10 Vowel Loss in Runic Inscriptions

The two chapters following this one will each deal with different aspects of metrical evidence for Norse prosody. None of the poetry in question is likely to predate the year 800 or so, and most of it is clearly rather later than that.¹ This means that it largely comes after some very significant prosodic developments in North Germanic: (at least) two major waves of vowel reductions that had an immense effect on the details and distributions of syllable weights.² These are often referred to as syncope periods, though this term is somewhat inexact: strictly speaking, syncope should refer to reductions in the middle of a word, and apocope to reductions in absolute finality. In North Germanic, however, there were a number of vowel reductions and losses that do not divide neatly into distinct syncope and apocope processes. There is also the question of what ‘final’ really means: if a loss can occur in true finality and before an extrametrical final consonant, but not before other consonants, this would seem to be a type of apocope rather than syncope (Riad 1992: 142). I favour neutral words such as loss or reduction (including reduction to zero) in the following discussions.

The workings of these loss periods is mainly attested in runic inscriptions, and though their general chronology is well known in the phonological literature,³ a number of outstanding problematic issues remain. I won’t try to resolve all of them, or deal with every messy philological issue involved, but instead will focus on trying to extract the main implications for foot structure: Do these vowel losses provide evidence for foot structure during these periods, what kind of consonant extrametricality might be at work, and what was the state of the language after all this was said and done, going into the later Viking Age and the manuscript age that followed?

The earliest direct written records of any Germanic language (not counting names recorded in Hellenised or Latinised forms) come from inscriptions in the Older Futhark script. This is an alphabetic script, almost certainly derived from Alpine alphabets of northern Italy (Mees 2000; Stifter 2020), used to write over 200 surviving inscriptions on stone, wood, and metal. Though the corpus is not large, and many inscriptions are problematic in one way or another – due to brevity, damage, and/or lack of context for interpretation – enough inscriptions can be read with enough confidence to give a fairly good picture of the state of North Germanic in the period of roughly 200–500 AD. I call the language represented in these inscriptions Early Runic (following Nielsen 2000), but it’s known by many names: especially Proto-Norse,⁴ Ancient Nordic, or Primitive Norse, and terms such as Urnordisch and Urnordisk are sometimes taken over into English as well.

The value and difficulty of Early Runic can be seen in one of the longer inscriptions, the famous Tune stone (KJ 72),⁵ found some 70 kilometers south of Oslo. The inscription is generally dated to around 400 AD. Transliterating the runes in bold, as is the normal convention when emphasising the precise reading of an inscription, and indicating partially legibile letters with a dot underneath, the two sides of the stone read:⁶

(155) a. ekwiwazafter · woduri

dewitadahalaiban : worahto · ṛ

b. zwoduride : staina ·

þrijozdohtrizdalidun

arbijasijostezarbijano

A boldfaced block like this might seem entirely impenetrable, and there are indeed many real obstacles to interpretation. Word boundaries are only sporadically indicated, two or three words have probably been lost due to damage, there is at least one likely scribal error (dalidun is probably for da[i]lidun), and certain phones and important phonological features are not systematically indicated (including nasal consonants before stops). That said, generations of interpretative work on the stone have produced a consensus reading of most (not all) words. Here is the text as analysed by Eythórsson (2013), to my mind the most convincing reading to date (the text is no longer in boldface, to show that a basic linguistic interpretation has been imposed):

(156) a. ek Wīwaz after Wōdurīdē witanda-h^alaiban wor^ahtō r[…]⁷

‘I, Wiwaz, made [runes] in memory of Woduridaz, the bread-guard.’

b. [… z]⁸ Wōdu-rīdē staina […]⁹

‘[?Wiwa]z [set up] a stone for Woduridaz.’

c. Þrijōz dohtriz da[i]lidun arbija, sījōstēz arbijanō.

‘Three daughters shared the inheritance, the very closest of heirs.’

The superscript vowels are epenthetic, and are potentially of considerable prosodic interest. Such vowels are common in Early Runic, but not normally reconstructed for Proto-Germanic and certainly absent in classical Norse. They tend to appear between a liquid and a neighbouring consonant, but aren’t universally present in such contexts: dohtriz, twice arbi-.

Epenthesis aside, this inscription preserves many vowels whose reductions and losses will be the subject of this chapter. Final-a in staina and arbija, and before z in Wīwaz are all preserved, as is the sequence-iz in dohtriz. Strictly speaking, there is no overt indication that the vowels in the endings-ē(z) and-ō(z) are still long, but the fact that they are not lost in later Norse – only reduced to-i(r) and-a(r) – suggests that they are indeed still bimoraic, as they were in Proto-Germanic. Medially, unstressed u is also preserved in Wōdu-, where it would eventually be deleted.

This is not to say that there have been no changes at all to unstressed vowels by this date. Proto-Germanic final *-ō appears to have been shortened and raised to *-uin Northwest Germanic (Ringe & Taylor 2014: 15–16), and this is reflected in words such as mīnu ‘my (fem.nom.sg)’ (Opedal, KJ 76, c. 350 AD). This falls together with Proto-Germanic short *-u, which are both retained during the main period of Early Runic, and share the same patterns of retention and loss in the later periods. Another inscription, the Vetteland stone (KJ 60, c. 350 AD), has the verb ist ‘is’, from Proto-Germanic *isti, which shows the loss of a final high vowel in the second syllable of a low-stress, high-frequency word. This is the only example I know of for loss of *i (or indeed any final vowel) in Early Runic before the major loss periods discussed below, and the vowel is normally retained in inscriptions in words such as hal(l)i (Strøm, KJ 50, probably after 450 AD, perhaps as late as 600; Düwel 2008: 34). There have also been monophthongisations (Early Runic unstressed ē is often from *ai). On the whole, such changes to Early Runic unstressed vowels are of little real significance, and outright losses are very restricted indeed before the 6th and 7th centuries. It is the following centuries – the language of which may be called ‘Transitional Runic’, defined negatively as the period after classic Early Runic and the late Viking Age language attested in abundance starting in the 10th century – that see the vowel losses that are the main focus of this chapter.

