1.1 The current chronology of pre-Old English sound changes and its critics

There is a widely accepted chronology of sound laws in Old English philology, covering the transition from Proto-West Germanic to Old English, starting with Luick (Reference Luick1921-40) and later modified by scholars such as Campbell, Brunner, Fulk, Stiles, Hogg and Ringe. This comprises sound laws such as fronting, breaking, i-mutation and Anglian Smoothing. I will refer to this theory as the Standard Theory. Extensive overviews and discussions are given by Campbell (Reference Campbell1977: 50–112), Hogg (Reference Hogg2011 [1992]: 74–168) and Ringe & Taylor (Reference Ringe and Taylor2014: 167–336), to mention a few.

Adjustments of the standard chronology have been put forward on various occasions, such as by Ball & Stiles (Reference Ball and Stiles1983), who altered the relative chronology of smoothing and back mutation, and Toon (Reference Toon, Koopman, van der Leek, Fischer and Eaton1987; referring to criticism by Kuhn from Reference Kuhn1939) and Toon (Reference Toon and Hogg1992), claiming that Anglian Smoothing was still active in the oldest attested texts and hence not ‘pre-historical’. Both refinements did not affect the foundations of the theory.

The crucial features of the Standard Theory are a postulated general fronting of PGmc *a, except before *w and nasals, which was followed by breaking of the *æ > ea before *rC, *lC and *x. Furthermore, the fronted, unbroken *æ could be restored to *a under influence of back vowels in the following syllable. Still, Anglian dialects in particular show much fewer instances of broken vowels than predicted by these three steps. This is supposedly the result of i-mutation, turning a broken *ea into e, and Anglian Smoothing (before ‑(l/r)k/g/x), with a similar effect (æ). Other vowels, in particular short *i and *e, could also be broken; the details can be found in the handbooks. A synoptic overview, also marking the differences between the dialects, is provided in table 1. This shows the main sound laws as reconstructed by the Standard Theory up till the earliest attestations, which are older in Northumbrian and Mercian than in West Saxon. The earliest Northumbrian and Mercian texts will be the basis of my evaluation of the earliest history of Old English (see section 1.2).

Table 1.

Sound laws affecting PGmc *a for Early Old English per dialect according to the Standard Theory with references to the relevant sections in Hogg (Reference Hogg2011). Sound laws that cancel the effect of a previous one, are highlighted in grey.

Some authors have raised questions about the plausibility of some of the reconstructed sound laws, e.g. Howell & Somers (Reference Howell and Somers2008) with reference to Anglian Smoothing. They state that this process is ‘difficult for scholars to adequately explain in either theoretical or phonetic terms’ and that the breaking-smoothing sequence in *nēh > *nēoh > nēh ‘near’ is an ‘apparent paradox’ (p. 187). Still, the historical reality of smoothing is not questioned by the authors.

Krupatkin (Reference Krupatkin1970: 50) is bolder in his criticism when discussing the theory of restoration of fronted PGmc *a, describing the theory of fronting and restoration as an implausible ‘zigzag’ a > æ > a’. Doubts have also been raised by Kortlandt (Reference Kortlandt1999, Reference Kortlandt2008), who, referring to the various stages as given in Fulk (Reference Fulk, Bremmer, Johnston and Vries1998), formulates his objections against this chronology in the following way (Kortlandt Reference Kortlandt1999: 46):

The core of the Standard Theory, and – in my view – its most problematic aspect, is the claimed nearly unconditional fronting of PGmc *a, followed by widespread breaking or retraction, all of this taking place before the implementation of i-mutation.Footnote ¹ An earlier attempt at inverting the relative order of i‑mutation and breaking, focusing on West Saxon Old English, was presented by Collier (Reference Collier1987), but it has not been met with wide enthusiasm and acceptance. An interesting statement is found in Hogg (Reference Hogg2011: 92), unfortunately without any further references:

Dyvik (Reference Dyvik1978: 34–5), who revised older theories about breaking and back mutation in North Germanic, assumed a common basis for the tendency towards breaking in pre-stages of Old English and Old Norse, which he dated very early, to the times before the landnam of the Anglo-Saxons.Footnote ³ The latter aspect seems unlikely in the spirit of the conclusions from Kuhn (Reference Kuhn1955), and I will work from the assumption that few to none of the specific North Sea Germanic innovations should be dated before the Anglo-Saxon landnam (Nielsen Reference Nielsen2000: 271–6; Stiles Reference Stiles2013). Haugen (Reference Haugen2004: 56–60) places i-mutation before breaking in his chronology from Early Runic Scandinavian to Old Norse. Old Frisian has some limited application of breaking and back mutation phenomena as well, which are usually dated later than i-mutation (Bremmer Jr Reference Bremmer and Rolf2009: 32–6).

An entirely different approach to the origins of Old English phonology is offered by Schrijver (Reference Schrijver2014: 70). He suggests that Celtic substratum influence was the trigger of many of the Old English vocalic changes, causing a fairly abrupt restructuring of the vowel system by imposing Celtic pronunciation habits on West Germanic vowels, with the result looking like the outcome of a complex set of consecutive sound laws (cf. Nielsen Reference Nielsen, Askedal and Nielsen2015: 274 and Laker Reference Laker2019).

1.2 Trying an alternative approach

Considering the aforementioned criticism on the current chronology of Old English sound changes and the proposed chronologies for Old Norse and Old Frisian, one may consider whether an alternative chronology could be possible. Old English breaking would appear much later than hitherto assumed, namely after the i-mutation.

