4 The Hēafudu-problem: Early Old English Foot Structure

The most direct evidence for prosodic patterning in Old English comes from a process commonly known as high-vowel deletion (often abbreviated to HVD). This refers both to a historical change in prehistoric Old English, and to a set of morphophonemic alternations in recorded Old English that maintain the effects of this sound change in the morphological system. Especially in its earliest operation, high-vowel deletion is an important window into Old English phonology. It attests to the presence of the bimoraic trochee foot in Old English, and gives an anchor for working out the chronology of vowel shortenings and reductions in the language.

High-vowel deletion involves the loss of unstressed, short, high vowels – *i and *u – in certain phonological contexts. The basic operation can be illustrated most easily with the nominative-accusative ending of strong (a-stem) neuter nouns, which in prehistoric Old English had the shape *-u. This was originally a simple suffix, and words such as *scip ‘ship’, *word ‘word, utterance’, and *ję̄r ‘year’ had corresponding plural forms *sci-pu, *wordu, and *ję̄-ru. Note where the syllable boundary falls in the plural forms: in*sci-pu, the initial syllable is *sci-, ending in a short vowel and so counts as a light, monomoraic syllable (cf. §2.3), while the other two both have heavy, bimoraic initial syllables, ending either in a consonant (*wor-) or in a long vowel or long diphthong (*ję̄-).

This distinction in syllable weight was a key conditioning factor in high-vowel deletion as a sound change. ‘Light-stemmed’ neuters – those which had a light initial syllable once an ending was added – such as *scipu retained the ending throughout the historical Old English period (written as scipu or scipo). The heavy-stemmed neuters, by contrast, lost the ending, making the nominative-accusative plurals the same as the singular, word and gēr (West Saxon gēar, etc.). Exactly when this took place as a new sound change is hard to pin down with complete precision, but it was clearly in the relatively late prehistoric period (certainly after umlaut), and an estimate of very roughly around the year 600 is probably approximately correct (Luick 1921: 287; Ringe & Taylor 2014: 292). The alternations between forms with and without-u created by this change were subsequently retained for several centuries. They are very regular in texts produced up to around the turn of the millennium, though they are gone by the time we get texts conventionally thought of as early Middle English.¹

The light-heavy distinction in monosyllabic bases – between scip and word – already suggests that high-vowel deletion has a prosodic dimension of some sort. This impression is strengthened when we consider polysyllabic bases. The most robust cases are words such as werod ‘troop, war-band, army’,² which may be notated as LL-stems (§2.3). The first syllable, we-, is always light, and the second is light after the addition of a vocalic inflection, as in the historical plural form *we-ru-du.

Words of the werod-type lose their final vowel, just like the word-type and in contrast to the scip-type: prehistoric *werudu becomes werod. Traditional grammars usually just list types of words that show high-vowel loss and those that do not, without making any real attempt to generalise why word and werod should be in the same group (Campbell 1983: 144–147, Sievers 1965: 124–127). However, if we think in terms of the bimoraic trochee – a foot type introduced in §2.5.1 – it is easy to see how those two should form a group against the scip type.

The following trees show the three type-example words (ω) arranged into bimoraic trochee feet (F), each of which has two moras (μ). Syllables that don’t fit into a foot are left stray, and it is these stray syllables that are of particular interest here:

Under this arrangement, which involves a simple, typologically normal division into bimoraic feet, a potential conditioning factor for high-vowel deletion clearly emerges: the high vowel (in this case *u) is retained when it is included in an initial bimoraic trochee, but is otherwise deleted. In the more compact bracket notation, the distinction is: *(sci-pu) versus *(wor)-du and *(we-ru)-du.

The equivalence of the sequences wor-and weru-may be compared to the metrical phenomenon of resolution (see §3.1.1), a link whose importance was emphasised by Kuryłowicz (1949); cf. Fulk (1995). Although resolution is properly speaking a metrical term, it may be usefully applied to phonology as well, and I will sometimes refer to the grouping of two monomoraic syllables into a single prosodic unit as phonological resolution.

Although the importance of a foot-based analysis along these lines has been recognised for decades, the precise formulation of high-vowel deletion has remained a matter of considerable controversy. This is partly a matter of differing formal or theoretical assumptions, but also (and more importantly) a question of working out just which outcomes of high-vowel deletion in more complex words are regular, and which are the result of analogical restructurings. This is partly a philological problem of determining precisely what the early and dialectal evidence says.

Theoretically, many models have sought to find a way to single out the position immediately after a heavy foot as in some way ‘special’, characterising this as the target environment for high-vowel deletion. Keyser & O’Neil (1985: 8–10), in a flawed but foundational analysis, posited ordinary bimoraic feet, which they claimed are right-headed (making them bimoraic iambs). For them, high-vowel deletion involves reductions immediately following the strong element at the end of each foot. Thus (marking strength with an acute), *(we-rú)-du would delete the final vowel because it follows the strong head of the foot. This view is taken up in all essentials in the grammar of Hogg (2011: 222), though he avoids the word ‘foot’ as such.

This view has a number of theoretical problems, the most obvious of which is that foot structure is, among other things, the basis for stress assignment, and it is clear that Old English stress is left-headed, not right-headed. The stronger syllable of werod is certainly the first. This is established both on metrical grounds (Minkova 2003: 24–34), and by the phonological development of the word itself (see further Hutton 1998). The initial syllable is subject to diphthongisation, often becoming wĕorod, etc., a process which is limited to stressed syllables. By contrast, the second syllable is frequently reduced: the lowering of werud to werod is typical only of unstressed u,³ as is the eventual further reduction to [ə] suggested by the very frequent spelling wered. Keyser & O’Neil thus have to adopt one analysis for foot structure to explain high-vowel deletion, but posit another, unrelated rule for stress assignment on the initial syllable of a root.

These criticisms were raised in an important article by Dresher & Lahiri (1991), who proposed a different kind of foot – dubbed the ‘Germanic foot’ – to better account for both high-vowel deletion and other stress phenomena (as well as metrical resolution) through a single formal apparatus. Specifically, they argue for a complex foot that has two components, an initial bimoraic ‘strong branch’, and an unstressed and monomoraic ‘weak branch’. Under this view, a word such as *werudu would be entirely footed, with (we-ru-) filling out the strong branch, and (-du) attached as the weak branch. High-vowel deletion would then simply be the elimination of high vowels in a weak branch.

In practical terms, the results of this analysis are virtually the same as Keyser & O’Neil’s. A bimoraic unit of some sort is still posited (though recast as merely the strong branch of a larger foot rather than the entire foot), and the immediately following position is highlighted for vowel deletion. Theoretically, however, Dresher & Lahiri’s approach represents a significant advance by formalising the idea of metrical coherence (already implicit in the rather briefer treatments of Kuryłowicz 1949, 1970),⁴ which rightly remains fundamental in all investigations of Old English foot structure (for the metrical dimensions to this concept, see §3.5). Metrical coherence holds that we should aim for a unified account of a language’s prosody (both in phonology and verse), and that severe discrepancies such as Keyser & O’Neil’s right-headed feet but left-headed word stress should be avoided. Although a closer review of the data in the following sections will suggest that the ‘Germanic foot’ of Dresher & Lahiri is not necessary or sufficient to account for the original operation of high-vowel deletion, the principle of metrical coherence remains an essential contribution to the problem.