By around the year 600, runic inscriptions had begun to show unmistakable losses of at least some unstressed vowels in at least some contexts. Because of difficulties in the data, there isn’t a full consensus about the exact lines of development, but these early changes are often grouped together into a single first loss period.

A useful anchor point in a period where the data is typically sparse, hard to interpret, and difficult to date comes from a group of closely related stones from the area of Blekinge, in what is now the far South of Sweden. The three earliest of these are the Stentoften, Gummarp, and Istaby stones (DR 357/KJ 96, DR 358/KJ 95, and DR 359/KJ 98, respectively), and a fourth, the Björketorp stone (DR 360/KJ 97), probably dates from slightly later than the others in the Blekinge group. The following list gives a selection of important forms from the earlier three inscriptions, illustrating both retentions and losses of unstressed vowels in (probably) the beginning of the seventh century. Note that the old rune for a is often used to write epenthetic vowels, with full [ɑ] being represented instead by the letter (which has variant forms) that used to stand for [j].¹⁰ This new a-rune is transcribed as a, and is very common (ls). As above, all epenthetic vowels are written in superscript (on the uncertain final vowels of 157c and 157e, see below):

(157) a. haþu-wol^Afz < *haþu-wulfaz (Stentoften)

b. hari-wol^Afz < *harja-wulfaz (Stentoften)

c. haþu-wol^Af^A < *haþu-wulfaz (Gummarp)

d. af^atz < *aftir (Istaby)

e. hari-wul^af^a < *harja-wulfaⁿ (Istaby)

f. haþu-wul^afz < *haþu-wulfaz (Istaby)

g. h^Aeru-wul^afīz < *heru-wulfijaz (Istaby)¹¹

On the data from the ‘curse formula’ portion of Stentoften, see §10.2.1 below.

Three more data points might date from a similar period, though they all involve extra problems:

(158) a. Hrōzēz< *Hrōzijaz (By, KJ 71, 6th century)¹²

b. Wīz < *Wīwaz (Eikaland, KJ 47, 450–600)

c. haukz < *haβukaz (Vallentuna, before 650)¹³

This small dataset is just for what might be called ‘Early Transitional Runic’, from a narrow window in the late 6th and early 7th centuries. It is small, and decidedly ambiguous on a number of points. Still, it does show, very clearly, that at least some words that had once had a final *-az (157a, 157b, 157f, 158b, 158c), have lost the unstressed *a. Most of these examples occur after historically heavy syllables, though in Wīwaz > Wīz the original length of the root vowel is uncertain (Kroonen 2013: 590).

There is also reduction in the ending *-ijaz (157g, 158a), which most likely represents a slightly earlier and distinct change (Birkmann 1995: 176, Schulte 1998: 83–87, 97). Certainly the example of Hrōzēz (158a) suggests this, since it occurs alongside unreduced forms such as irilaz in the same inscription. That this was a special development is further suggested by its outcome as a long vowel, *-ī-, which escapes any further deletion: contrast *hirðijaz > hirðir ‘shepherd’ with *niþjaz > niðr ‘relative’ and *winiz > vinr ‘friend’.

One point is unclear, but doesn’t necessarily have a strong bearing on the wider picture: this is the accusative singular *-aⁿ, reflected in-wolafa (157c) and-wulafa (157e). These have a final graphic a/a where this original *-aⁿ stood, and where epenthesis would be fairly unexpected. Perhaps the nasalisation of the vowel made *-aⁿ more resistant to deletion than was *-az, but just how much of a gap there was between the losses of the two vowels is fairly unclear. At least by the time of the Eggja inscription (c. 700), nasalised *-aⁿ had disappeared (land < *landaⁿ, probably also stain < *stainaⁿ, if this really is accusative).

So far, what can be really safely concluded is that the non-nasalised *a was lost after at least some heavy monosyllables and in the ja-stem ending *-ijaz by the early 7th century. More than this requires a closer examination of some problematic data, which may or may not support a further division of this early vowel-loss period into two periods – and which involves a closer consideration of syllable weight during this time.

One of the most interesting features of the Blekinge group is that the Stentoften stone shares part of its content – the so-called curse formula section – with the later Björketorp stone, with the latter showing some identifiable linguistic innovations. Reading the two side by side is almost like watching Early Runic take small steps towards later Norse right before your eyes. There are three forms in this repeated section that Kiparsky (2009: 25, ex. a4) cites as evidence for a hypothesis that the earliest vowel deletions should be separated into two distinct phases – phases Ia and Ib, let’s call them – with the dividing line falling sometime between Stentoften and Björketorp.

Specifically, Kiparsky sees the first phase as allowing vowel loss only when two constraints are satisfied:

For Kiparsky, epenthetic syllables are taken as real and important factors in determining how many syllables a word has, and how heavy they are. He is rather unusual in taking this approach, but at least in the Blekinge group, epenthesis does seem relatively stable and consistent, and it seems fair to explore the idea that it might be prosodically significant. Just as importantly – though I think wrongly, as I will argue shortly – Kiparsky considers all final consonants to be extrametrical, invisible for purposes of syllable weight.

