This is the second part of a four-part series. It offers an examination of the attested evidence of Celtic prosody.
In the last installment of this series, we covered the basics of what dīkantlom ‘prosody’ is, and explained why a language’s accent is important to speech and versification. We also touched briefly on the different accent styles that are found in Proto-Celtic’s Indo-European relatives and neighbors. Now, for this installment, we will dive deeper into examining what kind of accents existed throughout the Celtic language family to determine what kind of accent Proto-Celtic itself might have had. This way, we can make decisions for how to speak, write, and sing in this ancient tongue.
Note that pronuncation guides below are written in the International Phonetic Alphabet (IPA). In IPA ⟨ˈ⟩ before a syllable indicates primary stress and ⟨ˌ⟩ indicates secondary stress.
Modern Irish (Gaeilge) and Scottish Gaelic (Gàidhlig) predominantly feature an initial stress accent. This means that the first syllable of a word is stressed, even if the first syllable of a noun or adjective is a prefix. Irish is also known to apply stress to each root in a compound word1 (e.g., seanbhean /ˈʃanˌvʲan/), although Gaelic typically stresses only the first one.2,3
Scottish Gaelic, on the other hand, has developed a secondary accent based on pitch. This pitch accent is traditionally assumed to be borrowed from Norse influence, although some have argued that it could have developed independently as an inherent part of the prosodic structure after certain sound changes occurred.4 Gaelic sometimes also features a secondary stress in dialects where svarabhakti developed in certain words.2 However, neither of these later Gaelic developments are consequential to the history of Proto-Celtic accentuation.
Another variation in Goidelic accents occurs in both Southern Irish dialects and Manx (Gaelg). Here, syllables with long vowels may attract the stress accent away from a word’s first syllable.2,5
Historically, there were no long vowels that could attract stress away from the initial syllable like this, because the initial stress accent of Old Irish (Goidelc) had had the effect of shortening unstressed vowels. But later, various sound changes as well as foreign borrowings conspired to produce a number of words that have these long, non-initial vowels.5
There is some debate over whether or not this dialectal variant is a result of Norman influence or if it could have occurred independently after certain sound changes in the language. The Norman theory seems strong based on the chronology and geographical extent of this accent change both aligning with the Norman conquest.2
Returning to the predominant initial stress accent in the Goidelic branch, this has been present since Old Irish times. However compound verbs and adverbs in Old Irish received stress on the first syllable of the second element rather than necessarily stressing the prefix as would be the case for an adjective or noun.6 Example:
This seems to remain the case for some verbs and adverbs in modern Irish and Gaelic (e.g. atá /əˈt̪ˠɑː/, an-diugh /əɲˈtʲu/), which Gaelic helpfully marks with a hyphen (not unlike the middle dot once used in Old Irish). However, there are numerous counterexamples which would suggest a predominant shift towards first syllable stress regardless of the syllable’s lexical nature (e.g. athraigh /ˈaɾˠə/ and atharraich, descendants of the aforementioned OIr. ad·reig).
Unfortunately, the initial stress accent of the modern languages is of so little interest to researchers and is so ill-documented that the Tegos has not found an accessible scholarly citation that elaborates any further on its particulars. But in any case, for our purposes we will mainly need to focus on the older situation as exemplified by Old Irish, anyways.
Old Irish evidently had initial stress in its earliest days. Syncopation of non-initial syllables as well as the shortening of long vowels in non-initial positions attest this.
Peter Schrijver also points to the the transformation of the 1st sg. impf. verb ending (also attested in Brittonic) as pointing to a very early initial stress:
This sound change was not the regular apocope that took place at a later stage in Goidelic and Brittonic, but instead represents an early syncope where unstressed vowels were lost between identical consonants if they were preceded by an unstressed syllable. In other words, the syncope would not have occurred if there had been a penultimate stress. This is taken as demonstrating that initial stress even predates Goidelic. Thurneysen also concluded that Goidelic initial stress must be very old.7,8
The general consensus then, which most historical linguists seem to take for granted, is that Old Irish inherited its initial stress dating back to at least the earliest days of the Goidelic branch.
