This is the third part of a four-part series. It offers an analysis of the evidence and theory about ancient Celtic prosody.
  • See Part 1 for an introduction to what prosody is and why it matters.
  • See Part 2 for an examination of the attested evidence of Celtic prosody.
  • See Part 4 for our suggested ancient Celtic prosodic systems, based on the historical evidence and theories.

In the previous post of this series, we reviewed the details of accentuation in modern Celtic as well as the evidence for it in ancient Celtic. To briefly recap our findings:

  1. Although there are a variety of accents in modern Celtic, most of these variations stem from changes that happened in relatively recent history.
  2. Both the Goidelic and Brittonic branches exhibited initial stress in their earliest days, suggesting that Proto-Insular-Celtic probably did too.
  3. Brittonic shifted to a penultimate stress sometime after the Roman conquest. Gaulish seems to have had a penultimate stress, based on toponymic evidence. A metropolitan antepenultimate stress might have emerged later in both Britain and Gaul, though did not apparently survive.
    • Whether one views Brittonic and Gaulish as sharing a node on the Celtic tree, or as simply sharing a Sprachbund, the two languages shared many common developments. Accents in particular are easily shared language features, so it would be reasonable to assume there might have been a common accent between the two. Since we know the better-attested Brittonic originally was initial stress, it could be that Gaulish also was prior to becoming penultimate.
  4. There is evidence that Iberian Celtic had had a stress accent that fell before the penult (either antepenultimate or initial) and that this accent was earlier and more predominate throughout Iberia than a later penultimate accent was. Similar evidence also exists for Celtic in Italy, although the penultimate stress is what appears to have been more predominant there.
  5. Alliteration is a frequent form of verbal art found in all the Celtic branches. It is especially associated with and suitable to initial stress. Although languages such as Welsh may use it in the absence of initial stress, it could still point to a previous stage of the language that did feature such an accent, which is consequential for Proto-Celtic (the earliest stage of Celtic).

So taking this evidence into account, it seems reasonable to imagine that there had been an old and widespread initial accent throughout the Celtic speaking world that only more-or-less was overtaken by penultimate stress later on, except for in the Goidelic branch which never shared that development. While there are prominent scholars who continue to dispute or debate this conclusion, there are also those who generally accept it. While it may not be settled fact, it so far seems to be the most parsimonious explanation of the collected evidence.

Or as Salmons quite fairly put it:

Since the nineteenth century, Proto-Celtic, Brittonic and Gaulish accent have proven thorny problems [...] In summary then, the best indications of earliest Celtic accentuation seem to point toward initial stress, if still very inconclusively [...] Gaulish evidence remains especially inconclusive, though initial stress accent cannot presently be eliminated from consideration.1

Before we simply accept the school of thought that initial stress is old and was bequeathed from Proto-Celtic, though, we should answer where Proto-Celtic’s initial stress could have come from. After all, Proto-Indo-European had a mobile, musical accent. It’s quite a change from that to a fixed, initial stress. And if some thinkers disregard the initial stress theory, what other possibilities are there? We should at least understand some alternative views.

To start with, I have presented Salmons, Stifter, and soon to be Vennemann, who all seem to accept Schrijver’s work that indicates Brittonic originally had an initial stress accent. This group generally seems to find the idea of Proto-Celtic initial stress at least plausible, if not nearly certain. So who opposes the idea?

Most researchers who don’t subscribe to this view seem to simply remain agnostic, unwilling to throw their hat in the ring. The evidence is just too murky to form an opinion at all, in their view. This is fair enough from a scholarly perspective, but it’s not useful for a practical application of the language. Some decision must be made, in practice.

I myself have found only two notable scholars who seem to counter the narrative of initial stress: de Bernardo Stempel and Koch, both of whom are giants in the field on par with Schrijver. Unfortunately, as noted in the previous installment of this series, de Bernardo Stempel’s piece that most likely elaborates on her view is no longer in print. And neither have I found a comprehensive argument on the topic put forth by Koch. In brief, Koch has opined:

(Pers. comm. 25 July 2015):
The clearest evidence for initial stress in Celtic is in Goidelic. Not everybody would agree with Schrijver that the whole branch went through such a stage. Isaac and I, for example, think that the PIE accent position (as in Vedic) survived for some time in Celtic. Patrizia De Bernardo also has ideas about stress position without initial stress being generalized in Celtic. It’s complicated, and of course the accent position could have changed more than once, as clearly was the case within the literate period. But initial stress generalized across Italo-Celtic would be a problem. The evidence Schrijver has for initial-stress in Gaulish and Brythonic is far less compelling and clear-cut than that for Goidelic. Even with such provisos what did happen most clearly in Goidelic may be significant in this connection, but an awareness of the less obliging Gaulish and Brythonic evidence should be acknowledged.
(Vennemann 2016)

Setting aside de Bernardo Stempel’s unrevealed opinion on the matter, the competing camps may not be entirely irreconcilable, though, as will be seen below.