Kiparsky’s first constraint, his ban on overheavy syllables, seems to have some merit – though the evidential basis for it is rather slight. He is certainly correct that in the early Blekinge group (and a couple of other inscriptions, some of which do not provide good evidence), vowel loss never produces any overheavy syllables, assuming that epenthesis counts as real, and that at least final-z is extrametrical. For instance, a-fat〈z〉 (157d) would have one light syllable, a-, and a second with two moras: one from the epenthetic a, and another from the t, with the z being discounted as extrametrical.

Such negative evidence is not very strong. The amount of data is very small, and a mere absence of words overheavy from vowel deletion could very easily just be due to chance. Positive evidence is much more important, and there is indeed a little: specifically, three words from the curse formula section of Stentoften, which might show vowel deletion being blocked by a constraint against overheavy syllables. In all three cases, I also give the parallel word from the slightly later Björketorp version, which Kiparsky (2009: 25, ex. a4) sees as evidence that the overheavy constraint was relaxed during the intervening years:

(159) a. b^Ariutiþ ‘breaks’ < *briutiþ (Stentoften)

b. b^Arūtz [bərȳtr^j] (Björketorp)

(160) a. hidez ‘?bright(ness)’ ?< *haiðiz (Stentoften)

b. haidz (Björketorp)

(161) a.-lasaz ‘-less, without’ < *-lausaz (Stentoften)

b.-lausz (Björketorp)

Example (159) is particularly complicated and important – though it provides no support for Kiparsky’s chronology – and I discuss it separately in §10.2.2, before considering (160) and (161) in §10.2.3, and the qualified evidence they do provide for the theory of two distinct early phases of deletion.

The bariutiþ of (159a) has been reduced by one final syllable from Proto-Germanic *briutidi, but this change may well have happened centuries earlier, perhaps in parallel with the reduction of *isti to ist mentioned above. If so, then the source form in Early Runic was *briutiþ, and this clearly underwent no further vowel loss by the time the Stentoften inscription was carved. This contrasts strikingly with the matching form barūtz from Björketorp (159b).

For Kiparsky, the lack of vowel loss in bariutiþ could be a matter of his constraint against overheavy syllables during this earliest loss period (though he doesn’t explicitly cite this particular form in that context, or indeed comment on its vowel retention at all). If the final vowel were lost, the word would be left as ^xbariutþ [bəriutθ] or the like. The stressed syllable would have two moras from the diphthong, and a third from the t. Even if the þ were extrametrical (as per Kiparsky’s rules), and not contributing a fourth mora, the syllable would already be overheavy. Kiparsky, as noted above, suggests that the overheavy constraint was relaxed by the time of Björketorp, allowing a form such as barūtz to arise through vowel loss.

This particular pair, however, won’t sustain this kind of analysis. Specifically, the lack of vowel loss in bariutiþ can’t be due to an archaic constraint with chronological significance, since this form would, most likely, have never undergone deletion at all. This can be shown by the development the second-person plural verbal ending, which was probably identical to the third-person singular in Early Runic: both *briutiþ. This plural form remained unreduced in literary Norse, which has brjótið ‘you (plural) break’.¹⁴ It seems clear that the change between Stentoften’s bariutiþ and Björketorp’s barūtz was not phonological, but purely morphological: the replacement of the old third-person ending in-iþ with the second-person singular form in-(i)z >-(i)r (literary Norse-r). This change affected not only the consonantism of the ending – something this inscription has long been famous for illustrating (Nielsen 2000: 263–264, with references) – but the vocalism as well.

The development of these forms can be shown schematically in table 10.1, with > indicating (mostly) regular sound change, and → major analogical replacement. This lack of deletion in brjótið is particularly important because it suggests that *briutiþ and *briutiz did not have the same prosdoic structure in Early Runic. In particular, it suggests that Kiparsky’s rule that all final consonants count as extrametrical is likely incorrect. More likely, only the word-final-z/-r was extrametrical.¹⁵ The status of final-þ as fully metrical and moraic is what accounts for the lack of deletion in *(briu)(-tiþ) > brjótið: the final consonant made the final syllable closed and heavy, and not elligible for vowel deletion of any sort. In *(briu)-ti〈z〉, the unfooted *i is deleted, to give eventual brýtr. This different view of final-consonant extrametricality will lead me shortly (in §10.3) to depart from Kiparsky’s conclusions on vowel losses.

More immediately, the non-deletion in brjótið means that bariutiþ is simply uninformative about the chronology of vowel deletion. It retained its vowel not to satisfy an archaic overheavy constraint, but simply because that vowel was never open to deletion at any point in the history of Norse. This means that the possible existence of the early overheavy cosntraint turns largely on two other words from Stentoften.

Stentoften’s hidez (160a) looks rather peculiar, especially when taken on its own. The matching word on Björketorp is haidz (160b), which is somewhat easier to make sense of. The latter form points to a root syllable in *hai-, which probably lies behind Stentoften’s spelling with i. This is most likely just a simple error for *ai, though it could potentially represent a dialectal monophthongisation (compare Swedish heder, Danish hæder, with East Norse monophthongs). Either way, the root syllable is heavy, with two moras from a diphthong or monophthongised long vowel.

The second syllable of Stentoften is spelled e, which under the traditional reading of this inscription stands for unstressed *i, either as an orthographic variant or as a phonetically reduced vowel. This interpretation of hidez as [hɑiðiz], [hæːðɪz], or some other variant along such lines, might seem like a rather drastic interpretation of the letters in question, but is nonetheless probably the best conclusion – see Schulte (1998: 113–119) for a thorough discussion of the philological and etymological problems involved. He concludes that it does indeed stand for *haiðiz, a neuter s-stem meaning ‘brightness’. Despite some reservations and the many uncertainties involved, I agree that this is the most plausible view.