Before returning to the question of how to understand high-vowel deletion, and what model of prosody best accounts for it, there is a major data problem that needs to be addressed. So far, I have introduced three relatively simple word-shapes: LL (*scipu), HL (*wordu), and LLL (*werudu). There is no real doubt about high-vowel deletion in words of these shapes: LL words escape it, the other two are affected by it.⁵ There are, however, three potential further types of polysyllabic bases with unstressed high vowels to consider: LHL, HHL, and HLL. The first pattern will be discussed later on, in §4.5.1.2, and the second is relatively uninformative in terms of prosody. The third type – HLL (along with its inflectional variant HLH), which can be exemplified by the neuter plural hēafudu – is by contrast potentially highly informative, but also presents the most complexities in determining what the regular outcome of high-vowel deletion really is. The following sections will outline the main issues by focusing on the inflection of hēafud ‘head’; I will return to the fuller complexities of the data in §4.4.

The neuter noun hēafud or, much more frequently, hēafod ‘head’ presents a particularly interesting case for high-vowel deletion. In this form (the nominative-accusative singular), no deletion takes place, nor would it be expected to since the final syllable is closed. There is, however, regular deletion of the historical *u in the genitive and dative forms, both singular and plural, such as dative singular hēafde and plural hēafdum.

Complications arise when we look at the remaining case forms, the nominative-accusative plural. Historically this was formed with the same suffix *-u we have been dealing with so far, and the pre-deletion form was something such as *hæ⁀uβudu,⁶ with two potentially deletable vowels following the heavy root syllable. There are four logical possibilities for how high-vowel loss could affect this word:

As implied by the lack of any qualifying ^x, possibilities 2–4 all actually occur in the corpus of surviving Old English manuscripts (both as such and with further variations that I will gloss over as not directly relevant, e.g. hēafodo). And even the non-occurring ^xhēafd has been alleged to be the regular outcome of high-vowel deletion, with the attested forms all being analogical reformations (Ringe 2002; Ringe & Taylor 2014: 301–302, 377–378). To my knowledge, no one has seriously argued that option 4, hēafud, is the regular outcome of high-vowel deletion, but all the others have been argued for in recent scholarship.

There are two principle techniques for determining which of the many attested variants is the original, and which are later restructurings. The first is close philological evaluation of the attested forms (privileging archaic and pan-dialectal forms). The second is the consideration of morphological change: if some (or even all) of the attested forms are later analogical creations, these need to have arisen through plausible morphological processes. Both approaches fortunately converge in this case, strongly supporting option 2, hēafudu, as the regular form.

Philologically, we have relatively early forms of hēafud’s plural in Mercian and West Saxon, the former showing both hēafud (5x) and hēafudu (2x), the latter showing hēafdu or its morphological variant hēafda six times, alongside one example of hēafudu.⁷ In later Mercian, relevant forms are few, but show only the form hēafud (1x, with two more occurrences in a Late West Saxon text likely copied from Mercian). In West Saxon proper, hēafdu and hēafda predominate. A reasonable assumption is that in both dialects hēafudu is an archaic form replaced by varying innovative plurals: endingless hēafud in Mercian, syncopated hēafdu in West Saxon.

This is, however, not conclusive on its own. The evidence of Northumbrian is open-ended: all three attested variants are found, despite the number of overall attestations being low; all the evidence is late; and the possibility of influence from other dialects (including the Late West Saxon scribal quasi-standard) is very possible. It is possibly significant that hēafodo occurs at all – it is the only plural found in all three dialects, and its appearance in Northumbrian can hardly be attributed to Late West Saxon influence – but the number of tokens is small enough, and the number of variants large enough, that while hēafudu is the most plausible candidate for an original plural form, its position can hardly regarded as a certainty on philological grounds alone.

Consideration of morphological change can help us rule out possibilities 1 (^xhēafd) and 3 (hēafdu). If either of these were the original regular form, it would mean that the medial *u had been deleted regularly, and so where it does occur (i.e. plurals hēafudu and hēafud) it must be restored by analogy. It is particularly difficult, however, to explain hēafudu as the result of such an analogy, especially given that it co-exists with forms such as hēafde and hēafdum, where medial deletion is regular in all but the latest Old English texts.

To be specific, if the original form was hēafdu (or ^xhēafd), alteration to hēafudu would have to occur on the basis of the singular, hēafud. That is, the morphophonemic alternation of syncope created by high-vowel deletion would be eliminated through paradigm regularisation. This is a normal enough kind of development, but if it happened, it would be expected to actually regularise the paradigm. Either the medial-u-should be restored throughout (hēafud, hēafude, hēafudu) or else the supposedly original syncope should be retained (hēafud, hēafde, hēafdu). But this is not what happens, not at first in any dialect and not at all in Mercian. Instead, the plural hēafudu is found in early texts and in all dialects in which this word is attested, while forms such as hēafode are late and dialectally restricted.⁸ What analogical process could have restored the medial vowel in the nominative-accusative plural, while failing to do so in the dative singular or genitive plural?

On the other hand, if we start from a paradigm that already had a distinction between nominative-accusative plural hēafudu and dative singular hēafde in its oldest form, all the attested variants can be easily accounted for. Occurrences of hēafudu are simply archaisms, repeating the original pattern unaltered. The common West Saxon form hēafdu (or hēafda) represents an analogical extension of medial syncope from forms such as hēafde and hēafdum, generalising a regular rule: drop the medial syllable when a grammatical suffix is added.

The plural form hēafud follows a different analogy, based on the tendency for strong neuter nouns to have identical forms in the singular and plural nominative-accusative: just as the plural of word is word, so the plural of hēafud could be hēafud (compare Luick 1921: 286, who, less plausibly, looks to werod as the analogical basis for restructuring). This analogy would have been assisted by the presence of forms with secondary epenthetic syllables. The strong neuter *wuldr ‘glory’, for instance, originally had a plural *wuldru. This was precisely parallel to *word and *wordu, and the plural ending was dropped by regular sound change. However, unlike *word, *wuldr (now both singular and plural) ended in a consonant cluster of rising sonority, an awkward situation which was resolved by adding an epenthetic vowel: *wuldr > wuldur. Words affected by vowel epenthesis formed a distinct class in West Saxon, and were altered further (Bermúdez-Otero 2005: 22–24, 49–53), but in Mercian they provided solid class of HL neuters with identical singular and plural forms in the nominative and accusative. It is therefore unsurprising that it is precisely in Mercian that the plural hēafud is best attested (Fulk 2010: 134–135). The following sets of forms make it easy to see how wuldur would provide a ready model for an innovative plural hēafud:

On the other hand, it is difficult to see what kind of analogy might have produced hēafudu as secondary in this dialect, without also producing forms such as ^xwundru or ^xwuldru. Such forms are absent in the Vespasian Psalter, a valuable source for Mercian of, probably, the earlier 9th century (Kuhn 1965: v–vi; Toon 1983: 80).

Both philological and, especially, morphological considerations accordingly point to an original nominative-accusative plural hēafudu, which occurred in the same paradigm as regularly syncopated forms such as hēafde, and which was adjusted by straightforward analogies variously to hēafdu (especially in West Saxon) or hēafud (especially in Mercian). Using the relatively archaic variants from the Old Mercian of the Vespasian Psalter to represent early-ish historical Old English, the full paradigm would be as follows (the dative and genitive plurals of this word happen not to be attested in this text, but their forms are not in any doubt):

If this paradigm is correct,⁹ it raises a further question, which leads to the crux of the matter: how did the contrast between the dative singular hēafde and nominative-accusative plural hēafudu come to be?