The most logical candidate for an alternative chronology is the one postulated for Old Frisian, which is the closest relative of Old English (Nielsen Reference Nielsen1985; Ringe & Taylor Reference Ringe and Taylor2014: 139). Its earliest reconstructed history is much ‘simpler’ than the Old English one. A presentation of the generally accepted views on its history can be found in Bremmer Jr (Reference Bremmer and Rolf2009: 21–41). Up till now, a common stage or common chronology for Old Frisian and Old English has been commonly denied in the traditional approach, for instance, by Campbell (Reference Campbell1977: 52, § 131) or Ringe & Taylor (Reference Ringe and Taylor2014: 171, 181 and elsewhere), despite the many similarities between the two languages. Stiles (Reference Stiles, Faltings, Walker and Wilts1995: 211) concludes: ‘it is not possible to construct the exclusive common relative chronology that is necessary in order to be able to establish a[n Anglo-Frisian] node on a family tree’. Archaeological evidence, however, supports the idea that the Germanic settlers of Britain and the post-Migration Frisians shared their origin and maintained close cultural contacts during the first three centuries after the landnam (e.g. Colleran Reference Colleran2016; Nicolay Reference Nicolay2005, Reference Nicolay, IJssenagger and Hines2017; Hines & Behr Reference Hines, Behr and Ludowici2019), suggesting a common demographic basis for the two Germanic dialects: the fifth-century Frisians and the Germanic settlers in Britain were the same people.

In this article, I will compare and assess the performance of two theories that aim to account for pre-Old English phonological developments: (i) the Standard Theory, for which I will rely on Hogg’s (Reference Hogg2011) presentation of the chronology,Footnote ⁴ and (ii) a chronology based on pre-Old Frisian, with the following core concepts: conditional fronting of PGmc *a, i-mutation, breaking and back mutation, where the relative chronology of the latter two is also stipulated by the Old Norse evidence. This latter theory will be referred to as the Anglo-Frisian hypothesis, and the actually reconstructed chronological series of events as the Anglo-Frisian chronology. I will develop these chronologies in more detail, when comparing their performance.

The performance of the theories will be tested on the oldest extensive Old English sources: the few pieces of Early Northumbrian, as well as the Épinal and Erfurt Glossaries, all from the late seventh or early eighth century (for details see section 2). It makes sense, in my view, to test a reconstruction of prehistoric developments in Old English on the oldest attestations. I will apply both sets of sound laws to the reconstructed Proto-Germanic forms with short *a in the attested lexical items from the aforementioned sources. Either set of sound laws will produce a prediction of an Early Old English form, which can be compared to the actually attested form. The analysis will test which of the two sets of sound laws performs better overall and in terms of specific aspects, such as the prediction of words with and without breaking.

This article concentrates on the pathways of development for PGmc *a in stressed syllables, for which there is a practical and a theoretical reason: the practical reason is the obvious length limitation on a journal article. The theoretical point is that it is in particular the development of PGmc *a that is most complicated in the current approach: 4 out of 8 sound laws presented in table 1 apply to *a only. A theory that works for *a stands a good chance of working for the other vowels as well. A couple of issues will therefore have to be relegated to subsequent studies, in particular:

• • The development of other vowels than PGmc *a in stressed syllables; relevant for the understanding of the breaking of PGmc *a may be the development of PGmc *au, since they emerge as OE ea and ēa respectively. An alternative hypothesis, assuming a transition from PGmc *au > pre-OE *ā > OE ēa would strengthen the interpretation of short *a in this paper but is not crucial;

• • The palatalisation of velars before front vowels (churn, chaff, yard); although closely related with the development of PGmc *a and relevant for some of the squestions concerning the relative chronology, this issue would simply require too much space. Both the timing and phonological conditioning for the palatalisation in Old English are problematic, including in current theories (see e.g. Stiles Reference Stiles, Kastovsky and Fisiak1988).Footnote ⁵

• • The implications for further developments in late-Old English and Middle English.

The sources under consideration are considered to be ‘Anglian’ and therefore my conclusion can stricto sensu only apply to the non-Saxon branch of Old English. However, the fundamental steps in the Standard Theory for Anglian are the same as for Saxon (West Saxon and Kentish), and with the addition of Anglian Smoothing even more complicated for Anglian than for Saxon. A revision of the ‘Anglian’ prehistoric sound laws may be an incentive to reconsider the ‘Saxon’ history as well, which will, however, stay beyond the scope of this article.

The article is organised in the following way. After a discussion of some methodological issues in section 2, section 3 provides an outline of an alternative theory of the prehistorical changes in pre-Old English. The predictive power of my alternative approach is compared to the performance of the Standard Theory (following Hogg Reference Hogg2011: 74–163) in section 4, based on the material from the two Mercian glossaries. Section 5 discusses various sound changes and the performance of the two theories in more detail, while section 6 considers the Early Northumbrian material. Section 7 brings together the argumentation in a conclusion.Footnote ⁶

3.1 Proto-Germanic *a

A pivotal role in the reconstruction of the earliest changes in pre-Old English is taken by PGmc *a. Any alternative reconstruction must find a reasonable solution for its historical developments, including the lexical distribution of breaking to ea. An alternative theory will limit the lexical spread of early fronting and breaking, in order to eliminate retraction and monophthongisation by i-mutation and Anglian Smoothing. The question is: what alternative pathway of developments can be reconstructed? This will be the topic of section 3.3.

The building blocks for a new chronology of changes in Old English, working from an Anglo-Frisian perspective, are the following:

1. 1. Early rounding before nasals. A notable feature of EpGl, in contrast to ErfGl, is the consequent rendering of PGmc *a before nasals as <a>;

2. 2. Some form of phonologically selective fronting, a priori blocked by back vowels and various consonant combinations;

3. 3. The subsequent i-mutation is supposed to cause raising and/or fronting, rather than monophthongisation;

4. 4. Breaking is considered a relatively late process, as it is in Frisian and North Germanic:Footnote ¹² in the earliest sources (traditionally labelled as non-West Saxon) breaking of PGmc *a is almost completely limited to the context before ‑rC: 24 out of 29 instances in the database with <ea> in Épinal and 11 out of 15 instances in Erfurt, where breaking is less widespread.