If this interpretation is right, it provides one positive example of the kind of non-deletion that Kiparsky proposed for the earlier Blekinge group, and which is meant to motivate the overheavy constraint. This *haiðiz couldn’t undergo vowel loss at the earliest stage, because the resulting ˣhaidz would contain an overheavy syllable: even discounting the final-z as extrametrical, the diphthong already provides two moras, and the d a third. The introduction of vowel deletion by the time of Björketorp’s haidz really does seem to represent a new development taking place during the 7th century, between the time of the two inscriptions.

This conclusion is probably supported by (161). As with hidez, the exact interpretation of the letters here is not entirely clear-cut, but the most likely view is that-lasaz (161a) is a form of *-lausaz. The etymological diphthong would then be represented by a single vowel sign, again either as an error or due to monophthongisation (Schulte 1998: 136). The key feature is again the lack of vowel loss in the final syllable, and Kiparsky’s overheavy constraint does account for this well. Björketorp’s-lausz (161b) would then seem to show the loss of the overheavy constraint, and the introduction of deletion, in between the two inscriptions.

In general, then, the Blekinge group does seem to offer support for Kiparsky’s view. A plausible interpretation of the evidence is that vowel deletion was blocked in the Stentoften period, when it would produce an overheavy syllable. Deletion was able to take place in-wulafz (157a, 157b) even at this early stage due to vowel epenthesis. By the time of Björketorp, however, the constraint against overheavy syllables was lost, so that deletion could also operate in forms such as-lausz.

Unfortunately, despite the attractiveness of Kiparsky’s overheavy constraint, uncertainties remain. Kiparsky posits a plausible and prosodically interesting constraint – ‘don’t produce overheavy syllables’ – which is never violated in the early Blekinge group, and which could explain both the variations on Stentoften itself, and between Stentoften and Björketorp. The tiny number of data points is of course reason enough for caution already, since slight reinterpretations of the inscriptions could easily remove one or both forms from consideration entirely.

It has also been suggested that the curse formula section of Stentoften could be earlier than the rest of the inscription (Krause & Jankuhn 1966: 214), a possibility that would entirely undermine the overheavy constraint. Instead of seeing a contrast between forms such as-wulafz (with vowel loss) and-lasaz (without) in the same linguistic stratum,-lasaz would simply be an older, pre-loss form and-wulafz a younger, post-loss form. Schematically, the two possible chronological implications are:

That is, if Krause & Jankuhn are right about the curse formula, then Kiparsky’s two early deletion periods – separated by the presence or absence of the overheavy constraint – loses all empirical basis.¹⁶ Specifically, the curse formula could simply belong to the pre-deletion period altogether, while the remainder of Stentoften, along with the other Blekinge inscriptions (including Björketorp), and most other inscriptions down to around 800 AD or so, would all show the effects of one single round of vowel deletion, a general phase I not restricted by an overheavy constraint, and not to be subdivided into Ia and Ib.

How seriously we should take the possibility of the curse being older is hard to say. Krause & Jankuhn’s suggestion is broadly plausible given the layout of the inscription, but hardly seems necessary on general runological grounds. The idea is instead based essentially on linguistic criteria – that is, on precisely the unstressed vowels under discussion. But the possibility, along with the general extreme slenderness of the data, means that the question of whether there was just one early loss period or two is rather hard to answer decisively – not an uncommon kind of conclusion when working with earlier runic data. On the whole, I am inclined to think that Kiparsky’s two-loss model has merit. It is simplest to take Stentoften as a single inscription, and Kiparsky’s hypothesis is an elegant way to explain which words on it show vowel loss, and which ones don’t. This approach would also fit well with a larger view of early Germanic prosody as centred around the bimoraic trochee: just as with Kaluza’s law, strict bimoraism may have been preferred where it could be achieved during the very earliest vowel reductions. Possible further reinforcement, though of a very indirect kind, for Kiparsky’s two phases of early vowel loss may also come from umlaut, as I will argue in §10.4. But the doubts just raised make it hard to lean on this conclusion very strongly, and the supporting evidence of the umlaut is very indirect and involves a specific analysis of a highly contested issue.

As Kiparsky (2009: 25–26) notes, if there ever was an overheavy-syllable constraint on vowel loss, this was relaxed by the time of Björketorp, and is absent in subsequent inscriptions. Overheavy syllables resulting from vowel deletion are amply attested (by runic standards), not just by the Björketorp data given in (159–161), but by many other inscriptions from the following century and a half or so:

(162) a. mænnz < *manniz (Eggja, KJ 101, c. 700)

b. fiskz < *fiskaz (Eggja)

c. Rhōaltz < *Hrōþu-waldaz (Vatn, KJ 68, c. 700)¹⁷

d. Þiaurīkr < *þeuda-rīkz (Rök, Ög 136, c. 800)

e. Hraiþ-marar < *Hraiði-maraiz (Rök)

These examples, which are illustrative rather than comprehensive, show deletion of all the Proto-Germanic short unstressed vowels – *a, *i, *u – after heavy syllables (with nasalised *-aⁿ, at least, perhaps being very slightly slower to be deleted than the others). This creates an array of new overheavy syllables, significantly changing the overall frequencies of different syllable weights in the language. Some overheavy monosyllables had certainly existed since Proto-Germanic (§13.1.4), but they were becoming notably more common. Whether or not there was a short phase of more restricted vowel deletion, as per Kiparsky, the bulk of the inscriptions from c. 600–800 attest to a language that had readily lost vowels after heavy syllables in a way that often created overheavy syllables. As in Old English (chapter 4), this was an important prosodic shift in the language that, though it probably didn’t disturb the bimoraic trochee as such, was still of considerable importance prosodically.