In a synchronic grammar of ‘classical’ Old English (whether the Early or Late West Saxon literary norms, or the Mercian of the Vespasian Psalter), a distinction in vowel loss between hēafudu and hēafde is difficult to motivate. A coherent explanation may, however, be formulated for late prehistoric Old English, which is after all when high-vowel deletion first applied. Simply put, my proposal is that syncope of the medial *-u-depended on the weight of the following syllable: an *HLL sequence underwent no vowel deletion, but an *HLH sequence did, to become HH.

For two case endings, this contrast can be seen even in later Old English. The only thing that distinguishes the nominative-accusative plural ending-u from the dative plural-um is an extra final consonant – that is to say, an extra mora. In pre-deletion Old English *hæ⁀u-βu-dum would therefore have been an HLH word, while *hæ⁀u-βu-du would have been HLL. The operation of deletion can then be explained easily by applying a simple bimoraic foot structure: *(hæ⁀u)(-βu-du) can be exhaustively parsed into two bimoraic feet, while *(hæ⁀u)-βu(-dum) cannot. The monomoraic medial syllable would not fit into either foot without making it trimoraic. It was left as a weak, unfooted syllable, which made it open to deletion.

Much the same also applies to the genitive singular hēafdes, from *hæ⁀uβudæs, but what about dative singular hēafde and genitive plural hēafda? This account requires that they end in long vowels at the time of high-vowel deletion, while the nominative-accusative plural ending *-u was short: dative *hæ⁀uβudǣ and genitive *hæ⁀uβudā, against nominative-accusative *hæ⁀uβudŭ.

This length contrast is plausible (indeed, necessary), but the chronology of final-vowel shortenings in Old English requires a bit of discussion. All of the endings under discussion historically come from long vowels or diphthongs if we go back to Proto-Germanic:-u from *-ō, dative-e from *-ai, and genitive-a from, according to the mainstream view, *-ôⁿ.¹⁰ They all ended up as short by ‘classical’ Old English, and were further reduced stepwise to schwa (Kitson 1997) and then lost over the course of Middle English. The key problem concerns the precise dates of shortenings (for there was, I argue, more than one round of shortening) before the ‘classical’ Old English stage.

The raising of absolutely final *-ō (the plain kind, neither trimoraic nor nasalised) to *-ū̌ appears to be a common Northwest Germanic development. It took place universally in West Germanic, and is witnessed in Early Runic inscriptions (Krause 1971: 88) – the earliest likely example to be directly attested is the form mīnu ‘my (fem.nom.sg)’ on the Opedal stone, dated to c. 350 by Antonsen (1975: 40), and to the early 400s by Krause & Jankuhn (1966: 177–178; Nielsen 2000: 85). This raising was probably closely followed by, or simultaneous with, a general shortening of high vowels in absolute auslaut (Ringe & Taylor 2014: 14–16): Proto-Germanic *hildī ‘battle’ became Northwest Germanic *hildi, and the vowel is treated uniformly as short throughout North and West Germanic. There is therefore little doubt that the neuter plural ending *-u, which has featured so much in the discussion so far, had been short for many centuries by the period of Old English high-vowel deletion.

Non-final high vowels were not shortened as part of this process. This can readily be seen in North Germanic, where short *i (including when newly shortened) was generally lost, but preserved *ī was retained, and only shortened later on: *gastiz ‘guest’ > gestr, *anþi ‘and, still’ > enn, and *hildī > *hildi → hildr (the addition of the-r is analogical); but *gastīz ‘guests’ > gestir, with the non-auslaut *ī escaping early loss, attesting to its length at that time. The length of medial *ī was also preserved through the syncope period in combining forms in compounds, such as *hildī-tanþu ‘battle-tooth’ > hildi-tǫnn, an epithet of an early Danish king.

A similar contrast is also apparent in West Germanic, as evidenced by high-vowel deletion itself. In contrast to North Germanic, there was a very early loss of final *-z, so that *gastiz became *gasti. This final short *-i stuck around long enough to trigger umlaut, but was eventually lost by high-vowel deletion, producing gest, giest, gyst, gist, etc. (the vocalism varies dialectally). Originally final long *-ī shows the same development, as shown by, e.g., *hildī > *hildi > attested hild. By contrast, in originally non-final position the length is retained through high-vowel deletion, preventing the loss of the vowel. Plural *alβīz ‘elves’ accordingly becomes later Old English ylfe,¹¹ without undergoing high-vowel deletion. This suggests that the sequence of sound changes was as follows:

This also fits nicely with 1 being a common Northwest Germanic change, and 2 being a strictly West Germanic innovation. As in Old Norse, compound forms such as *hildī-burdaⁿ ‘battle-board, shield’ also escape early shortening, and consequently also high-vowel deletion, becoming hilde-bord.¹²

New instances of long *-ī were also created from the reduction of the common endings *-ijaz and *-ijaⁿ, e.g. *rīkijaⁿ ‘dominion’, which became prehistoric Old English *rīcī, later rīce. This seems to have fallen together with the unshortened instances of Proto-Germanic *ī, giving late prehistoric Old English, at the time of high-vowel deletion, a clearly reconstructible contrast of vowel length between unstressed (including word-final) *i and *ī, both from various sources.¹³

The question remains of how the various non-high vowels fit into this system quantitatively. Qualitatively,¹⁴ these final vowels and diphthongs (and sequences which eventually became final) showed a gradual tendency to merge, first into a set of three non-high vowels in West Germanic, *-ō, *-ā, and *-ē, and then into just two, *-ɔ̄ (from *-ō) and *-ǣ (by a merger of the other two). This last development occurred in the so-called Ingvaeonic (or North Sea Germanic) subgroup, which included Old Frisian and Old Saxon as well as Old English. At this stage there was a ‘square’ system of unstressed vowels: high *i and *u, non-high *æ and *ɔ (representing only quality, with no prejudice yet as to length).

It is probable that the non-high vowels were unaffected by the Northwest Germanic shortening of final *-ī and *-ū, a view reflected in my labelling of this as ‘high-vowel shortening’. The monophthongisation of *-ai (and *-ōi) to *-ē postdates the divergence of North and West Germanic (Ringe & Taylor 2014: 24–27), suggesting that long non-high final vowels continued to play a role in the system even after the shortening of the high vowels. Further evidence for this retention of length comes from Kaluza’s law in Beowulf, which will be discussed in the following chapter. To anticipate it slightly, there is metrical evidence that the earliest Old English poetry still made length distinctions in final vowels, and that *-ǣ and-ā (from Ingvaeonic *-ɔ̄) were both still long. There is no positive evidence for the shortening of final non-high vowels at any point in the prehistory of Old English.

The ‘square’ Ingvaeonic system survives into early Old English, and is robustly attested in Old Saxon (Klein 1977; Boutkan 1995: 152–162). In the majority of Old English manuscripts, however, it has been reduced to three, a ‘triangular’ system, through the merger of i and æ as e. This shift from a square to a triangular system likely took place in the middle of the 8th century in Mercian, and the 9th in Northumbrian (Dahl 1938: 196; Fulk 1992: 386–390). It seems likely, though not certain, that there was a stage (though perhaps a short-lived one) between general final-vowel shortening and the merger of the two front vowels. The development of vowels in (absolute) final position according to these assumptions is laid out in table 4.3 (nasalisation is only noted where this makes a difference, so *i also includes *iⁿ, etc.).¹⁵

If we allow that final *-ǣ and *-ā were indeed long in late prehistoric Old English, then the operation of high-vowel deletion in the paradigm of hēafud falls out nicely:

Syllables are parsed into bimoraic trochees, and unfooted high vowels are deleted. This may also be represented using trees, using the nominative singular and plural and the dative singular as representative forms:

The great strength of this hypothesis is that it can account for the discrepancy between syncope in forms such as hēafde, and its apparent lack of original operation in hēafudu (and similar data presented in §4.4.1). It also explains why this morphophonemic alternation eventually broke down, and was prone to restructuring in the later dialects: once final *-ǣ did shorten (and eventually turn into-e), the motivation for syncopating before-e but not before-u – both now simply short vowels – was opaque. Different varieties of Old English adopted different solutions, West Saxon regularising syncope by extending it into the nominative-accusative plural, and Mercian generalising the endingless strong neuter pattern on the model of wuldur.