Welsh (Cymraeg), Breton (Brezhoneg), and Cornish (Kernowek) generally feature a penultimate (second-to-last syllable) stress accent. Welsh additionally has an oxytonic (final syllable) pitch accent. This ultimately descends from the penultimate stress that was present in Brittonic at some point after the 1st or 2nd Centuries CE.7, 9, 10
As Brittonic lost final syllables through apocope, the penultimate stress had become a word-final stress:
However, over time, this final stress shifted back to the penultimate syllable, but not without leaving a tonal accent in its wake in Welsh. This back-and-forth shifting of the stress in Brittonic is why we see these layered accents in Welsh today. Technically, the role that oxytonic pitch plays in Welsh accentuation can be seen as a component of the stress accent. Since it is not independent or driven by lexical factors, Welsh is not regarded as properly having a “pitch accent” system.4
In Cornish and Breton, George Ken writes:11
We may surmise that, in common with Breton and Welsh, the normal stress in [Cornish] polysyllables was originally on the ultimate syllable, and that it changed, perhaps c. 1050, to the penultimate. It is likely that all elements of the primary stress combined on the stressed syllable, as in Breton (and English), rather than being on different syllables, as in Welsh (Humphreys 1980).
Unlike the prevailing modern Brittonic situation, the Gwenedeg dialect of Breton features word-final stress.10 It remains unclear whether this could be a conservative feature preserved from an earlier phase of the language or a latter-day change, either of which may or may not have been influenced by French.7
Another accent variant occurs in another Breton dialect, this one spoken in the hamlet of Bothoa. The Bothoa dialect features a pitch accent development parallel to that of Gaelic, which is likewise considered to be probably an internal innovation.4 So this, too, is not consequential to Proto-Celtic accentuation.
Now that the modern Brittonic situation has been explained and accounted for, we will want to examine the earliest Brittonic accentuation that’s known so we can best understand Proto-Celtic.
Peter Schrijver’s Studies in British Celtic Historical Phonology (1994) provides the best synopsis I know of in its 2nd chapter, ‘The British Accent.’
Schrijver explains that, although it remained debated for decades since, in the 19th Century philologists such as Thurneysen had already posited that Brittonic had an initial stress accent prior to its well-known penultimate one. They reasoned this on account of the fact that word-initial stops were weakened in unstressed proclitics, while fully stressed words preserved them:
Although not all linguists were convinced by this point alone, Kenneth Jackson (Language and History in Early Britain, 1953) and Schrijver (1994) have together provided at least four more persuasive indicators of initial stress, one of which has already been mentioned in the “Goidelic” section of this article above. Another one, this time postulated by Jackson, is the development of Welsh cawr ‘giant’:
(cf. Galatian Cauaros, Gallo-Latin Cauarillus, Cauares)
Since the ‘a’ in the second syllable was syncopated, we can infer that it was unstressed. Hence, the stress would have had to fall on the initial syllable.
Without taking more space here to elaborate further on all the justifications, we can confidently say that there have now been identified many pieces of evidence which reveal that, prior to its well-known penultimate stress, Brittonic almost certainly had an initial stress phase.
Knowing what has been established for both early Goidelic and Brittonic, Schrijver concludes,
This strongly suggests that the system of initial stress that we find in Irish goes back at least to [Proto-Insular-Celtic] times.
Schrijver exhibits various early Latin loanwords borrowed in the wake of the Roman conquest, and the apocope they underwent in Brittonic, as continuing to exhibit an initial stress:
The initial stress remained, “hence at least in the first or second century AD.”
Schrijver goes on to cite Jackson (1953) for establishing that the shift to penultimate stress cannot have occurred later than the early 5th Century. Any more precise of a date range, though, cannot likely be ascertained.