It is clear that a number of vowels that disappeared by the time of literary Norse did manage to survive the initial 7th-century period(s) of vowel loss. These were lost in a second wave of deletions that, to judge by the runic evidence, took place sometime during the 9th century. As usual, there are complications. Most of the words in question concern high vowels after light syllables, but there are questions about whether all of these had survived the earlier deletion period, or if some of them had already been lost in round one.

In general, it is clear that *i and *u survived after stressed light syllables when no consonants at all followed. Examples of the elements haþu-, hari-, and h^Aeru-have already been cited (157a, 157b, 157c, 157e, 157f, 157g). There are, furthermore, a number of examples of accusative sunu ‘son’ from inscriptions up into the early 9th century (Birkmann 1995: 178).

Kiparsky (2009: 25–26) has a ready explanation for why such words would retain their final vowel: the mimimal word constraint (§10.2.1). This holds that vowel deletion can’t produce a ‘word’ (a minimal prosodic word: individual elements of compounds, including names, would count as words for this process) with fewer than two moras. The bimoraic minimum is a well-known and important constraint in Germanic generally (§13.1.1), but in this case, it is only relevant under Kiparsky’s view that all final consonants originally counted as extrametrical. This would mean that sunu couldn’t lose its final vowel, because the resulting *su〈n〉 would have been too light: one mora from the vowel, and that was it (since the consonant wouldn’t count). Note that Kiparsky argues for a divergence between the nominative and accusative forms of this and similar words: since nominative *sunur can freely lose its second vowel and still meet the minimal-word constraint, he thinks that it should have done so already in phase I. During the period between phases Ib and II, nominative *sun〈r〉 and accusative *sunu should (he predicts) exist side by side. Phase II,¹⁸ beginning around 800 or so, would be triggered by the elimination of final-consonant extrametricality, which would allow forms such as sun to finally meet the bimoraic minimum.

The hypothesis that the key change concerned final-consonant extrametricality can be criticised on theoretical grounds. Firstly, as discussed in §10.2.2, the idea that all final consonants were extrametrical is doubtful: only-r, and perhaps-s, are really likely to have been extrametrical. Secondly, it is unclear why there should be the same vowel-deletion impulse surviving for over two centuries, ready to raise its head whenever a new change in extrametricality or other general parameters takes place.

Beyond this, there is an empirical problem. A number of forms from the end of the ‘transitional’ period – shortly before the second major vowel losses took place – seem to show the retention of the high vowels not only when truly final, but also before-r. The simplest examples are the following, but the data in (164) below is also relevant:

(163) a. sunur < *sunuz (Gursten, Sm 144, 9th century)

b. magur (makur) < *maguz (Sparlösa, Vg 119, c. 800)

c. sitir < *sitiz (Rök)

Kiparsky tries to provide explanations for these forms, and draws attention to forms from the ‘transitional’ period (between Ia and II) that do potentially show high-vowel deletion. In the following subsections, I consider the relevant data from various angles: supposed evidence for deletion in sunur-type words at an earlier date (§10.3.1), evidence for earlier medial syncope (§10.3.2), and counterevidence from words eventually undergoing double vowel loss (§10.3.3), before returning to the counterevidence of words such as sunur (§10.3.5), and examining the implications of these patterns of vowel loss and retention for foot structure (§10.3.6).

Kiparsky (2009: 26) invokes a certain number of forms from 600–800 that might appear to support his view that early vowel loss would take place to the greatest extent possible, just so long as the word maintained an overall minimum of two moras (so also Riad 1992: 108–118). On inspection, however, much of this data doesn’t hold up very well, or is better explained in other ways.¹⁹

Two early forms mentioned by Kiparsky probably show consonantal developments rather than vowel deletion directly. Björketorp’s sba, assuming it really is a form of spǫ´ ‘prophecy’, from *spahu, reflects the loss of the medial *h with subsequent contraction, not the direct loss of the vowel (Noreen 1970: 167).²⁰ A similar, though more complicated, case is the nahli on Strand (KJ 18), if this really is from *nawi-hlewai ‘(for) corpse-protection’. Here the question is probably tied to the developments of *-VwV-sequences, which tend to result in a lengthening of the first vowel and loss of the *-wV-(Noreen 1970: 77). Pace Noreen, this is not likely to be simply due to vowel deletion alone, since that should have simply resulted in a diphthong in Norse: e.g. ^xstrauði ‘spread out, strewed’ from *strawiðē, rather than the actual outcome stráði. The same loss of *w is also probably seen in *hlewai > hli = hlē, where no vowel loss has taken place. Such words involving complex sequences of high-sonority consonants are very poor evidence, at best, for general vowel losses.

From rather later on in the period, the Sparlösa stone (c. 800) shows the reduced form sunr.²¹ This sunr does not, however, occur as an independent word, but as part of a patronymic formation, Airīkis sunr ‘son of Erik, Erikson’. I suspect this is an example of an exceptional reduction or remodelling in what amounts to the later part of a high-frequency type of compound word, comparable to the regular use of nominative-son rather than-sonr in patronymics in Old Icelandic (Noreen 1970: 274). Note that the same inscription goes on to supply the form magur ‘son’, without vowel reduction.

Three of Kiparsky’s examples involve the middle syllables of longer words, which makes sense: this would be an environment where the minimal-word constraint would not be an issue, and if deletion did occur inside the boundaries of the bimoraic foot during the earlier loss period(s), it would be in this position. Riad (1992: 114, 126, n. 24) also suggests that such medial syllables were deleted during the initial loss period, regardless of the weight of the first syllable. He would place the syncope of a word such as *katilōz ‘kettles’, which becomes literary Norse katlar, in this same period. The actual data is, however, at best ambiguous. There are four relatively clear examples where vowel deletion has taken place (see §10.3.4 on a complicated fifth potential case, Eggja’s nakdąn).²²

The first example is the very early reduction of *satiðē to sate, sattē ‘set, placed’ on Gummarp. This could be interpreted according to Kiparsky’s model, or taken as a simple haplology. Partial haplology is also a possibility in the second example: the participle fatlaþr ‘buckled’, from *fatilōðaz (Rök). Scribal error – omission of a single stave <ᛁ> i between <ᛐᛚ> tl – is also hard to confidently exclude. This is a potential example of a regular syncope, but a weak one.