Insofar as this model is successful in explaining the data of Old English, including this otherwise puzzling morphophonemic alternation, it in turn provides further evidence for the retention of length in final non-high vowels until at least the late prehistoric period (and thus serves as an important phonological corroboration for the evidence of Kaluza’s law). It also suggests that the simple and typologically common bimoraic trochee is not only sufficient to account for early Old English, but is a preferable model to any alternative proposed so far, at least for the late prehistoric period. The following sections will elaborate on each of this points in turn, first (§4.4) briefly (at least relative to the scope of the potential evidence) surveying the wider data of high-vowel deletion – both to strengthen the empirical case in favour of this model, and to address potential counterexamples – and then (§4.5) outlining more explicitly and comprehensively the theoretical picture of early Old English prosody suggested by this analysis.

The argument so far has focused on the illustrative example of hēafud, but an analysis is only as good as the data it accounts for. There are of course many more forms that have been affected by high-vowel deletion, and the picture they present is on the face of it very messy indeed. Under any analysis, a great deal of the data must be explained as secondary, arising from morphologically driven innovations. As emphasised already with regard to hēafud, the strongest arguments in favour of the model of high-vowel deletion advocated for here are morphological: that the current account can derive all attested forms either by sound change or straightforward analogy (not necessarily of the old-fashioned four-part type, but always in line with what is observable in normal linguistic development around the globe), while other accounts require unmotivated or typologically strange analogies to derive the full range of forms actually found. I cannot fully treat all the relevant data in this section,¹⁶ but there are several pieces of evidence that need to be addressed as part of the empirical basis for the bimoraic trochee in Old English.

This morphological dimension is what makes the paradigmatic alternations in hēafud so important. It is significant, therefore, that this pattern is not isolated to a single lexeme. As noted by Fulk (2010: 133–135), in the 9th-century Mercian of the Vespasian Psalter exactly the same allomorphy is found quite generally in words with a heavy initial syllable followed by a syllable of the shape *-uC-or *-iC-: syncope fails in, e.g., lȳtelu ‘little (neut.nom.pl)’, but occurs in lȳtle (masc. acc.pl), which is from *lūtilǣ. Fulk’s article should be consulted for an extensive review of this data.

Further important evidence comes from nouns such as the strong neuter nēten ‘animal, livestock’, which are distinguished by having had an originally long second vowel: pre-deletion (and pre-umlaut) *næ⁀utīn. In Mercian, these seem to have fully merged with the hēafud-type, with a nominative-accusative plural nētenu, but a genitive plural nētna. This alternation is, like that in hēafud, very challenging to explain under the traditional view of high-vowel deletion (where deletion after a heavy foot should either apply to all forms or none), or under Ringe’s alternative view (Ringe & Taylor 2014: 300–302).

Ringe sets up a relative chronology that splits high-vowel deletion into two phases, first a medial syncope, and later a final apocope. He suggests that medial shortening occurred between the two, which can yield the nominative-accusative plural well enough: *nētīnu undergoes no change by syncope, since the medial vowel is still long, and then shortens to *nētinu, which escapes apocope because the two short vowels form a single foot. Unfortunately, this theory also predicts a lack of syncope in the genitive plural, since the medial syllable of *nētīnā would still have been long when that took place. As with hēafud, Ringe’s theory has no means of explaining why syncope should take place before endings such as *-ǣ and *-ā, but not before *-u.

Instead, the simplest explanation for this Mercian data is to assume the following relative chronology:

The full paradigm may then be explained by adopting the same assumptions developed above for hēafudu, that words were formed into bimoraic trochees, and unfooted high vowels were deleted:¹⁷

These Mercian morphophonemic alternations, going well beyond the evidence of hēafud alone, constitute the central evidence for the model of high-vowel deletion I propose. Nonetheless, the wider picture is complex, and there are two potential complicating sets of data in particular that need to be addressed. The first concerns a potential challenge to the relative chronology I have just proposed, posed by the development of class II weak verbs. The second consists of the West Saxon cognates of nēten and other words of its type, whose inflections differ crucially in certain respects. Both complications can, I think, be satisfactorily addressed under the model I am proposing here, while the Mercian alternations remain effectively inexplicable under other approaches proposed so far.

The so-called second or ō-class of weak verbs in Germanic had a stem formant *ō or *ô throughout the paradigm. While the present stem was restructured in ways that are not of immediate concern (see Cowgill 1959 for details), the preterite stem was formed with *-ōd-in West Germanic, which became Ingvaeonic *ɔ̄d. The vowel had two main outcomes in Old English: either u~o or a.¹⁸ The conditioning for this divergence was long thought to be Van Helten’s rule, a proposed West Germanic sound change according to which *ō was raised to *ū before a following *u: this produced attested u~o, while unraised cases developed as normal for *ɔ̄, being eventually lowered to a. Under this proposal, a verb such as West Germanic *wundōdun ‘wounded (pret.3pl)’ would become wundodon, while *wundōdē (pret.3sg) or the past participle *wundōd would become wundade and (ge·)wundad.

This account has clear implications for the relative ordering of medial shortening and high-vowel deletion, and if accepted would strongly point to shortening only occurring after (medial) deletion. If *wundōdun regularly became *wundūdun, then medial shortening should give *wundudun, which should in turn undergo deletion to become ^xwundun or the like. There are no traces at all of such a form ever having existed. There would, certainly, have been analogical pressures at work to restore the lost vowel, both from short-stemmed verbs such as bodudon ‘preached (pret.3pl)’, where deletion would not take place, and from the tendency in some dialects to generalise the a-formant, creating clearly analogical forms such as bodadon, cēapadon ‘they sold’, etc. But that such analogies would have so thoroughly and so early on eliminated all traces of syncope seems unlikely, and it is much more preferable to seek an explanation under which high-vowel deletion simply never affected class II weak verbs at all, or did so in a much more restricted manner.

Fortunately, the validity of Van Helten’s rule in general has, on quite different grounds, recently been cast into serious doubt by Stausland Johnsen (2015), who has shown that in Early West Saxon, the data is strongly at odds with the predictions of that account. Instead, a different conditioning seems to have applied in prehistoric Old English, which is still statistically reflected in the distribution of variants in attested Early West Saxon forms: medial *ō was weakened ultimately to u~o, while in final syllables it became a. Thus both *wundōdun and *wundōdē develop alike, to wundodon and wundode, respectively, while the participle *wundōd is the original locus for the development into a, wundad.

This proposal is congruent with the metrical evidence for class II weak verbs, which supports the chronology of medial shortening predating final-syllable shortening. This can be seen most clearly in verses such as the following (Sievers 1893: 126; Russom 1987: 45–46):

(56) þrēatedon þĕarle

‘threatened forcefully’ (Beowulf 560a)

Long syllables in the equivalent position are disallowed, and we do not find verses such as:

(57) ^xScyldinga þēoden

‘prince of the Shieldings’ (cf. Beowulf 1675a)

A word such as Scyldinga would have a clear secondary stress on its medial syllable,-din-, which would constitute its own bimoraic foot. This would produce a verse of the rhythm SswSw,¹⁹ an ‘^E*’ type that is prohibited in normal Old English verse (§3.1.6).

