

t(a)ráj-, t(a)raj-á-

tréjes, trijóm


dáraw-, daráw-, daraw-(a)n-á-

dóru, dróus, drunós


xwah-ánt-, xwah-ant-á-

xuhónts, xuhntós


wádar-, wadár-, wadar-á-

wódr, wedó:r, udrós


gwhán-t-, gwhan-ánt-

gwhénti, gwhnénti

to strike

háj-t-, haj-ánt-

héiti, h(i)jénti

to go

This page has been visited times since 24 Feb 2000