The other two examples are both past participles of strong verbs: Rök’s numnar ‘taken’ and bornīr (burnir) ‘born’. Kiparsky does not cite these, probably because the history of the medial vowel in such words is highly problematic, and (at least to my mind) nowhere near being satisfactorily resolved.²³ One key question – though not the only relevant one – is whether such forms still contained a reflex of *a during the period in which syncope was established in such words, or whether this had already been replaced by *i analogically at such a date.²⁴ If the former, this may be evidence for earlier loss of *a after light syllables.

All in all, it seems difficult to sustain Kiparsky and Riad’s view of early medial syncope taking place hand-in-hand with the earliest wave of vowel deletions in Norse. All of these four examples involve problems, and the least useful, sate, is also the only particularly early attestation. All the other forms come from Rök, one of the last ‘transitional’ runic inscriptions before the main second period of vowel loss. Even if these forms are taken at face value as examples of syncope, it could well be that Rök simply reflects a slightly earlier application of second loss to medial syllables, following the general tendency of medial vowels to be somewhat weaker (compare §4.4.1.1).

There are admittedly no positive examples of retained medial high vowels in the corpus: no examples such as *framiðun ‘carried out’, which would show that such medial vowels definitely were retained – contrast Old English fremedon, where the medial *i was retained just as much as it would be in a shorter word such as wine ‘friend’. This leaves an almost complete void of data, with no reliable forms before Rök at the start of the 9th century. On theoretical grounds, I find it very unlikely that Kiparsky and Riad’s ideas of early medial syncope can be sustained, resting as they do on untenable assumptions about general final-consonant extrametricality, but this is not a point that I can demonstrate empirically.

Some words show sequences of a glide plus a vowel where, once the vowel was lost, the glide might be expected to vocalise into a new vowel, whose patterns of loss and retention add a further layer to consider in reconstructing the history of Norse vowel loss. Kiparsky (2009: 25–26) would see such forms as losing the new vowel in what I am calling phase Ib:²⁵ that is, as long as the minimal-word requirement would still be meant, vocalised glides should also vanish as quickly as possible.

The best example Kiparsky cites for this is niþr ‘relative’ from Rök, which seems to show double vowel loss. The earlier Germanic form would have been *niþjaz, with the *a presumably dropping in the initial vowel-loss period. This should have produced *niþiz, with the glide vocalised into a vowel. This may actually be what the Rök inscription is meant to read, according to Grønvik (1983), though this interpretation involves taking the border line as simultaneously standing for <ᛁ> i. If Grønvik’s suggestion is rejected, then the lost vowel in niþr would indeed need to be explained.

As it happens, the Rök stone also contains two other examples of words with lost *a, which retain vocalised glides that would later be lost in Norse. A further example comes from the Oklunda slab, from a similar time period:

(164) a. fiaru < *ferh^waⁿ (Rök)

b. garur (karur) < *garwaz (Rök)

c. sækir < *sakjaz (Oklunda, Ög N288, 9th century)

These become later Norse fjǫr ‘life (accusative singular)’, gǫrr ‘prepared’ and sekr ‘guilty’, respectively, but in the Rök inscription this further reduction is only potentially present in niþr. (164a) is not problematic for Kiparsky – the *-aⁿ was lost early on, as was the *h, and he argues for retention of truly final *-u during his Stage 2 – but I am not certain how the other two forms might be explained under his model. He cites niþr as an example ‘of the earlier type of syncope which began at Stage 1’ (Kiparsky 2009: 26). This implies that syncope is meant to have operated cyclically in the earliest period, so that both the *a and *i (< *j) were syncopated in short order, which should, per Kiparsky, produce ^xgǫrr and ^xsækr already in the 7th century. Why didn’t this happen?

All in all, the evidence of niþr is ambiguous. This may not even be the form intended by the carver, and if it is, it is at odds with the majority of data concerning short high vowels after stressed light syllables. The best case that could be made for it, I think, is not that it is a reflection of the first deletion period(s) some two centuries earlier, but that it might – perhaps along with fatlaþr – be among the first examples of a new, initially variable vowel deletion that would, by the 10th century, go on to become the general rule in Norse.

Of interest to both the preceding sections is nakdąn from the Eggja stone, probably carved over a century before Rök. This is widely taken as an earlier form of nøcþan ‘naked (masc.acc.sg)’ (Birkmann 1995: 100–102), which would make it cognate with Gothic naqadana (nominative naqaþs). The starting form for Norse is something such as *nak^wiðanō (with medial *i for Gothic’s a attested by the umlaut), with *k^w at some point ejecting its labial element as a distinct consonant, *kw, and subsequently geminating to *kkw (cf. the Icelandic nominative nøcquiþr).

The absolute and relative chronologies of these changes are rather obscure, however, which makes it extremely difficult to use as evidence for medial vowel loss. The key question is whether the medial vowel was lost from a geminated form such as *nøkkwiðann > *nøkkuðann > *nøkkþann, or in a non-geminated form such as *nøkwiðann > *nøkuðann > *nøkþann.²⁶ In the former case, this word is uninformative about vowel losses after light syllables, since the doubled *kk would keep the initial syllable heavy throughout the word’s history. Only in the latter case would the word be evidence for early vowel syncope. It would then push the chronology for medial loss after light syllables back considerably earlier, by over a century, from the time of Rök to the era of Eggja. But I don’t think nakdąn can bear that kind of weight as evidence: its exact route of development is simply much too uncertain.