3.2 The lexical distribution of a, æ and ea

The theory involving widespread early breaking and the subsequent sound laws of retraction and smoothing grew from the following assumption, formulated by Campbell (Reference Campbell1977: 54, §139 fn.1):

The sister language, Old Frisian, has fronting of *a as well, but there it is generally accepted that fronting only took place in a more restricted set of contexts; see discussions in Siebs (Reference Siebs and Paul1901: 1177–89) and Bremmer Jr (Reference Bremmer and Rolf2009: 29–30). The early widespread fronting in Old English, according to the Standard Theory, suggests entirely different pathways for (nearly) identical Old Frisian and Old English forms. The result of this interpretation is that there are many more, unattested steps for English than for Frisian, with exactly the same results; compare the following examples:

A remarkable distributional fact is that breaking is (nearly) absent in a couple of instances in the glossaries:

1. 1. before ‑rC + i-mutation (n = 6), with the only exception feormat < *farmjan- (instead of e [Hogg Reference Hogg2011: 118]; see section 5.4). This finds a parallel in the consonantal conditioning theory of breaking in Old Norse (Dyvik Reference Dyvik1978: 11), with breaking of Proto-Norse e before ‑rC and ‑lC, but not when followed by an i-mutation factor.

2. 2. before ‑rC, with /r/ < PGmc *-z; the vowel is æ/e (n = 7; 3 instances with i-mutation); otherwise, EpGl has ea before ‑rC without i-mutation, with four exceptions (n = 28). The contrast is statistically significant with p < 0.01.Footnote ¹³ ErfGl has more <e,ae,ę,a> in the latter contexts, but ea is only found before r < PGmc *r (apart from a few other instances discussed in section 3.6).

This suggests that fronting and not breaking applied when there was a high (semi) vowel in the second syllable or when the /r/ was dental *[r] < PGmc *z. Breaking rather correlates with velar sounds, such as *[r] (see Ringe & Taylor Reference Ringe and Taylor2014: 189) or *[l], which are no obvious environments for fronting.

Broken vowels in dialects with extensive breaking often correlate with a in dialects with less breaking. Such an alternation is found before ‑lC: WS eald ‘old’ corresponds to ald in other dialects, but æld is rarely found. The broken vowel ea alternates with a in the only instance of breaking of *a in CH, and not with a fronted vowel: ‑gard ~ ‑geard and not *-gærd. According to the Standard Theory, the form gard should not exist at all: the sequence g_rd does not fulfil any of the exceptions to fronting in the Standard Theory, nor is it targeted by any of the later sound laws in any of its paradigm forms. One could therefore consider a chronological relationship between the two forms.Footnote ¹⁴ In the remaining seven Early Northumbrian (eNorth.) words with PGmc *a before ‑rC, the eNorth. texts have a (four times with preceding w‑), except for aerigfaerae (LR) ‘flight of arrows’ < *arxwō‑. In instances with the same coda consonants as in the eNorth. dataset (‑rd, ‑rð, ‑rn, ‑rƀ, ‑rp, ‑rx) and without i-mutation (n = 12 EpGl, 13 ErfGl), EpGl has 10 times ea, 1 <ae> and 1 <o> (suorn-). ErfGl has 4 x ea, 5 x a (4 times with preceding w‑) but 4 x æ/e. For the context ‑rx see section 3.4.Footnote ¹⁵

Finally, an indication for a chronological relation between a and ea without intervening early fronting to *æ comes from the earliest attested place names and personal names in Old English. Van der Schee (Reference Schee2015) presents the evidence for breaking of PGmc *a in names from Latin charters and manuscripts of Bede’s Historia from the seventh century (see esp. p. 234). The oldest attestations (679) spell <a> (bernhardi, gudhardi). Also, in the older Bede manuscripts (M, P, B), the spellings with <a> are dominant (p. 240), only gradually giving way to <ea>-spellings. There is hardly any trace of æ, and certainly not as an intermediate stage between a and ea. This supports the idea that ea developed directly from a. Under the Standard Theory, one has to assume either: (i) that speakers/scribes of a dialect that retained ‑arC gave way to speakers/scribes that developed ea from an earlier not attested *æ, or – as Van der Schee does – (ii) that <a> stands for an ‘underlying’ */ɛa/ and that the whole alternation is merely a spelling issue. A problem for the first interpretation is that various attested words (such as -hard- in the onomastic material and gard in Early Northumbrian) do not represent contexts for retraction in the Standard Theory, implying that they are not supposed to exist at all in any dialect in the seventh century. The second interpretation assumes influence from petrified Latin or Franconian spellings.

All things considered, it seems for the most part that the broken diphthong ea < PGmc *a alternates diachronically and dialectally with a and not with æ. A retention of the a in breaking contexts is also likely from a comparative Germanic perspective: the Old English ‘breaking environments’ have a velarising effect in, for example, Old Frisian and Old High German, as mentioned by Laker (Reference Laker2019: 603).Footnote ¹⁶ This implies that fronting in these contexts was not the most logical development and that *a may have formed the input for the breaking. The gard ~ geard-alternation in BD and the seventh-century names in Bede suggest that breaking only developed in the seventh century and gradually spread during the eighth century. The widespread al-spellings in early West Saxon and Kentish sources imply that this process – at least in particular phonological contexts – was still active in the late ninth century (Versloot Reference Versloot2021: 80–2).