The fact that þrēatedon + Sw is acceptable (scanning as type A, SwwSw) while Scyldinga + Sw is not points to a difference between shortened medial *-ɔ̄-and unshortenable medial *-in-. The divergence is further reinforced when we look at verses such as:

(58) ĕorlscipe efnde

‘performed (acts of) heroism’ (Beowulf 2133a, 3007a; cf. 2535a, 2622a)

Here the second syllable is historically short, *-sci-, and is allowed to form part of the dip of a type A verse (like 56, it scans as SwwSw). That is to say, the medial syllable of þrēatedon patterns metrically with historically short vowels in open syllables (which were always light), and against closed syllables (which were and remained heavy). The clear inference is that medial long vowels had already become short.

Since Kaluza’s law in Beowulf suggests that absolutely final long vowels were still long (see chapter 5), the relative chronology would have to be medial-syllable shortening first, followed by final-syllable shortening. That is to say, a singular form such as *þræ⁀utɔ̄dǣ (the singular counterpart to *þræ⁀utɔ̄dun > þrēatedon) would have gone through a stage with a short medial but long final vowel: *þræ⁀utɔdǣ or the like.

The relative place of high-vowel deletion in the chronology is more difficult to resolve. In comparison to Van Helten’s rule, Stausland Johnsen’s account of class II weak verbs reduces the number of problematic forms greatly: much of the data would simply have no relevance to high-vowel deletion at all. Specifically, no finite forms of such verbs would have undergone high-vowel deletion, whatever the relative chronology, as long as we assume that the initial change really was one of shortening alone. If prehistoric Old English *wundɔ̄dǣ became *wundɔdǣ by medial shortening, then its stem vowel would not be a high vowel. That it later merged with *u to give a single round unstressed vowel, eliminating unstressed *ɔ as a distinct phoneme, would not be surprising, as this would have been present only in a single morphological category (though a very common and productive one), but this raising all the way to *u is not a necessary immediate consequence of shortening.

Stausland Johnsen notes this, but argues on other grounds that this shortening took place after high-vowel deletion. His pool of evidence is rather small, however, coming entirely from inflected forms of the past participle – namely the feminine nominative singular and the neuter nominative-accusative plural with the ending *-u, e.g. *wundɔ̄du ‘wounded (fem.nom.sg)’. If we reckon with straight phonological development alone for such forms, Stausland Johnsen’s conclusion is reasonable. The order of shortening followed by deletion would incorrectly give *wundɔ̄du > *wundɔdu > ^xwundudu (^xwundodo), an almost non-occurring type (see below for the sole possible example). He therefore prefers the ordering of deletion first, and then medial shortening, with *wundɔ̄du becoming first *wundɔ̄d, and then the attested wundad.

It is not clear, however, that we should expect all forms of the past participle inflection to be sound-change outcomes of older inflections. In particular, there would have been no shortening in common forms such as the nominative-accusative singular for the masculine and neuter, and the unshortened variant *ɔ̄ could well have been levelled from there into other forms. In Anglian, this analogy proceeded to the point that-a-is the usual formant, not just in the participle but throughout the preterite for heavy-stemmed members of the class (Sievers 1965: 335–337; Dresher 1985: 47), and examples such as ge·myclade ‘made great (masc. nom.pl)’ occur in Early West Saxon. If such restoration took place before high-vowel deletion, then loss of the ending would be regular: *wundɔ̄du > *wundɔdu → *wundɔ̄du > *wundɔ̄d > wundad.

Furthermore, and perhaps just as importantly, the inflection of these participles was potentially also influenced by short-stemmed members of the class. In *luβɔdu ‘loved (fem.nom.sg)’, deletion of the inflectional vowel would have been regular, even with early medial shortening: *(lu-βɔ)-du > lufad^*. During the period of textual production, it would not be surprising for the lack of inflectional-u to have been generalised in class II participles, regardless of the weight of the root syllable.

Direct evidence for the ordering of medial shortening before high-vowel deletion is not something the class II weak verbs can easily supply, once we reject Van Helten’s rule. A possible relic of the original development – if medial shortening is indeed earlier than high-vowel deletion – is ge·āgenudu ‘owned, i.e. own (neut.nom.pl)’ (Cura Pastoralis 9.3.1), from *ʝi-āɣnɔ̄du. This is an isolated form in a relatively early text, and so while it could be a sporadic analogical innovation, it could also plausibly be a residue of the phonological outcome of this inflection.

Clearly the evidence of class II weak verbs is not definite. Stausland Johnsen’s rule gives good evidence for shortening occurring earlier in medial syllables than in final ones, but when high-vowel deletion fits in is harder to determine. It comes down to a relatively small number of past participle forms, which are reasonably viewed as being subject to multiple analogical pressures. Against this we may set the Mercian inflection of nēten. One of the two, either feminine wundad or plural nētenu, should be analogical.²⁰ With the rejection of Van Helten’s rule, the remaining evidence of wundad is not, in my view, particularly significant, being synchronically transparent in later Old English and easily explained as the secondary product of analogies. I continue to prefer the chronology based on accepting nētenu/nētne as the regular outcome of sound change, and not a plausible analogical innovation.

The Early West Saxon inflection of nīeten ‘animal, livestock’ (the dialectal equivalent of Mercian nēten) presents an interesting puzzle for any account of high-vowel deletion. In contrast to the Mercian forms, which show medial syncope before historically heavy endings but retain the vowel medially before historically light endings, the West Saxon forms show no syncope at all: the nominative singular is nīeten with a corresponding plural nīetenu (such as Mercian nēten, nētenu), but the genitive plural is nīetena, and the dative plural nīetenum (Mercian nētna, nētnum).

These forms present something of a paradox. Given a starting point such as *nı⁀etīnu and *nı⁀etīnā,²¹ neither ordering of high-vowel deletion and medial vowel shortening will give us the right outcomes. If shortening occurred first, then we would expect the equivalents of the Mercian forms, following the developments laid out in table 4.5 above. That is, we would expect syncopated ^xnīetna rather than attested nīetena.

On the other hand, if vowel deletion occurred first, then we would expect the following developments:

The retained medial length would ‘protect’ that syllable from being syncopated in the genitive plural, correctly yielding nīetena – but this is at the cost of predicting high-vowel deletion in the nominative plural. The actual form is nīetenu, not ^xnīeten.

The most influential way of resolving this paradox is based on Luick (1921: 290), who suggested that the original secondary stress of the long vowel was retained even after shortening, protecting the now-short medial vowel from syncope. This account has been fairly widely followed in the grammars (Sievers 1965: 135; Campbell 1983: 149; Hogg 2011: 230). Bermúdez-Otero (2005: 25–32) has more recently attempted to place the process on a more solid theoretical footing, while keeping the essential logic of the proposal.²²

This line of thought has not been universally accepted, however, and a particularly noteworthy alternative solution comes from Fulk (2010: 137).²³ He asserts that ‘West Saxon has generalised the disyllabic stem in these nouns’, i.e. that it once had forms such as *nı⁀etna, but has analogically extended the medial vowel to give nīetena. Fulk is on the right track, I think, but some adjustment is necessary to head off some reasonable objections to the proposal as stated.