It should now be abundantly clear that the data from runic inscriptions is not straightforward to work with. Nonetheless, the balance of evidence seems to point rather to the forms such as sunur (163) and sækir (164) as being regular developments, without reduction of a final (high) vowel after a light syllable. It is relatively easy to explain the few potential counterexamples, such as Sparlösa’s sunr, as either doubtful or as reduced by independent processes, but hard to account for the examples of vowel retention as anything but evidence that early vowel loss was less widespread than Riad or Kiparsky predict.

For example, in order to explain away sunur, etc., Kiparsky (2009: 26) invokes the highly improbable suggestion by Birkmann (1995: 178, 313) that such forms were in fact reduced, but then inserted epenthetic vowels – that is, that Rök doesn’t have sitir, but rather sitⁱr. It seems an unlikely coincidence that such epenthesis should always just happen to exactly match the etymological vowel in question, and only take place after light root syllables – there are no examples in Rök of this supposed epenthesis in words such as-rikr (-rīkr) or histr (hæ̨str).²⁷ Indeed, it is doubtful that in these inscriptions epenthesis took place in the neighbourhood of r at all, since the only compelling example in the late transitional inscriptions is Rök’s uintur, where the consonant in question is the phonologically distinct r. This attempt to explain apparent non-deletion through epenthesis can be safely dismissed.

The upshot is that Kiparsky’s explanation of vowel loss after light-root syllables is not only theoretically problematic (§10.3), but empirically makes the wrong predictions. Forms such as sunur, magur, sækir, and garur regularly retained their second vowels through the ‘transitional’ period, and were disyllabic right up until the second major wave of vowel deletions removed them, along with truly final vowels such as those in sunu, fiaru, and so on.

Kiparsky’s explanation for the second vowel-deletion period is simple and elegant – final consonants stopped being extrametrical, which meant that vowel loss could take place without violating the minimal-word constraint – but it doesn’t seem to hold up either theoretically or empirically. This naturally raises the question of just why the vowels of sunur, sunu, etc. survived the first vowel-loss period(s) in the 7th century, but were then lost a couple of centuries later in the 9th.

The simplest explanation for the initial retention of such vowels is they are within the bimoraic foot, and so protected from the scope of the first deletion period (phases Ia and Ib both). This would imply that the development of Norse fairly closely paralleled that of Old English in this way, as in so many others (notably breaking and umlaut) – without suggesting that they were identical in every detail, or even directly connected processes. Rather, it was simply that vowels in initial feet were, in general, more resistent to being reduced than were unfooted vowels. This explains most of the data very straightforwardly:

(165) a. ^*(hrai)-ði-> Hraiþ-

b. ^*(briu)-ti〈z〉 > b^Arūtz > brýtr

c. ^*(man)-ni〈z〉 > mænnz

d. ^*(stan)-di〈z〉 > stątr, stændr

(166) a. ^*(si-ti〈z〉) > sitir

b. ^*(su-nu〈z〉) > sunur

c. ^*(su-nuⁿ) > sunu

d. ^*(gar)-wa〈z〉 > ^*(ga-ru〈z〉) > garur

e. ^*(sak)-ja〈z〉 > ^*(sa-ki〈z〉) > sækir

f. ^*(briu)(-tiþ) > b^Ariutiþ > brjótið

This does not present a very detailed picture of the interactions of foot structure and vowel loss, not compared to the details of medial syncope recoverable for Old English (chapter 4). The case of longer words in particular is very difficult to recover. Assuming a general bimoraic trochee, the forerunners of fatlaþr and katlar would have been footed as:

(167) a. ^*(fa-ti)(-lō)-ða〈z〉

b. ^*(ka-ti)(-lō〈z〉)

Were these protected after all, giving ‘transitional’ *fatilaþr and *katilār? Or was syncope introduced early, despite the protection of the bimoraic foot (Riad 1992: 116–118; cf. further Schulte 2004: 10–11 for a defence of initial degenerate feet in longer words)? It would be very helpful indeed to have a clear answer to this question, but the data is, to my mind, too sparse and ambiguous to warrant any conclusive analysis (§§10.3.2, 10.3.4).²⁸

If it correct that footed (high) vowels were saved from the earlier vowel loss period – at least in final syllables, where the situation seems clearest – then a major shift in the importance of the bimoraic trochee took place during the second major wave of vowel losses during the later 9th century. One possibility is of course that the foot structure changed so that words such as sunur were no longer contained in a single foot, and there was another round of deleting unfooted vowels. It is possible to imagine that the syllabic trochee of modern Icelandic (Hayes 1995: 188–198) was already introduced at this very early stage.

On the other hand, it may be more likely that the bimoraic trochee was maintained, but with a shifting preference to align feet and syllables more closely. Compare the foot structures of sunur and sunr before and after the second round of loss, assuming a bimoraic trochee (and extrametrical r/r) for both:

In both instances, the foot remains bimoraic, with the difference lying in whether the two moras come from two different syllable nuclei, or from a single syllable’s rhyme. The old alignment could, indeed, still be maintained in forms such as the plural synir, whose vowel was long until after the operation of the second deletion period:

This structure is supported by the metrical evidence discussed in the next chapter, but on the phonological evidence alone, the foot structure of synir after the vowel losses of phase II would not be certain.

As a brief final note, I can hardly let a discussion of runic vowel losses or prosody pass without acknowledging the potential relevance of the umlaut for both. It has long been noted that Norse shows a striking weight-based contrast in how i-umlaut played out. A classic example is Norse *dōmiðē > dǿmði ‘judged’ versus *framiðē > framði ‘performed, promoted’. These are both class I weak verbs, in which the same grammatical desinence resulted in umlaut in the heavy stem, but not in the light.