3.3 The phonetic conditioning for the breaking of *a to ea

The crucial issue is: how does one get from /a/ to /ɛa/ in a phonologically plausible way if one does not posit an early widespread fronting of *a > *æ? The ‘prototypical’ context for breaking is the position before ‑rC and this context must be the prime focus when searching for a possible explanation for the change.

A solution may lie in lengthening of /a/ before ‑rC. Such a lengthening is also assumed for Old East Frisian (Siebs Reference Siebs and Paul1901: 1178), especially when the following consonant is a tautosyllabic d,m,n,v (< *w,f,ƀ). No lengthening takes place before ‑r + k,g,s,p. The contexts where EpGl and ErfGl both have ea, and which for that reason appear to constitute the core of breaking contexts are: ‑rd, ‑rđ, ‑rt, ‑rb, ‑rw, ‑rm and ‑rr. There is a large overlap between the contexts sensitive to breaking in Old English and the contexts sensitive to lengthening in Old East Frisian; contexts without lengthening in Old East Frisian (‑r + k,g,s; there is no *‑arp in the data) have no breaking in ErfGl (n = 5), but all of them have e,æ, one with i-mutation. The contexts ‑rk, ‑rg are labelled as smoothing contexts in the Standard Theory.Footnote ¹⁷

This hypothesis, which will be referred to as the ‘breaking-through-lengthening’ hypothesis, is also supported by Old English itself. It is generally assumed – although never shown in the spelling – that vowels could be lengthened before ‑ld, ‑r + l,n,d,ð,[z], ‑mb, ‑n + d,g (Hogg Reference Hogg2011: 208). The contexts with a nasal do not apply here, because PGmc *a was rounded in that context. According to Brunner & Johnston (Reference Brunner and Johnston1970: 9), this lengthening had already taken place in the eighth century, which brings it close to the date of the breaking as suggested in this article. The context with ‑ld constitutes a major dialectal contrast between Saxon and Anglian in terms of breaking. The context is one of the most robust contexts for lengthening. An explanation for the lack of breaking in Anglian before ‑lC will be presented below. The remaining phonological breaking contexts are very similar to the Frisian lengthening contexts. The common denominator seems to be r + voiced, non-velar consonant. The details may differ per language variety and lemma; cf. PDE wind [i] (noun) – wind [ai] (verb). Unfortunately, not every context is equally well represented in every dataset.

The aforementioned lengthening is not controversial, but the transition of lengthened /a/ to [ɛa] is crucial to the whole debate. It was not without reason that scholars like Luick and Campbell assumed that fronting had to be a widespread phenomenon in pre-Old English to find a phonetic rationale for the [ɛa]-quality.

Faroese provides an interesting potential parallel here. Old Norse (ON) had a long /a:/, which was ‘dark’ and appears as [ɔa] (with breaking!) or [ɔ] in Modern Faroese. But next to this old long /a:/, Faroese developed a new low long vowel through lengthening of ON short /a/ in light syllables. The interesting parallel between Faroese and Old English is that the lengthened ON a is always realised as [ɛa], but without lengthening it remains [a] (Küspert Reference Küspert1988:207, Þráinsson Reference Þráinsson2004: 34,395). Examples can be found within one paradigm, e.g. tak [t^hɛak] ‘roof (nom.sg.) ~ taks [t^haks] (gen.sg.); spakur [spɛakʊr] ‘calm (masc.nom.sg.)’ ~ spakt [spa^hkt] (ntr.nom.sg.). This means that only the lengthened vowel was broken to [ɛa] and that general fronting of the vowel in its short phase was not a prerequisite for the breaking. A similar length-conditioned breaking of PGmc *a can be found in West Frisian, resulting in [iˑə] from Middle Frisian ā (from OFri. short a with lengthening, but also from OFri. ā); e.g. Old Frisian nama ‘name’ kept a short vowel, appearing as Modern West Frisan namme [namə], but in the verb namia ‘to name’, the vowel was lengthened and subsequently fronted and diphthongised (‘broken’ in the Old English terminology) to neame [niˑəmə].Footnote ¹⁸ The parallelism between the Old English ea and the West Frisian transition ā > ea has also been observed by Laker (Reference Laker2007: 179).

The phonetic rationale for breaking suggests that the lengthened vowel was indeed most likely a phonetically somewhat fronted *[a̟ː] or *[æː]. Present-day Dutch has a similar phonetic contrast between /a/ = [ɑ] and /a:/ = [a̟ˑ]. The timbre of the vowel can offer an explanation for the lack of breaking in Anglian before ‑lC. If we assume that the Anglian /a/ was darker than in West Saxon (perhaps because there was a difference between Anglian dark [l] and [l] in West Saxon), this would explain that Anglian had *[ɑˑ] against *[a̟ˑ] in West Saxon. The West Frisian language offers an exact parallel to such a contrast. Fifteenth-century charters from the northeast spell the Old Frisian word ald ‘old’ as <auld>, later <aud>. This comes out as âld [ɔːt] in the modern language (aud [aˑut] in the northeastern dialect of Schiermonnikoog). Charters from the southwest spell <aeld>. The southwestern dialect nearly disappeared during the sixteenth and seventeenth centuries and only survives in the modern dialect of Hindeloopen, where the positive degree is aald with [aː], but the comparative and superlative show breaking to [ɛ_ɔ]: ealder, ealdst. Footnote ¹⁹

Given the following observations:

• – OE ea through breaking shows a distributional overlap with a in varieties without breaking, both diatopically and diachronically;

• – Broken vowels are nearly absent in contexts where fronted vowels are expected, such as before i,j in the next syllable or an r [r] from PGmc *z;

• – Faroese and West Frisian provide examples of a short a that was only broken/diphthongised after lengthening;Footnote ²⁰

• – The core-context for breaking, before ‑rC^{+voice,-velar}, overlaps strongly with acknowledged contexts for lengthening in Old English, with parallels in Old Frisian;