The main problem with Fulk’s analogy is why regularisation would occur in different ways in nīeten and hēafud. In the former, Fulk sees paradigm regularisation as restoring the lost medial vowel (*nı⁀etna → nīetena), while in the latter regularisation is achieved by extending syncope to all open medial syllables (*hæ⁀uβudu → hēafdu). Fulk indeed articulates the morphological reasonableness of the latter type of analogy very well, and argues that it is also responsible for the creation of Early West Saxon forms such as īdlu ‘empty (fem.nom.sg.)’ ← *īdilu (Mercian has, of course, īdelu). Why, then, would the paradigm of nīeten not simply have regularised by creating the nominative plural ^xnīetnu?

This objection can be answered, I think, by placing the regularisation of nīeten much earlier than that of hēafud, with this early levelling motivated by the particularly large amount of variation that had arisen in the stem vowel of this and other words with an originally long medial vowel. If the ordering of changes really was, as I have been arguing, first medial shortening, then high-vowel deletion, and only later final shortening, then there would have been a stage, between the second and third of these changes, where pre-West Saxon nīeten had three different forms of the (historical) second syllable:

This triple variation was surely more unstable and prone to restructuring than the merely two-way variation found in words such as *hæ⁀uβud and *īdil with originally short medial syllables: these only alternated between presence and absence, without any variation between long and short vowels.

I suggest that at this stage of pre-West Saxon, words of the nīeten-type levelled out this variation, in general by extending a single vowel throughout the paradigm. The illustrative forms would then develop as follows:²⁴

I leave it open whether the generalised vowel was long or short: hence the noncommittal notation *ī̆. If long, it would have been shortened regularly in the nominative-accusative singular by the shortening of vowels in final syllables. Probably any remaining long medial vowels would also be (re)shortened at this point, as contrastive vowel length in unstressed syllables seems to have been eliminated at that point (alternatively, the short vowel could have spread analogically). If it was the long vowel that was generalised, this would have (re-)created *nı⁀etīnu at a stage after the operation of high-vowel deletion. If such a form existed, I would assume that by this time the ending was a stable part of this word’s inflection, and so rather than applying a second round of mechanical high-vowel deletion, the *-u was retained on a morphological basis. This kind of morphologisation is well paralleled, with synchronic exceptions even in early texts to high-vowel deletion in contexts where it should be able to apply very transparently. For instance, the 8th-century Corpus Glossary (entry 514) gives the verb frigno ‘I find out by enquiry, consulo’, historically from *friɣnu where the vowel was in the most classic context imaginable for high-vowel deletion, presumably restored on the basis of both light verbs such as cumu ‘I come, uenio’ and heavy ja-stem verbs such as do͞emu < *doē miu²⁵ ‘I (will) judge, iudicabo’.²⁶

Once a paradigm with a uniform stem vowel – *nı⁀etī̆n, *nı⁀etī̆nu, and *nı⁀etī̆nā – had developed, the attested forms would develop by straightforward sound changes: the shortening of any remaining unstressed long vowels, and the lowering of unstressed *i to e. It is important to stress that this scenario does not work under a different relative chronology. If high-vowel deletion took place before shortening, then the *-u of the nominative-accusative plural would already be gone, with no reasonable basis for being restored later on.

In the final view, the West Saxon forms are simply not very probative. While I believe an analogical account along the lines proposed here can account for the paradigm in the most economical way, the traditional account based on Luick is also feasible, and is compatible not only with Luick’s own view of high-vowel deletion (that it affected the position immediately following a heavy foot), but also with the current proposal (as I argued in Goering 2016a: 186–187). Ringe’s theory of double loss, outlined briefly above, derives the West Saxon paradigm entirely by regular sound change – but any advantages this may seem to provide are entirely undermined by the inability of this framework to account for the Mercian forms, which have no feasible phonological or analogical explanation under that theory, and indeed ought to be precisely the same as the West Saxon ones. The overall conclusion is that the West Saxon forms are relatively easy to account for (and are synchronically transparent), while the Mercian paradigm (with its synchronically erratic distribution of syncope) demands a much more constrained and specific explanation, and should be privileged as important evidence in favour of the model of high-vowel deletion proposed here.

Most further data with a potential bearing on high-vowel deletion’s original operation, and its relationship to other sound changes, comes from morphologically volatile environments, where analogies are plausible or, in come cases, certain. There are isolated early forms, such as hirnitu ‘hornet’ (Erfurt Glossary 275; as hurnitu in Corpus 603) and aelbitu ‘swan’ (Épinal 718, Corpus 1439), which support the same pattern of no deletion in HLL sequences suggested by hēafudu and nētenu, and which are unlikely to have been morphologically restored from forms with vowel loss (Goering 2016a: 177).

Forms can also be found that seem to speak against the theory of high-vowel loss I have been developing. The largest class of words that appear at first glance to be at odds with the retention of HLL sequences are feminine abstract nouns made with the synchronic suffix-þ(u), such as strengþ(u) ‘strength’ and frymð(o) ‘beginning’. In late prehistoric Old English, this suffix had the form *-iþu, and I would predict that the regular forms in attested Old English would be ^xstrengeþu and ^xfrymeþ, neither of which actually occurs. Forms such as strengþu might be seen as supporting a Luick-style high-vowel deletion, occurring immediately after heavy feet, while Ringe (2002) points to the strengþ-type as a regular outcome of his proposed double loss.

It is important to note, however, that under no theory is either frymð or frymþu possibly regular: an original *frumiþu should, under any model, develop in parallel to *werudu, and become ^xfrymeþ. There has clearly been a widespread restructuring of this class of abstracts, and the issue is not to seize upon this or that form as ‘regular’, but to best explain all the extant forms through reasonable changes.

Originally, the *-u would have been limited to the nominative singular, and all other case forms would have had a heavy ending. Here is the paradigm for strengþ immediately before and after high-vowel deletion:

In every single form except the nominative singular, the following ending was heavy, and the medial syllable should have been lost. That the now-anomalous *strængiþu should then have been remade to *strængþu is an entirely run-of-the-mill analogy. Some such generalisation of the suffix without the *i must have taken place, since the replacement of *frymiþ by frymð has to have been a morphological process.

The variable fate of final *-u then follows naturally. Once *strængþu had come into being, it would (at that early stage) have been relatively anomalous. The inflection of þ-abstracts would in the other case forms have been standard for heavy strong feminine nouns, such as hild ‘battle’ or lār ‘teaching’, which had lost their nominative singulars due to high-vowel deletion (< *hildi, *lāru). There were two potential paths open to deal with this anomaly, and both are taken in different varieties of Old English:

Option 1 was mainly elaborated in the Anglian dialects, where analogical forms such as frymþu can be found. The-u was even extended from the nominative into the oblique: dative singular ermðu ‘misery’ in the Mercian Vespasian Psalter is a representative example. Meanwhile, West Saxon texts favour forms such as strengþ, with what I argue is an analogically removed *-u, and retain the historically expected oblique forms such as strengþe. See further Hogg & Fulk (2011: 29, 120–121).

The traditional account of Luick or Dresher & Lahiri, under which the regular forms would be strengþu and ^xfrymeþ, can of course also explain all these forms in one way or another – for this subset of data, that approach is no worse, but it is also no better. Ringe’s double-loss approach, however, faces serious difficulties. Even though he can derive strengþ regularly, and frymð by the same type of analogy as everyone else, he has no easy way to bring the ending *-u back into the picture, it having been universally lost in this class. He proposes a rather elaborate account involving the īn-stem feminine abstracts, but this rests on pivot-forms that simply would not have existed in early Old English, before the merger of *-ǣ and *-ī; see Goering (2016a: 192–193) for details.