While Schrijver never articulates it, he leaves this reader with the impression that Brittonic’s penultimate stress could have been a result of Gaulish and/or Latin influence, possibly as a result of the Roman conquest. As will be discussed below, Gaulish appears to have had a penultimate stress of its own. Classical Latin also featured a stress accent that often fell on the penult (although unlike Brittonic, Latin’s accent could be pushed back or forth, depending on certain syllable weight considerations). Considering that Brittonic had an initial stress upon first contact with the Romans, but shifted to a penultimate stress after some centuries of intensive contact, it may be possible that the accent change was an introduced feature.
And it is well known in historical linguistics that prolonged contact with another language is liable to influence accents. As Lyle Campbell (2013) observes:
borrowed stress rules are not uncommon, such as first syllable stress of many of the languages in the Baltic linguistic area [...], or the rule which places stress on the vowel before the last consonant (V → V/—C(V)#), shared by several unrelated American Indian languages of southern Mexico and Guatemala.12
In short, stress rules can easily be influenced or changed by sustained contact with neighboring languages, even unrelated ones.
So while the exact origin of penultimate stress in Brittonic remains unproven, the coincidence of the shift with the Roman conquest bears consideration.
The evidence from ancient Celtic languages known on continental Europe would be instrumental in establishing Proto-Celtic’s likely accentual situation. But of course, none of these languages survived much past the Early Middle Ages at the latest, and attestation is much more sparse than we see in the Insular branch. So extracting valuable information from Continental Celtic is a much more difficult task. But nonetheless, we have some key pieces of evidence to consider:
The best-attested and longest-lived language in Continental Celtic is Gaulish. Unfortunately, none of the attested evidence marks or describes the accent of the language. As such, we have to turn to historical linguistics to hunt for clues as to what the accentuation of Gaulish was like.
In order to understand the first piece of evidence that will be shown, we must point out that Gaulish, like Brittonic, was a “P-Celtic” language. This means that it underwent a sound change from Proto-celtic kʷ > p. However, the suffix -kʷe ‘and’ (cf. Latin -que) was inherited into Gaulish as -c, rather than the *-pe we would expect (and as is seen in another Continental P-Celtic language, Lepontic).13 As such, we can surmise that early or pre-Gaulish had undergone apocopation of that final vowel before the kʷ > p change took place. But what does this apocope tell us?
Let us refer to a concise observation made by Andrew Sihler (2008):
An opinion sanctified by tradition is that stress accent CAUSES [vowel] reductions and losses; but Finnish, Hungarian, and Czech, for example ... [show] that vowel weakening is not an inevitable concomitant of stress accent. The converse is probably more nearly true, namely that accent defined by pitch changes alone is not likely to result in phonetic differences between atonic and tonic vowels.14
So if I may paraphrase Sihler’s statement as a logical axiom, we can reason the following:
Conclusion: Pre-Gaulish was likely to have had a stress accent.
This does not, however, reveal where in the word the stress fell, beyond the obvious observation that it was not on the ultima (final syllable). As Schrijver (1995) points out, apocope of word-final syllables occurred in both Goidelic and Brittonic at a time when they had initial and penultimate stresses, respectively, which shows that either position could lead to the same in Gaulish. But there may be other methods that could suggest the location of the stress.
One such method is toponymical analysis. Schrijver (1995) cites several researchers, including Holger Pedersen, Georges Dottin, and again Jackson for producing work on this topic as it pertains to Gaulish accent. But the most contemporary and prominent contribution to be cited is François Falc’hun’s 1981 publication, Perspectives nouvelles sur l’histoire de la langue bretonne which is unfortunately now out of print. Still, Schrijver, at least provides a brief treatment of it for us so it is not entirely inaccessible.