I, therefore, hypothesise that breaking took place via lengthened pre-Old English *a and not a short fronted PGmc *a. This lengthened, phonetically somewhat fronted /a/ was diphthongised to ea. The ‘lengthening’ proposed above was not necessarily phonemic (possibly *[a̟ˑ]), but still enough to allow for some form of glide to develop into a phonetic diphthong. The phonemic status of <ea> is also open to various interpretations (for a longer discussion, see Hogg Reference Hogg2011: 16–20). The general interpretation is that the broken vowels still counted as ‘short’, as evidenced by their use in verse and their development in Middle English (Brunner & Johnston Reference Brunner and Johnston1970: 10). Minkova (Reference Minkova, Honeybone and Salmons2015: 78) concludes that the spelling <ea> for broken short /a/ represents a phonetic reality [e͝a] which was a ‘not-yet-integrated semi-contrast’, in an allophonic relation with [æ]. Whether forms such as PDE arm, hard, OE earm, heard are the result of some form of a Middle English phonological monophthongisation process (i.e. the result of the merging of the two semi-phonemes /e͝a/ and /æ/ with /a/; see Lass & Laing Reference Lass, Laing, Denison, Bermúdez-Otero, McCully and Moore2012: 76), or come from dialects with little or no breaking (cf. Brunner & Johnston Reference Brunner and Johnston1970: 12–13) can be left open here. It is not the primary question for a reconstruction of prehistoric sound changes in Old English.Footnote ²¹

The distribution in the earliest attestations suggests that breaking was a fairly young process that may have started only in the seventh century and was still spreading in the eighth century. Such an approach avoids the early fronting, the early and widespread breaking and with it Anglian Smoothing. It stays in line with all the evidence of unbroken spellings in early sources, does away with unattested intermediate broken forms and eliminates the necessity to invoke processes like dialect mixture and scribal confusion to ‘account’ for many actually attested forms.

3.4 Breaking before *x and Anglian Smoothing

The PGmc velar fricative *x is in the Standard Theory mentioned as a source of breaking. The context before *(C)x is also a trigger of so-called Anglian Smoothing, and forms with attested breaking are only expected in words where *x disappeared before the vowel could be monophthongised by *x and when it was not monophthongised by i-mutation, which leaves very few lemmas. *x is consistently associated with fronting in the glossaries (EpGl): naecht- ‘night’ < *naxt‑, aec(h)that ‘deliberates’ < *axtōþ, fex ‘kelp’ < *faxsa-; and with i-mutation: ambechtae ‘servant, office’ < *ambaxti-. *x is only preserved in the combination *xt and *xs, but disappeared in other contexts, e.g. steeli ‘steel’ < PGmc *staxlja-, or eo‑ (attested in eorisc ‘water-rush’) < PGmc *axwō-. The monophthong in steeli can be ascribed to i-mutation, but eo- is considered to be a key-witness for the Standard Theory: ‘*aho > *æho > (breaking) *eaho > (smoothing) *æho > ǣo’ (Hogg Reference Hogg2011: 79). Breaking and subsequent smoothing are required in the Standard Theory to avoid retraction: *axu > *æxu > *axu, eventually leading to **ā. This form ā, without fronting, is the Old Frisian form (aha as a name in an even older Latin context), just as e.g. OFri. nacht ‘night’; apparently *x had the sound value *[x] in Frisian and blocked fronting.

A much simpler scheme for the Old English glossary data is obtained if we assume that *x was realised as *[ç] in Anglian (pre-)Old English and allowed fronting to operate. The phoneme /x/ may have had different realisations in pre-Old English and Old Frisian.Footnote ²² An Anglian *[ç] also solves the issue of eo‑, which can now directly be derived from earlier *ax(w)ō with fronting before *[ç] and subsequent contraction after loss of the *x(w): *ax(w)ō- > *æxo > *æo > eo. Footnote ²³

Let us now have a look at the evidence in the glossaries for the effects of i-mutation and Anglian Smoothing of PGmc *a in words that are supposed to have breaking in the Standard Theory. The chronology of the processes can be summarised in the following way:

The actual results in EpGl are shown in table 2.

Table 2.

Breaking and smoothing according to the Standard Theory (ST). The theory predicts the tokens to fall in the grey cells (77 per cent correct for non-i-mutation contexts).

The Standard Theory’s prediction about ‘monophthongisation-through-i-mutation’ is borne out well. But the complications of fronting and breaking are not needed: simple i-mutation of *a – or *æ when fronted – will lead to æ or e in the glossaries (see further section 3.5).

The smoothing is not fully effective (7 out of 11) and the actual distribution in EpGl can be predicted in a more efficient way by the assumption that /x/ was *[ç] and that in the contexts of *ax- and *arx, the vowel was fronted and not further affected by any change. General breaking of non-fronted *a before *rC suffices to predict the instances of mear(c) ‘mark’ and uuearg-rod ‘gallows’. The improved result of such a set of sound laws is shown in table 3.

Table 3.

Predicted Anglo-Frisian (A-F) developments in the ST-breaking and -smoothing contexts from table 2 (91 per cent correct for non-i-mutation contexts)

The single exception with a broken vowel instead of a fronted vowel is leax (EpGl), which appears as lex in ErfGl.; for a discussion see section 5.4.