An alternative view (dissoi logoi)

In the absence of de Bernardo Stempel’s hypotheses, we should play a bit of dissoi logoi (counterargument) for the sake of checking potential bias or one-sidedness here. If Proto-Celtic had not had initial stress at the time that it diverged, what could it have had?

The original musical accent of PIE seems unlikely, in light of the pre-Gaulish apocope that was noted in Part 2 of this series. A stress accent could have been adopted from a non-Indo-European language, but the known non-IE languages in the region were all initial stress (as will be explained below). So perhaps the PIE accent transformed itself into a stress accent. This has happened in other daughter languages, so what did those transformations look like?

In Koine Greek, the stress accent is mobile and can more-or-less fall on the last three syllables of a word, based on syllable weight. Hindustani likewise has a moveable stress that is sensitive to syllable weight, but which is not necessarily restricted to the last three syllables. In Russian, the stress can fall on any syllable of the word. These are all plainly reflexes of the mobile PIE pitch accent. Of course, these are just a few examples, not an exhaustive account of the Indo-European family. These would lead us to anticipate a movable stress, if a stress accent did indeed develop internally to the language.

But there are examples of fixed stress elsewhere in the Indo-European family, so let’s take a look at those. For comparison, in the West Slavic family there are some examples of fixed stress, either on the initial, antepenult, or penult. The initial stress in some varieties can be explained through language contact with Germanic, but the other two positions (as in Polish penultimate stress and Macedonian antepenultimate stress) remain unexplained. Some researchers have suggested that those languages also had adopted Germanic’s initial stress but then later shifted it to their current position (similar to what Schrijver suggests Brittonic had done). Others may give internal prosodic explanations, or loss of final stress for their developments.2

So this would suggest we shouldn’t rule out the possibility of the PIE accent yielding a fixed stress accent, but it certainly seems a less likely and more poorly-supported scenario.

The West Slavic situation could actually further buttress the initial stress hypothesis if we propose that Continental dialects had first adopted initial stress from a pre-Indo-European substrate and then later undergone the same sort of shift as Brittonic and perhaps Polish and Macedonian did. But to maintain the dissoi logoi, we will work with the assumption that these fixed stresses might have ultimately developed independently out of the PIE mobile pitch accent.

So, if we assume the PIE accent developed into a non-initial Celtic stress on its own, it could have been mobile before later becoming fixed. Or, less likely, it could have fixed itself on the penult or antepenult at an early date before diverging into the daughter languages.

If we accept the first possibility, it could perhaps explain why we see varying evidence of penultimate and antepenultimate accents on the continent. We could propose that a mobile accent had eventually fixed itself, but did so differently or unevenly across different Celtic dialects.

If we assume the latter, Celtic would probably have to be idiosyncratic for producing a fixed stress out of the PIE mobile accent. We are also left with the open question of where the stress was originally fixed and why some later dialects relocated it.

Under both hypotheticals, we would have to either discard Schrijver’s work on Brittonic (which Koch sounds willing to do, although I have not seen any proper refutation made), and say that only Goidelic ever moved its stress to the initial position (probably due to substratic influence). Or, we can leave Schrijver’s work intact and assume that Proto-Insular-Celtic had shifted from an earlier stress system to fixed initial stress, but then Brittonic later shifted again, this time to the penultimate, perhaps from some kind of Gallo and/or Roman influence after the conquest of Rome.

The dissoi logoi leaves some unanswered questions and presents a more complicated problem to solve than the initial stress hypothesis does. But that being said, its greater difficulty and murkiness does not make it false, and there’s really no part of it that could be said to be disproven. While the notion that Proto-Celtic had developed a fixed penultimate or antepenultimate stress would seem dubious, the mobile stress option is quite workable, albeit that it has fewer suggestive indicators than the initial stress hypothesis seems to have.