The essence of this toponymical theory is the fact that multiple French placenames with Gaulish etymologies in common syncopated or apocopated differently. These differences can be explained if the stress accent held a different position in each one. For example:
Falc’hun has ascertained that a penultimate accent is most common throughout rural or marginal areas, while an antepenultimate accent is found in larger urban centers and coastal areas that presumably had more traffic and contact from foreign speakers such as Greek. He thus proposes that the penultimate accent of Gaulish preceded the antepenultimate one, which may have been an innovation from foreign contact. Falc’hun notes also that these same patterns hold true for British cities (e.g. Br-La *Eburācum /eb.ˈur.aː.kum/ > Eng York). Falc’hun considers that the antepenultimate position for these toponyms would sometimes be possible according to Greek accent rules, but not Latin ones. From this he reasons that Greek influence is what resulted in the newer, urban accent.
Schrijver is not convinced that Greek influence would result in an antepenultimate position, noting that the penultimate stress often already agrees with Greek accentuation rules. So the exact origin or cause of the antepenultimate stress remains mysterious. But Schrijver otherwise finds Falc’hun’s theory persuasive.
While I would like to add de Bernardo Stempel’s 1995 work ‘Gaulish Accentuation: Results and Outlook’ to the discussion, it is unfortunately out of print and totally inaccessible to anyone who lacks university library access. Schrijver merely describes it as, “[mentioning] a number of vowel alternations in Gaulish words, some of which might have something to do with stress.”
A final consideration for Gaulish is its many shared innovations with Brittonic. While Schrijver (1995) believes that these post-date the Roman invasions in Brittonic and are thus nothing more than a South-to-North migration of Gallo-Roman influence, he notes that John Koch is of the opinion that there is more of a Gallo-Brittonic unity. As you can see in the model below, de Bernardo Stempel likewise and more recently considers there to have been a unity beyond just post-Roman contact. This should hardly be surprising, in light of the fact that tribes such as the Atrebates held territory on both sides of the English Channel.
So it would be cogent to think that Brittonic and Gaulish had likewise shared accentual properties, especially since accents are known to be easily transmittable from sustained contact. Since we know that Brittonic originally had an initial accent before the Roman period, but shifted to its characteristic penultimate position perhaps one or two centuries later, Gaulish may likewise have shared this initial stress in earlier times. But, only an oxytonic stress placement can be entirely ruled out.
In 1995, Schrijver bluntly intoned, “Nothing is known about the stress in Celtiberian and Lepontic.” But there has been some analysis of these lesser-understood Continental languages since then.
One such notable work is Arenas-Esteban and de Bernardo Stempel’s 2011 article, ‘Celtic Dialects and Cultural Contacts in Protohistory.’15 In it, they indicate that there is some evidence for penultimate stress in Iberian Celtic, but that this accent was not generalized to the whole peninsula. Instead, they suggest that an older, antepenultimate stress had predominated. The evidence they present for this seems similar to what Schrijver says de Bernardo Stempel had provided for Gaulish back in 1995.
The two researchers note the following sound changes from Proto-Celtic to Celtiberian in certain words, such as:
Since we know that stressed vowels typically preserve their integrity, and that vowel reductions especially occur in unstressed syllables, the researchers have collected a number of words that essentially demonstrate that the stress accent must have been further back than the penultimate.
Although they describe this as an “antepenultimate accent”, it must be noted that seven of their ten examples (see table 5, p. 123) are di- and tri-syllabic words—or put another way, words where the antepenult is identical to the initial position. Furthermore, if we examine their examples that do exceed three syllables, there is no clear reason to rule out initial stress there, either:
As we can see, each of these represent a vowel reduction in the penultimate and sometimes perhaps the final position, which could just as well have occurred under initial stress. It’s true that the antepenult is never reduced in these examples, but neither is the initial. So while I would not rule out the possibility of an antepenultimate stress by any means, at the end of the day this evidence could just as easily be marshaled in favor of an initial stress hypothesis.
The researchers go on to explain that both accents also appear within the Italian evidence, although there the penultimate accent predominates (in reverse of the Iberian situation). If we recall my suggestion that penultimate accent could relate to Classical Latin influence, this scenario would align perfectly. It should be noted, however, that their evidence retrieved from Italy is smaller in number and less striking in quality compared to that of Iberia. So any conclusions that may be drawn from it should be done conservatively.