It should be clear from this section that the data of Old English is messy, and it is simply not possible to account for it all through purely phonological means (not that we should expect phonological change to be the only factor at work). In a case like this, it is easy for explanations to proliferate, and hard to decide between them. Once the data is collected (a philological project), the analysis turns more on morphological judgements than phonological ones: which cases are easiest to explain as remodelled morphologically, and which ones remain outstanding once reasonable morphological solutions are exhausted?

Here, there is a core of data – including the early inflection of hēafud, and the many Mercian nouns of the nēten-type – that shows a peculiarity which resists any analogical explanation: a failure to undergo high-vowel loss in HLL sequences, alternating synchronically within a single paradigm with medial syncope in historical HLH conditions. Since this cannot be the result of morphological innovation, it must have a phonological explanation, such as the model proposed here: the formation of bimoraic trochees and the deletion of unfooted high vowels.

Even if we accept the bimoraic foot in early Old English, questions remain about overall prosodic structure. There are two major issues. First, what is the status of overheavy syllables and initial LH sequences: are there any contexts under which feet with three moras (or more) are permissible? And second, what is the relationship between final feet and stress, and is there any kind of extrametricality or stress demotion?

The model of high-vowel deletion developed above implies that trimoraic feet are strongly avoided. If *hæ⁀uβudum could be footed as either *(hæ⁀u-βu)(-dum) or as *(hæ⁀u)(-βu-dum), then the medial vowel would not be open to deletion. While precisely bimoraic feet were clearly strongly preferred, there is evidence that trimoraic feet were tolerated, at least under some circumstances.

The first type of evidence for overheavy feet comes from initial syllables (or stressed monosyllables) which would seem to have more than two moras. Monosyllabic examples are legion: land ‘land’, bōc ‘book’, torht ‘bright’, lēoht ‘light’, frēond ‘friend’, and many more. These could only be made bimoraic by assuming (optional) final-consonant extrametricality: that is, freedom to disregard the weight from segments in the final coda of a word (Bermúdez-Otero 2005: 9–10). Some very limited form of extrametricality may well be plausible (§4.5.2), but it will not be able to account for non-final overheavy syllables (Bermúdez-Otero 2005: 15): e.g. inflected forms such as frēon-des (gen.sg), lēoh-te (dat.sg), or torht-ne (masc.acc.sg).

We could perhaps play around with the syllabification or moraic assignment rules, but these would generally be unsatisfactory approaches. Alternative syllabifications such as frēo-ndes, lēo-hte, and tor-htne would result in onsets that violate sonority sequencing and are highly unusual for Old English: nd-, ht-, and htn-are not valid word onsets in Old English, and are not very plausible word medially. It seems better to accept that the initial syllables of words such as torhtne and lēohte really are heavier than the ‘ideal’ norm, but that this was acceptable specifically and only in initial root syllables or feet. Such tolerance – which I will call the overheavy licence for initial feet – is a natural response in a language that privileges the initial syllable of a word in many ways: it bears the primary stress, it allows the greatest range of vowels, and, in verse, its onset provides the material for alliteration. Such syllables must be footed, and the preference for feet to contain precisely two moras of weight is secondary by comparison.

That single overheavy syllables need to sometimes be incorporated into ideally bimoraic feet is probably not surprising. A more complex problem is posed by ĹH sequences: how would words such as weruld ‘world’, and its inflected forms such as werulde, have been treated? One thing to stress immediately is that there is no stress shift in werulde comparable to what is found in modern German lebéndig ‘living’, where the heavy syllable has attracted the word accent (potentially leaving the initial syllable unfooted). The initial stress is shown both by the developments of the vowel (e.g. diphthongisation to wĕorold-), and by the metrical behaviour of the word, which alliterates on w-. This implies that it is part of a foot – but in what way? A strictly bimoraic initial foot is not a possible option, given that moras are tied to syllables and the moras within a syllable cannot be split between feet (Hayes 1995: 121–123). So either we have an initial ‘degenerate’, light foot (we) (-ruld), or else an overheavy, trimoraic foot (we-ruld), as argued for by Idsardi (1994: 525–526) and Sohn (1998: 4–8). Note that either way, high-vowel deletion would be expected to and did in fact apply to the nominative singular: weruld is from *weruldu.

A hypothetical foot structure of (we)(-ruld) would involve an initial light foot, with a single mora. There are typological parallels for languages that use a bimoraic trochee, and which tolerate such light, monomoraic feet in some contexts. For instance, Cahuilla, a Uto-Aztecan language of California, largely shows a bimoraic trochee and root-initial stress system relatively close to that of early Germanic, though with the notable difference that the only consonant to count as moraic is the glottal stop [ʔ] when it occurs in a syllable coda (Seiler 1977: 26–43; Hayes 1995: 117–118, 132–140). In Cahuilla, there is a clear distinction between words beginning with LL and LH. An LL word such as táxmuʔat ‘song’ has initial primary stress followed by an unstressed syllable.²⁷ The foot structure may be taken to be (tax-mu)-ʔat (remember that [x] is non-moraic in this language). By contrast, the word súkàʔti ‘deer’ has the same initial stress, but also a secondary stress on the medial syllable. This implies a different foot structure, along the lines of (su)(-kaʔ)-ti (remember that [ʔ] is moraic in codas, unlike [x]). That is, since the three moras of su-kaʔ-cannot all be accommodated into a single bimoraic foot, a light initial foot is formed and carries the primary stress, while the medial syllable then forms a second, fully optimal bimoraic foot, which translates into secondary medial stress:

A typological analogue is, however, not evidence, and a closer examination shows that Cahuilla provides more of a contrast than a parallel to Old English. If we apply a system with light, monomoraic feet to Old English, we would expect, alongside the nominative (we)(-ruld), a dative singular *(we)(-rul)(-dǣ). Such a form should (just like súkàʔti) have a secondary stress on the medial syllable: ^xwérùlde. This predicted stress, however, is not to be found. Phonologically, the second syllable develops entirely as a normal unstressed vowel, with u lowering to o in forms such as werold or wĕorold, and the vowel is, in the later Old English period, eventually lost entirely: wĕorld-, world-.

Metrical evidence also weighs against footings such as (we)(-ruld), and instead supports the accommodation of LH feet, at least under certain circumstances. As noted in §3.1.1, resolution is a central feature of Old English poetry, and its operation is robustly evidenced in the surviving corpus. As I will discuss in more detail in chapter 5, resolution is most general with LL sequences, and a word such as wine ‘friend’ (for earlier uuini) will resolve in any context. But LH sequences also frequently resolve:

(60) þurh hwæt his worulde ge·dāl

‘through what his separation from the world’ (Beowulf 3068a)

(61) worulde līfes

‘of the life of the world’ (Beowulf 2343a)

In these verses (which are representative), the underlined sequences are resolved, and the half-lines would be unmetrical without resolution. Taking (60) as having a medial secondary foot would imply a verse of the rhythm wwwSswwS, which is wholly unparalleled in Beowulfian metre. A half-stress instead of resolution in (61) would result in the illegitimate rhythm SswSw, which is not only unparalleled, but one of the most conspicuously avoided patterns in Beowulfian verse (§3.1.6).

Negative metrical evidence reinforces this picture. We can find type-E verses with the rhythm SswS where the secondary stress is filled by a heavy derivational syllable (underlined in the following examples), but only when the root syllable is already heavy or resolved (italicised):

(62) ēhtende wæs

‘was a persecutor’ (Beowulf 159b)

(63) æþelinga bĕarn

‘children of nobles’ (Beowulf 1408b)

If Old English really had a Cahuilla-type system, with a degenerate initial foot followed by a regular bimoraic foot, then there ought to also occur verses such as the following, but with LH instead of HH or LLH:

(64) ^xworolde brēac

‘enjoyed the world’ (cf. Beowulf 1062b)

(65) ^xcyninga bĕarn

‘children of kings’ (cf. Beowulf 1408b)

All this evidence taken together points in one direction: LH sequences could, at least when word-initial, resolve together, and form a single foot.