Now that we have provided an alternative viewpoint, let us return to what a possible scenario could be under the assumption of initial stress:

Initial Stress Development

Schrijver may be the most enthusiastic proponent of the initial stress hypothesis. He also determinedly defends his theory of Proto-Italo-Celtic against a host of critics. According to his theory put forth in Celtic From the West 3 (2016), he proposes Northern Italy as the origin site of Proto-Celtic and gives an overview of how he believes Proto-Italo-Celtic had developed. According to the theory, it’s noted that there are two sound changes which could only have occurred during the presence of the Proto-Indo-European mobile pitch:3

  1. Dybo’s rule: Long vowels are shortened if they preced a resonant plus a vowel that was accented in Proto-Indo-European.
  2. Pretonic CHIC > CIC: A controversial sound change that Schrijver asserts was unique to the Italic and Celtic branches.

Immediately after the above changes, Schrijver suggests fixation of initial stress took place. He considers this change to be typologically trivial so that it could have happened spontaneously without requiring external influence. He explains his relative chronology saying, “I assume that identical innovations in Italic and Celtic occurred in the same order if there is no reason to think that they did not,” and says of initial stress: “there is no reason for thinking it was not a common innovation [...] If this is an Italo-Celtic development, it postdates (1) and (2), both of which presuppose the presence of a PIE mobile accent.”

So essentially, Schrijver proposes it happened at that point because it might as well have and there’s no evidence that it did not. Not to mention Koch’s objection from earlier, though, there are other researchers who suggest reasons why the initial stress may have appeared later.

There are indeed some issues with Schrijver’s assumptions about initial stress. For one thing, if fixed initial stress were so typologically trivial and prone to develop in the Indo-European family, we would expect to see other examples of it. It occurs in some West Slavic languages, but this is explained as an adoption from Germanic. It occurred in the Germanic branch, but Schrijver himself (as well as many others) explains this as an influence from the Finno-Uralic languages,4 so external to the Indo-European family. Italo-Celtic would have to be totally unique, then, for an Indo-European language developing this internally.

Salmons (2003) provides an overview of how scholars have generally agreed that there was an initial stress zone in the North and West of Europe. They have quibbled over the details of the nature of this zone and how it developed (such as whether or not these multiple instances of initial stress are related or merely coincidental), but nonetheless it is widely agreed that Germanic, Italic, and possibly Celtic all shifted to initial stress as well as the fact that a number of non-Indo-European languages with initial stress also existed in this region. Salmons also observes:

Evidence from numerous modern languages suggests that language contact, even relatively moderate contact, correlates positively with shift from movable pitch accent or tone systems toward fixed, stress systems.

He also notes that some scholars have proposed Italic initial stress was influenced by that of Etruscan, not unlike how Germanic was influenced by Finno-Uralic. So again, an external influence. This is a suitable explanation and would easily account for Celtic as well if one accepts Schrijver’s Italo-Celtic hypothesis.

But there is perhaps one more thing we should consider about Proto-Celtic’s relation to initial stress, which is that Proto-Indo-European p was weakened at the beginning of words in Proto-Celtic to φ/f. This seems counterintuitive if the initial syllable had been stressed, since the emphasis of that stress would seem more likely to strengthen or preserve the sound rather than weaken it. For comparison, Proto-Germanic similarly underwent a p > φ sound change (see Grimm’s Law), and this is known to have occurred at a time when Germanic still had the PIE musical accent.

However, fortified as ‘f’ in Germanic and persisted to the present day:

PIE *ph₂tḗr > PGmc *φadēr > Eng ‘father’

Meanwhile in Proto-Celtic, word initial *φ/f was lost entirely, which is further counterintuitive to initial stress:

PIE *ph₂tḗr > PCl *φatīr > Ir athair

One would anticipate that, if anything, might strengthen to **f as it did in Germanic rather than be eliminated altogether. Granted, I am not aware of any research that would prove these changes could not occur under initial stress. But given what we normally witness elsewhere, it seems that it would be a highly unusual development, especially compared to Celtic’s immediate, stress-initial neighbors. And if there are any more cogent explanations for why this took place besides, “it just did,” we should examine them.