Poetry is naturally one of the most illustrative media for accentuation. Had it survived intact, it would answer a great number of our questions. Some researchers have hunted for poetic texts in the Continental repertoire that could possibly provide more clues. Of course, nothing there is unambiguously labeled as ‘poetry.’ But careful analysis combined with some a priori assumptions has helped to identify certain remains that may possibly be poetic.
David Stifter (2016) has summarized this field of study as it relates to accentuation:
To start with the accentual metrical approach, its fundamental weakness lies in the fact that the stress systems of Ancient Celtic languages are unknown. Only more or less sophisticated hypotheses have been proposed about their nature. In many instances where accentual metrics are involved, the implicit assumption is that the texts have to be scanned with initial dynamic stress. One argument in favour of it is the fact that alliteration, which is particularly suited for stress-initial languages, is of relatively frequent occurrence in Ancient Celtic texts. Initial stress automatically entails that the language tends towards a trochaic or dactylic rhythm in its poetry. Given also the fact that due to the Indo-European character of Ancient Celtic words in these languages tend towards a structure of two or three syllables [...], a trochaic or dactylic dimetre will result most naturally for any given bimembric phrase, for instance for sequences of two elements such as noun + adjective, subject + verb or verb + object. This natural rhythm can be exploited by a poet for versification, but it can also be a coincidental feature of prose.16
So while many of these texts scan poetically if we assume an initial stress, researchers have not yet ruled out that this may be a red herring that simply owes to the typology of Indo-European languages. As Stifter mentions, though, alliteration is another piece of suggestive evidence from poetry to be considered.
To illustrate, a few selections of alliteration are provided below.
Perhaps one of the oldest Celtic texts in the world, and which exhibits verbal art that could be considered poetic, is the Lepontic inscription of Vergiate:
Eska & Mercado have described how this demonstrates alliteration, ring structure, and near-rhyme.17
A Celtiberian example is the rock inscription from Peñalba de Villastar, which Stifter notes has a “remarkable number of alliterations”:
The Larzac tablet, Gaulish’s longest surviving text, also displays “a high incidence of alliterative pairs.” Those which Stifter considers the most striking, from the tablet’s first six lines, are:
The bnanom brictom pair is especially noteworthy, as it has cognates that appear in Old Irish poetry, e.g. brichtu ban/brechtaib ban ‘magic by women.’ The notion that Gaulish might have shared poetic formulae with the stress initial tradition of Irish could have huge implications, and not just for our understanding of Gaulish accent. But as this remains an isolated example from the sparse survivals of Gaulish literature, we can only take its possibilities so far.
To elaborate further on the significance of alliteration, though, Stifter helpfully explains:
Barring a typological study of the phenomenon, from a purely European point of view there is a strong correlation between stress-initial languages and systematic, not just occasional, alliteration. Even outside of the Indo-European speaking area, alliteration is linked with fixed initial stress for instance in the Finnic languages. It is suggestive, although unprovable, that the presence of alliteration constitutes an argument in favour of a stress-initial phase in Gaulish and British. Rather than thinking of it as an Indo-European inheritance, it is more likely that the systematic employment of alliteration had diffused at some unspecifiable time from an unspecifiable source across an area that had a shared prosodic feature, namely initial stress16
Of course, Stifter also notes that Welsh poetry has a well-known alliterative tradition, which would seem to be a blatant counterexample to the association of initial stress with alliteration. However, thinking back to Schrijver’s revelations about early Brittonic, Stifter duly notes, “the role that alliteration plays in Welsh poetry could be seen as an inheritance from a time when British was stress-initial.”
Now that this collective, attested evidence has been laid out for Celtic languages from the modern day to antiquity, we will need to finally look at some theories about the development of Proto-Celtic and the Western European linguistic area more broadly to put it all in context.
To Part 3.