This tolerance needs no further theoretical machinery to explain it beyond what has already been proposed. The overheavy licence already shows that Old English was willing to allow overheavy feet of more than two moras, when these occurred word-initially. It appears this licence applied irrespective of whether the feet in question were monosyllabic or disyllabic. The principles at work are then just the following (see §5.5 for an emendment to principle 2):

These principles will allow all the words discussed so far to be footed appropriately, from *(hæ⁀u)(-βu-du) to *(hæ⁀u)-βu(-dǣ) to *(sci-pu) to *(lēoht) to *(torht)(-nǣ) to *(cy-ning) to *(we-rul)-du to *(æ-þæ)(-lin)(-gā).

There is one further outstanding issue of foot structure: the possibility of secondary stress on final feet. As should now be clear from the data in the previous section, in initial and medial syllables, the heads of feet are usually stressed. The head of an initial foot bears primary stress, while that of a medial foot carries secondary stress. This is well illustrated by ǽþelìnga (see 63), which (given in its early Old English form to avoid anachronism) would be footed and stressed as *(ǽ-þæ) (-lìn)(-gā). As far as the first two feet go, there are no surprises. Bimoraic trochees should be seen as, in the first instance, fundamentally quantitative (Hayes 1995: 271–272), grouping elements of weight (phonologised perceptual duration), which we formalise as moras, into regular groups. These units are in turn used in stress assignment, based on a separate set of principles. In Old English, stress appears to proceed, like footing, from left to right, and to be left-headed: the leftmost syllable of the prosodic word is therefore always both footed and given primary stress, while medial feet are assigned secondary stress. The length of non-compound words generally means there will usually not be more than one medial foot at most.

Final feet, however, are treated as ‘unstressed’. The final syllables of words such as (hæ⁀u)-βu(-dæs) must be footed in order to account for the syncope that affects this word, but the final foot shows no evidence of stress. The vowel reduces to e in later Old English, showing the typical merger of unstressed æ and i, and metrically such syllables are treated as entirely unstressed. This does not necessarily mean that final footed (or indeed unfooted) syllables have no relative acoustic prominence at all, and things such as final phonetic lengthening are typologically common (Minkova 2021: §4.3), but actual ‘secondary stress’ (however exactly this was realised phonetically) could only be given to the heads of medial feet, not generally to those of final ones.

This exemption from stress should probably be thought of either as a type of foot extrametricality (Hayes 1995: 77–78), or of defooting. Either way, the practical results are much the same: a final foot is formed, and is relevant for high-vowel deletion but ignored when it comes to stress assignment. Final feet also appear to be less salient metrically, a matter discussed at greater length in §§5.6–5.7 in the next chapter.

A complicating piece of data is that, to judge by the metre, there is one type final syllable that does carry secondary stress. Particularly in Beowulf,²⁸ the final syllable of æþeling is able to condition the non-application of resolution (see further §5.5.2), which suggests that it is a partly stressed syllable:

(66) æþeling manig

‘many a noble one’ (Beowulf 1112b)

The scansion here is SsSŠw (type ‘A2k’; see appendix D), with no resolution taking place in manig. The behaviour of-ling in conditioning this suspension of resolution suggests that ǽþelìng, like ǽþelìnga, has real secondary stress of some kind. This is not normal, and most words that have secondary stress medially in inflected forms lose that stress when the syllable is final: a typical example is ṓþèrne ‘(an)other, second (masc.acc.sg)’ versus ṓþer (masc.nom.sg), with no secondary stress.

The general rule appears to be that overheavy final syllables such as-ing can (if not absorbed into a larger LH foot) bear secondary stress,²⁹ while bimoraic final syllables such as-er cannot carry secondary stress at all. This behaviour might be formalised in various ways, but it is likely related to the fact that final overheavy syllables are prosodically problematic. They are overheavy, and so perceptually relatively prominent, but they cannot be straightforwardly incorporated into the ideal bimoraic trochee. One possibility would be that a limited form of final-consonant extrametricality gets applied to these syllables (Russom 2001: 62–63). This could not be a general rule at the earliest period, since final extrametricality in *hæ⁀uβudu〈m〉 would incorrectly lead to footing as *(hæ⁀u)(-βu-du〈m〉) and escape syncope, but it is possible that extrametricality was allowed in a limited fashion. Alternatively, and perhaps more simply, the phonetic heaviness of these syllables could have led to the exceptional extension of the trimoraic licence to them. Either way, with special allowances taken to let syllables such as-ing and-end be footed, it is hardly surprising that they would be then exempt from the usual final foot extrametricality or defooting.

Putting all the pieces from this chapter together, I would propose the following rules for the foot structure of early Old English. Chronologically, most of the evidence comes from high-vowel deletion, and so describes a system active at perhaps c. 600 AD, though it is also congruent with the metrical data of Beowulf (probably composed c. 650–750, though the poetic register may reflect a slightly more conservative phonology than the daily speech of the time). The cover term ‘early Old English’ thus, in this particular case, spans the late prehistoric and early historical period.

The relevant prosodic factors may be summarised as follows. The foot type is the bimoraic trochee, with feet being formed starting at the left edge of the word (left-to-right foot formation).³⁰ An overheavy licence allows the creation of feet with more than two moras, in order to allow the footing of syllables that cannot be left unfooted (usually under-or overweight initial syllables, but also potentially overheavy final syllables). Light, monomoraic feet are strongly dispreferred (though they may occur in the second elements of compounds; see §5.5.1), but stray unfooted syllables are allowed where not excluded by other factors (namely the requirement to foot initial syllables). The leftmost foot is assigned primary stress, and all remaining non-extrametrical feet take secondary stress; the final foot is usually extrametrical for the purposes of stress assignment.³¹

More schematically, the rules and principles are:

This foot structure persists through a number of sound changes affecting the quantitative structure of words, at least until (and in large part potentially after) change 3:

A few localised analogies also took place early on, including possibly the restoration of vowel length in class II participles between changes 1 and 2 (e.g. *wundɔdu → *wundɔ̄du); see §4.4.1.1. More limitedly in terms of dialectal scope, sometime after change 2 pre-West Saxon extended a single stem vowel in noun paradigms where there was three-way alternation between a long vowel, a short vowel, and no vowel (through syncope) within a single paradigm (e.g. *nı⁀etnā → *nı⁀etī̆nā); see §4.4.1.2.

The combined foot structure and vowel deletion rules generate the data for early Old English very robustly, as shown in the following list:

A certain number of very early exceptions, such as the creation of forms like strengþu ‘strength’ and frigno ‘I learn by enquiry’, are due to natural and simple morphological pressures. Later, after change 3, the prosodic basis of high-vowel deletion became opaque. The relatively early Mercian of the Vespasian Psalter often preserves the original alternations, but analogical readjustments would become increasingly common as time went on. These changes did not obscure the fundamental distinction between words such as scip on the one hand, and word and werud on the other, but they do suggest that the presence or absence of high vowels was increasingly morphologised. It is not until the 12th and 13th centuries, in the period conventionally known as early ‘Middle English’, that new and more direct evidence for foot structure will emerge – this will be the matter of chapter 6. But before moving forward in time, I will deal with the metrical evidence of resolution and Kaluza’s law in Beowulf in the coming chapter.