Enter the theory of Vasconic substratic influence. Vennemann is well-known (perhaps infamous) for his theories about Vasconic (the proto-language of the Basque language family). While some of his claims are too broad and have been rightfully pilloried, we should not throw the baby out with the bathwater. Some of his more sensible propositions include the idea that Vasconic lacked the ‘f’ sound and that it could have featured a fixed initial stress. He points out that this lack of ‘f’ exerted an influence on Latin in the past, such that we see the effects of it on certain Spanish dialects today.5 While I would not accept his notion of a “Vasconic Europe” where this language family covered wide swaths of the continent, it certainly would appear plausible that Proto-Celtic may have existed far enough West and Vasconic far enough East for there to have been contact between them. If we agree with Schrijver’s placement of the Proto-Celtic homeland in Northern Italy, it would be very simple to imagine contact between the two. And with the more extreme theory suggested by Koch that Celtic’s origins were in Iberia, it would be guaranteed that there was contact.

So it could be said, whether or not Italic and Celtic were ever unified branches, that Proto-Celtic had originally had the PIE musical accent (or at least that it lacked initial stress) at which time the word-initial p weakened to φ/f. Later, Proto-Celtic came into contact with Vasconic, whose speakers could not pronounce this φ/f sound and who stressed each word at its start. In this way, Proto-Celtic simultaneously lost the φ/f sound and fixed a stress accent to initial syllables. This is a rather elegant and parsimonious way to solve two tricky problems, and the historical circumstances could very well have existed for this to happen.

Shrijver (2016) depicts the weakening of p > φ/f as an identifiable moment when Celtic was driven away from Italic. It’s possible, if we dismiss his idea that they both generated fixed initial stress together, that the converse is true: Perhaps Pre-Italic speakers had adopted initial stress from Etruscan, further separating them from pre-Celtic speakers who instead remained with the PIE musical accent. The initial stress in pre-Italic could have favored the preservation of *p not long before pre-Celtic speakers began leniting it, thus representing a tell-tale moment where the two branches had begun to diverge into Italic and Celtic, respectively.

Then, of course, we could imagine at a later date that the Celtic speakers had moved North or West where they developed a relationship with Vasconic speakers, and the rest is history.

Or, as Koch quipped, “To put it in layman’s terms, Celtic is Indo-European spoken with an Iberian accent.”6

Or a ‘Vasconic’ accent, as Vennemann (2016) would have it.

So now we have outlined a logical, if necessarily speculative, explanation of how the development of initial stress in Proto-Celtic could align with the current, collected evidence. But how might the fixed penultimate or even antepenultimate stress accents have arisen within some Celtic dialects?

Penultimate and Antepenultimate Stress Developments

There are uncountable and imponderable mechanisms by which initial stress could have shifted to the second- or third-to-last syllable. But I will put forth a few propositions, if only for the sake of being able to imagine some possibilities, or to brainstorm potential avenues of study.


To begin with, we should note that “initial stress” does not mean “only the first syllable is ever stressed”. We see in Goidelic, going back to its earliest attested days, that a secondary stress falls on subsequent members of compound words. For example, seanbhean /ˈʃan.ˌvʲan/ < senabenā /ˈˌbe.naː/. Since many Celtic words are disyllabic like both of these roots are, it is easy to see how such a system would regularly stress the penult in such compounds. Not to mention the fact that for disyllabic simplexes, the penult and initial syllable are synonymous (e.g. /ˈwi.ros/, /ˈbe.naː/). Additionally, many noun-forming suffixes are disyllabic (e.g. -ati-, -ikno-, -geno-, etc.) and may have been stressed on their initial syllable as well.

So, since in a number of situations secondary stress might have fallen on the penult, perhaps this secondary stress had become generalized over time to all situations. And since the secondary and primary stresses would be indistinguishable in disyllabic simplexes, perhaps this ambiguity eroded the distinction between the two stresses. And once that had happened, it could leave open an opportunity for the secondary stress to prevail while the original, primary stress somehow fell to the wayside.

Many adjectival suffixes are likewise disyllabic, but furthermore also feature long vowels in the penultimate position (e.g. -āko-, -īno-, etc.). This is significant in connection to the fact that long vowels and stress often correlate:

As O’Rahilly (loc. cit.) rightly points out a language, such as Middle Irish, with a strong initial stress (not pitch) accent and long vowels in later syllables would show an inherent instability. This can be resolved in one of two ways. Either the long vowels in post-initial syllables are shortened (this is the solution in Ulster Irish, O’Rahilly, loc. cit.) or they remain long and attract the stress as in Munster Irish. Western Irish lies in between these two poles and would be, on O’Rahilly’s analysis, an unstable system as it allows long vowels in unstressed syllables.7

Thus, the possible “instability” posed by these suffixes could have drawn stress to the penultimate, which was subsequently generalized throughout the dialect.