This subject has been treated extensively from many angles,²⁹ but it seems hard to avoid the impression that the operation of umlaut either depends directly or, perhaps, indirectly on metrical structure. A recent and significant argument for indirect influence has been put forward by Schalin (2017a,b), who suggests Proto-Germanic *i was frequently altered to a non-coronal – and so non-umlauting – sound that he notates *ï. Only when it remained coronal *ȋ could i trigger umlaut: this coronalisation happened before *z and – significantly – when *i stood outside of a bimoraic foot (Schalin 2017a: 10–12, 37–43). Thus:

(170) *(fra-mi)(-ðē) > *(fra-mï)(-ðē) > framði (no umlaut)

(171) *(dō)-mi(-ðē) > *(dō)-mȋ(-ðē) > dǿmði (umlaut)

Schalin’s approach is insightful and interesting, but involves one major complication whose rejection, I think, both simplifies the operation of umlaut, and potentially supplies some (very indirect) evidence for the operation of the overheavy constraint in the earliest Norse vowel losses (§10.2.1).

This complication comes from Schalin’s suggestion that (pre-)Proto-Germanic *e remained distinct from unstressed *i, and developed unconditionally to coronal *ȋ. The weight-and context-sensitive developments to non-coronal *ï only apply to old *i. That is, perhaps counterintuitively, it would be old *e that would most uniformly trigger umlaut, while the effects of *i would be much more variable. This idea is, however, rather weakly supported etymologically, and depends particularly on abstracts in Northwest Germanic *-iþu, like dygð ‘virtue’, whose suffix is etymologically at best ambiguous.³⁰ The other evidence Schalin brings to bear is weaker, including the third-person singular of verbs such as ferr ‘goes’, and the noun gleðill.³¹ A modified approach, still leaning on Schalin’s overall insightful explanation, might be to assume a uniform unstressed *i in Proto-Germanic (from a merger of *i and *e), which during the first vowel-loss period becomes non-coronal unless:

Long *ī would also, unsurprisingly, remain coronal, except when shortened medially (Stausland Johnsen 2012), with some morphological complications (Schalin 2017a: 43–45; cf. Schulte 1998: 205–223). This revised framing is itself not without exceptions and counterexamples, but most of these are explained easily.³³

A point of chronological interest in this view is that umlaut occurs both in*dōmȋðē (so before the loss of this medial *ȋ), and in *dugȋð (where the *ȋ would arise, by rule 2, only after loss of the final vowel of *dugiðu; I assume here an overheavy licence already, but see Schulte 2004: 9–12 for a different view). Those two forms should only coexist if there really were two phases of early syncope, one (Ia) operating with an overheavy constraint, the other (Ib) without, as shown on table 10.2.³⁴ Unfortunately, while I find this chronology attractive and plausible, the explanation of umlaut remains extremely tricky and controversial. The very clear role of syllable weight in, for instance, class I weak verbs, allows umlaut to be safely used in a general way as support for the bimoraic trochee during the relevant period, but it is hard to press this process for any more specific prosodic details without relying on speculative and highly contested hypotheses.

Despite difficulties with the runic data, vowel deletions in Norse can be broadly grouped into two major phases: an early one, which did not affect vowels protected by bimoraic trochees, and a later one, which was much more general and insensitive to foot structure in deleting short vowels in open syllables. Within these two broad periods, there is potentially some evidence for a finer-grained chronology. Among the earliest losses, it is likely that the unstressed sequence *-ija-was reduced to *-ī-even before other deletions took place (§10.2). More interestingly, there is some evidence – though none of it is decisive – that vowel loss was at first restrained by a limitation against creating overheavy syllables (§10.2.1 and §10.4). If this restriction really ever held, it was relaxed by the later 7th century at the latest. During these earlier phases, vowels protected by bimoraic trochees were retained, as shown by forms such as sunur and sitir, which are still in evidence on the Gursten and Rök stones at the start of the 9th century (§10.3). It was left to the second phase of vowel deletions to delete these remaining short vowels in open syllables and before final *-r. There is slight evidence that this process, too, may have taken place in two stages: first in medial syllables (§10.3.2), and later in final ones (§10.3.6). Taken together, the following overall chronology might be suggested:

Generally speaking, the maintence of forms such as sunur through the first deletion period(s) is most straightforwardly explained with reference to a bimoraic foot, whose presence is also suggested by the operation of umlaut (§10.4). There should be nothing surprising about this kind of chronology. Vowel loss waves need not form a single process, as the developments in Old English amply demonstrate: compare the loss of *u in *handu > hand ‘hand’ around perhaps 600 with the weakening of u to schwa in sunu > sune only achieved half a millennium later, and only being actually lost in southern English in the 14th century (or perhaps even later in some dialects). In some Flemish varieties, the final vowel of zeune [zyːnə] < *sunu still survives (Taeldeman 2013: 215–216), while that of hand was lost at an uncertain date before the earliest records of any kind of Dutch.

What is striking in the Norse losses of unstressed vowels is not that they should have taken place over multiple stages which operated over the span of three centuries or so (as if this were a long period of time for so many reductions), but that they should have been so swift and far-reaching. In a wider Germanic context, the deletion in words such as sunur occurs remarkably early in Norse.³⁵ Such deletions from the second major phase can be framed in terms of the moraic trochee, but they represent a significant innovation prosodically, leading to a much greater alignment of foot and syllable than had been the case previously. While this in itself did not lead North Germanic to shift at once to a syllabic trochee, it may have allowed that shift to occur more easily later on.