We can therefore conclude that the actual EpGl attestations in contexts with *x and other ST-smoothing contexts do not support the complex set of sound laws from the Standard Theory (fronting, breaking and subsequent monophthongisation by i-mutation and smoothing), and are better predicted by the application of i-mutation to unbroken forms, with a subsequent later, restricted application of breaking.Footnote ²⁴

3.5 The impact of i-mutation

Both EpGl and ErfGl show variation between <ae> and <e> in the i-mutation product of PGmc *a. Hogg (Reference Hogg2011: 118, 123–4) mentions the relevance of the vowel quality: pre-OE *æ is mutated to <e>, whereas *a and *ɑ become <æ>, although the latter soon became <e> as well (Ringe & Taylor Reference Ringe and Taylor2014: 228–9), in particular in light syllable stems (Hogg Reference Hogg2011: 128). The <ae>-spellings are more prominent in EpGl (43 per cent) than in ErfGl (28 per cent), which is in line with the dates of the manuscripts. Hogg’s criteria explain c. 80 per cent of the distribution of the spellings in EpGl. Detailed consideration of various factors shows that the quality of the historical i-mutation trigger, /i/, /i:/ or /j/, is an additional factor, where /i/ has a weaker effect than /i:/ and /j/. Such a difference in effect has hitherto not been observed for i-mutation in West Germanic languages, but see Schulte (Reference Schulte2018: 57–8, and especially Reference Schulte1998: 158–73) for a lengthy discussion of this phenomenon and parallels in North Germanic and other non-Germanic languages. Vaan (Reference Vaan2017: 301–5) highlights the relevance of the trigger /j/ in Dutch, whereas Dutch in general shows very few traces of i-mutation.

There are differences between the two glossaries in respect to the i-mutation factor. The input vowel quality is the most important factor in EpGl. This results in a probabilistic cline, as demonstrated in table 4.

Table 4.

The impact of the input vowel quality and i-mutation factor on the final outcome of i-mutation (only unambiguous ‑i and ‑ī/j; see the discussion below)

In ErfGL, the quality of the mutating factor is more important, as can be seen from the cline in table 5.

The features relevant for the final outcome of the i-mutation process are thus:

• • The quality of the mutated vowel: at the moment that i-mutation was active, PGmc *a had developed into [æ], [a] or [ɑ] (before nasals). The distinction between [a] or [ɑ] turns out to be irrelevant for the further development, and it is the contrast with /æ/ that matters.

• • There are three historical i-mutation factors: /i/ has a weaker impact than /i:/ and /j/.

• • In various paradigms, there is alternation of the mutating factor, such as in class 1 weak verbs, e.g. inf. *framjan – part.pt. *framid, or in the i-stems, e.g. nom.acc.sg. *mari, gen.dat.sg. *marī. We can expect analogical levelling here. Scrutiny of the data shows an effect of root syllable weight with heavy root syllables; these words ‘behave’ as if they had /i/ as a mutating factor, while after light syllables, the inclination to raising is stronger. This could be ascribed to Sievers’ Law, which predicts *‑ij- instead of *‑j after heavy syllables. However, a similar but weaker syllable weight effect can also be observed for words with an invariable mutation factor, so syllable weight should be added as an additional relevant factor that predicts the height of the mutation product.

Table 5.

The impact of the input vowel quality and i-mutation factor on the final outcome of i-mutation (only unambiguous ‑i and – ī/j; see the discussion below)

The detection of these factors explains a lot of the alternations between <e> and <ae> in the glossaries, even when they cannot easily be converted into discrete sound laws. A statistical model that factors in manuscript, mutation factor, syllable weight and input vowel quality can predict 87 per cent of all tokens in both manuscripts correctly. The details can be found in the Appendix.Footnote ²⁵

3.6 Palatal diphthongisation or back mutation?

One of the supposedly older developments in Old English that seems to have left some traces already in the glossaries is what Hogg calls ‘palatal diphthongization’. The process is ‘controversial or dubious’ in many instances (Hogg Reference Hogg2011: 104ff.) and is predominantly found in West Saxon and Northumbrian rather than in Mercian. I will concentrate on positive evidence for this process in the glossaries and the Early Northumbrian material. Where a digraph can be interpreted as the result of breaking, it is not considered to be evidence for palatal diphthongisation, such as in ‑geardas ‘gardens’, with breaking before ‑rd. I considered instances with the initial consonants *k-, *g-, *sk- (no instances with *ja- in the glossaries), followed by PGmc *a with fronting or i-mutation.

Both glossaries have the same two lemmas (out of 16 relevant tokens) with ea < *a that cannot be explained as breaking before rC: sceadu ‘shady place’, ‑sceadan/‑ae ‘damage’. EpGl also has once scaedu‑ alongside ErfGl sceadu‑. The only somewhat unambiguous context in the glossaries is therefore *skæ- < PGmc *ska-, but the glossaries have 6 (EpGl) and 4 (ErfGl) examples without <ea>, such as scaet ‘property’. These instances of sceadu and ‑sceadan/‑ae can therefore better be explained as ‘back mutation’ (Hogg Reference Hogg2011: 149ff.); see also the discussion in section 5.4. The Early Northumbrian material has two instances of ea in four potential candidates for palatal diphthongisation: sceppend (CH) ‘creator’, cæn[d]æ (LR) ‘knew’, sceal (LR) ‘shall’, geatum (LR) ‘ornaments’. The last one can also be an instance of back mutation. The data support the idea that palatal diphthongisation was not common in Mercian and is only found at an initial stage in Early Northumbrian.

3.7 *ax(w)‑ and *aw‑

Seven lemmas with the sequence *aw‑ and the word eo‑ ‘water’ (see section 3.4) will be kept out of the count in section 4 because of their unpredictable inclination to build diphthongs; compare (snid)streo ‘carline thistle’, str[e?]idae ‘spread out (3.sg.prt)’, thrauu ‘rebuke, threat’ < *strawa-, *strawid- and *þrawō-. The various developments are strongly associated with the quality and quantity of the following syllable and must originally have alternated paradigmatically. The direction of the levelling (which is difficult to establish for the language of the glossaries because of the limited number of tokens per lemma) depended probably on the frequency and specific phonology of the various paradigm forms. These forms can therefore not provide evidence about the mainstream developments.