Turning our attention now to antepenultimate stress found in metropolitan areas, Falc’hun has proposed Greek influence as the cause (see Part 2 for references and discussion). Schrijver points out that this explanation is not too convincing, because the Greek accent often would have found the penult suitable rather than necessarily falling on the antepenult. No other indications of the antepenultimate stress’ origin seem yet forthcoming.

But if we analyze the evidence critically, we may begin to question whether the antepenultimate stress perceived by Falc’hun existed at all, or if it was instead a mirage.

Consider for a moment how apocopation in Brittonic led to penultimate stress becoming oxytonic, before later shifting back to the penult. Perhaps this exact same phenomenon was happening at an earlier date in metropolitan areas than it did in marginal ones. And thus, what has been perceived as an ante-penultimate stress based off of Proto-Celtic reconstructions of toponyms, is a red herring: perhaps, in fact, the final syllables had been lost by that time, and what would have once been the antepenult was simply now the penult. E.g.:

*Eburācum /eb.ˈur.aː.kum/ had already become something like **Eburāc /eb.ˈur.aːk/
*Nemausus /ˈ had already shortened to something like **Nemaus /ˈnem.aus/

Schrijver informs that Jackson dated Brittonic apocope to the fifth or sixth centuries,8 but notes that the accent had already shifted back to the penultimate by that time. Perhaps this is as far back in time as the toponyms actually take us.

As discussed in Part 2 of this series, the “antepenultimate” stress observed at an early date in Iberian and Italian inscriptions by de Bernardo Stempel could easily be interpreted as initial stress, and thus not a part of the same accentual phenomenon proposed by Falc’hun.

Now that we have grappled with competing viewpoints and established plausible historical scenarios, we can finally settle on how to accent our poetry, speech, and song, in the next and final installment of the series.

To Part 4.


  1. Salmons, J. (2003). Accentual similarities between Germanic and Celtic. In Accentual change and language contact: Comparative survey and a case study of early Northern Europe. London: Routledge. (back)
  2. Hammel, R. (2009). 7. Fester Akzent (Formen und Funktionen. In G. Ungeheuer, H. Steger, H.E. Wiegand, S. Kempgen, P. Kosta, T. Berger & K. Gutschmidt (Eds.) Die slavischen Sprachen / The Slavic languages. (pp. 76-86). Berlin: Mouton De Gruyter. (back)
  3. Schrijver, P (2016). 17. Ancillary study: Sound change, the Italo-Celtic linguistic unity, and the Italian homeland of Celtic. In J.T. Koch, B. Cunliffe, K. Cleary & C.D. Gibson (Eds.) Celtic from the West 3, Atlantic Europe in the metal ages: Questions of shared language. Oxford: Oxbow Books. (back)
  4. Schrijver, P. (2014). Language contact and the origins of the Germanic languages. New York, London: Routledge, Taylor & Francis Group. (back)
  5. Vennemann, T. (2016). 18. Ancillary study: Celtic as Vasconized Indo-European? Three structural arguments. In J.T. Koch, B. Cunliffe, K. Cleary & C.D. Gibson (Eds.) Celtic from the West 3, Atlantic Europe in the metal ages: Questions of shared language. Oxford: Oxbow Books. (back)
  6. Koch, J.T. (2013). Out of the flow and ebb of the European Bronze Age: Heroes, Tartessos, and Celtic. In J.T. Koch, B. Cunliffe, K. Cleary & C.D. Gibson (Eds.) Celtic from the West 2: Rethinking the Bronze Age and the arrival of Indo-European in Atlantic Europe (pp. 101-146). Oxford: Oxbow Books. (back)
  7. Hickey, R. (2014). 3. Stress in Irish. In The sound structure of modern Irish.. Berlin: De Gruyter Mouton. (back)
  8. Schrijver, P. (1995). II The British accent. In Studies in British Celtic historical phonology (pp. 16 - 22). Amsterdam: Rodopi. (back)

Previous Post Next Post