4.1 Sound laws for PGmc *a for early Mercian according to the Standard Theory

The sound laws for PGmc *a deducible from Hogg (Reference Hogg2011), §§ 5.3–5.100, in particular for the early Mercian sources, are summarised as follows:

Table 6.

Predictive power for the development of PGmc *a in the EpGl, following the two sets of reconstructed sound laws. Only seven words came out with a different vowel (ā, eo, ē, o, ō) and were excluded from the overview. X-axis = predicted vowel quality; Y-axis = attested vowel quality. The dark grey cells mark the correctly predicted tokens. The light grey fields mark the confusion of e and æ in the prediction.

Ringe & Taylor (Reference Ringe and Taylor2014: 221) mention ‘Second fronting’ for EpGl, but Hogg (Reference Hogg2011: 137) says that this is no more than a marginal development in the glossaries and this was therefore not implemented as a sound law for the language of the glossaries.

4.2 The Anglo-Frisian sound laws

4.2.2 Sound laws for PGmc *a for Erfurt

Chronology of three steps:

1. 1. split of PGmc*a: a → [ɑ] = <a/o> / _N

   *a: a → æ except / w_r / _lC / _Cu/ō /_rC_non-pal (C_pal = x = [ç], Ci,j, [r] < z)

2. 2. i-mutation ‘cline’:

   1. a. a [a,ɑ] → æ (e) / _C(C)i and heavy syllable ja-stems, i-stems and class 1 weak verbs

   2. b. else: a, æ → e

3. 3. breaking: (C^-w )arC → *[aˑ] → ea.

Similar analysis steps as for EpGl can be applied to ErfGl. The account given by Hogg hardly differentiates between the two glossaries. ErfGl has a development to either <o> or <a> in the position before a nasal (also mentioned by Hogg). Both outcomes were considered to be correctly predicted in both approaches. ErfGl also has a tendency to preserve the *a in the sequence *warC.Footnote ²⁸ It also has a much stronger inclination to develop every mutated PGmc *a to <e> (Hogg Reference Hogg2011: 123) and, therefore, the Standard Theory predicts an e as the result of i-mutation for ErfGl (with good performance). Table 7 summarises the results from the sound laws and their actual match with the attested word forms in the glossary.

Table 7.

Predictive power for the development of PGmc *a in the ErfGl, following the two sets of reconstructed sound laws. Only five words came out with a different vowel (ā, æo, ē, ō) and were excluded from the overview. (See further table 6 for legend.)

The interpretation of the Anglo-Frisian hypothesis is based on the (fine-grained) categorical application of the i-mutation, which leads to a prediction accuracy of 70 per cent with i-mutation. The purely probabilistic cline provides an accurate prediction for words with i-mutation in 91 per cent of the words. This corresponds to 21 additional correct items: 218/274 = 80 per cent correct.

Altogether, the phonological shape of ErfGl is more difficult to predict in both approaches. As in the EpGl, the largest deviations are found in the interchanging of e and æ in either prediction or attestation. Part of it has to do with the variability of the effect of the i-mutation. Further implications of the differences between the two approaches will be discussed in section 5.

4.3 An alternative account: early fronting after all

Comparison between the initial fronting conditions in Frisian and English leads to an alternative interpretation for the context *arC. Frisian has fronted vowels in these contexts, except when preceded by *w, *b and *f, or when followed by a back vowel. This leaves only few exceptions, such as bern ‘child’ and garda ‘yard’.Footnote ²⁹ When we assume a similar early fronting in this specific phonological context (not throughout as in the Standard Theory), the input vowel to breaking is */æ/ in this approach, without the intermediate lengthening of *a. This interpretation will be referred to as ‘early-fronting’ hypothesis.

For EpGl, where breaking is almost entirely restricted to the context *arC, this scenario requires a general fronting to *æ before any *rC. The i-mutation causes raising of this *æ to e (4 out of 6); the remaining two instances are feormat and auuaerdid, the latter with æ after w- (see section 5.4). Breaking is predicted in the 28 instances without i-mutation with assumed *æ (under this approach), and indeed observed in 24 tokens. One of the exceptions is suornodun with PGmc *a between *w_r, which represents the context described as ‘combinative breaking’ by Hogg (Reference Hogg2011: 89). Altogether, 29 out of 34 relevant tokens (85 per cent) in the Épinal Glossary with PGmc *arC are correctly predicted by this set of rules. For the breaking-through-lengthening theory, this would be 30 (88 per cent), so both theories perform almost equally well for EpGl.

The alternative theory is particularly fruitful for the interpretation of Erfurt, when judged by conditions for fronting more similar to Frisian, namely not in the context *war‑. Breaking is then restricted to the contexts before ‑r + voiced, non-velar consonants, but not before ‑rn and ‑rw. The product of i-mutation is expected to be <e> in the etymons under consideration (rule 2b in section 4.2.2). This produces 22 correct predictions (out of 34 = 65 per cent) and 27 (79 per cent) if we are tolerant to alternations of <e> and <ae>. In the earlier presented version of the A-F theory, this was 19 (56 per cent; and 21 with <e> ≈ <ae>). Given variants such as uearte, uaertae and uuertae for ‘wart’ in ErfGl, there is no way to achieve 100 per cent correctness in any interpretation.

This theory has two advantages when compared to the ‘breaking-through-lengthening’ hypothesis, described in section 3.3:

1. 1. It is more straightforward in terms of the vowel quality for breaking;

2. 2. It increases the overlap with developments of fronting in Frisian, which is likely, because fronting is a very early change and therefore expected to be similar in Frisian and English and it strengthens the similarity between consonantal breaking in Old Norse (on /ɛ/) and Old English (on /æ/).

Its main disadvantage is that it leaves the observed diachronic and diatopic alternation of a and ea in the near absence of æ (section 3.2) unexplained. However, in the glossaries, there is a side-by-side occurrence of broken vowels and front vowels: in the 12 tokens where <ea> appears only in EpGl and not in ErfGl, the latter has 9 fronted vowels before ‑r + g,k,n,t: three times <e> and five times <ae>. Three of them have <a>, all with preceding labial w or b.

An aspect mentioned in this section, which tends to support the ‘breaking-through-lengthening’ hypothesis as described in section 3.3, is the fact that the instances of breaking in ErfGl are concentrated in contexts ‑r + r,ƀ,d,đ,m, which constitute a good overlap with the lengthening contexts, in particular in Old Frisian. Among the 9 tokens with a voiceless consonantal cluster, there is only one instance of breaking, strangely enough after w-: uearte ‘wart’ (next to uaertae and uuertae). The fact that the breaking in ErfGl is concentrated in the voiced contexts suggests that lengthening of the vowel may have been part of the process of breaking.

This allows for yet another interpretation of the observed correspondence between EpGl ea and in particular ErfGl æ: the <ae> may be interpreted as *[æˑ], the intermediate stage for breaking as described in section 3.3. Vowel length is after all not rendered in Old English spelling, and <ae> may represent *[æ], *[æˑ] or *[æː]. Breaking is more widespread in EpGl than in ErfGl, and the dialectal origin of EpGl should be sought nearer to the West Saxon region. The language of ErfGl resembles rather the early Northumbrian dialect (e.g. with respect to the impact of preceding w‑). If indeed breaking originates in West Saxon and gradually spread to the north and northeast, the instances with <ae> in ErfGl could reflect the middle step in the chain /a/ > *[æˑ] > [ɛˑa]. Finally, there is so-called ‘Second Fronting’ that might account for some of the instances of <ae> in ErfGl (see section 5.2), implying an earlier *a.

The conclusion is that both scenarios, the ‘breaking-through-lengthening’ hypothesis and the ‘early fronting’ hypothesis, have their pros and cons. They both perform well on the data from the glossaries, although the ‘early fronting’ hypothesis is slightly more flexible in accommodating the variation in the language of ErfGl. Under any circumstance (including the Standard Theory), we have to assume some variability in the application of sound changes. The ‘early fronting’ hypothesis also stays closer to Frisian fronting and Old Norse consonantal breaking, but leaves other aspects of breaking unexplained, such as the preference for voiced contexts in ErfGl.

However, it is important to notice that both variants of the Anglo-Frisian theory differ from the Standard Theory in a couple of respects:

1. 1. they limit the extent of fronting and operate without the fronting-retraction movement;

2. 2. they apply i-mutation to a stage of the language with unbroken vowels and do not cause the uncommon effect of i-mutation, entertained in the Standard Theory, namely monophthongisation;

3. 3. they operate with an implementation of breaking after i-mutation in fairly specific contexts, without a later correction by i-mutation and a process known in as Anglian Smoothing.

Both scenarios will be considered when discussing changes in words with *arC in section 5. They both mark a clearly different approach than the chronology of the ST.

5.1 General differences between the theories

The figures in section 4.2 show a very similar level of correct predictions for both tested theories. Tables 8 and 9 illustrate that both theories also make very similar predictions on the level of individual tokens: for 92 per cent of the tokens in the Épinal Glossary, the theories predict the same outcome, which is correct in 86 per cent of those cases. The figures for ErfGl are 91 and 74 per cent respectively.

Table 8.

Overlap and difference in the performance of the two theories for EpGl

Table 9.

Overlap and difference in the performance of the two theories for ErfGl

This implies that both theories are successful and fail in largely the same stock of words. The following sections discuss the predictive success of several hypothesised sound laws in either theory. The situation in the EpGl is taken as default in the discussion – derivations in ErfGl will sometimes be discussed separately. In general, both theories have more problems with ErfGl.

5.2 Fronting, ‘second fronting’ and retraction

Both theories are successful in the prediction of developments of words with *aN or *alC without i-mutation in 97 per cent of cases (n = 70). One exception is heamol ‘miserly, niggardly’ instead of predicted *hamol. This could be an instance of back mutation, common in later Mercian sources. ErfGl has healful here, which looks like a spelling error. The second exception is fealga (not in ErfGl), which shows a ‘West Saxon’ type of breaking and a violation of the Anglian Smoothing.

Words that go through initial fronting and subsequent retraction in the Standard Theory (n = 19) are correctly predicted in 11 instances by both theories. Both theories fail on seven cases, such as ‑hebuc (ErfGl ‑haebuc) ‘hawk’ or slęgu (ErfGl slægu) ‘dross, slag’ < PGmc *slagō- with unexpected lack of retraction (ST) or with ‘Second Fronting’. The words sceadu ‘shade’ and ‑sceadan (ErfGl scheadae) ‘damage’ are best explained as instances of back mutation, rather than palatal diphthongisation (see discussion in section 3.6) and presuppose Second Fronting as an early development in the Standard Theory. The A-F chronology has no proper explanation for them either, other than indeed some form of Second Fronting.Footnote ³⁰ The dataset contains 31 tokens, which could show second fronting of æ to e, all of them without i-mutation or Anglian Smoothing (ST). These words actually have <e> in only two instances in EpGl and six in ErfGl. Altogether, if there is anything like Second Fronting in the glossaries, its impact is limited. Such a process of raising would account for the raising of the i-mutation products to /ɛ/, which is stronger in ErfGl than in EpGl (see section 3.5 and next section).

5.3 i-mutation

The categorical interpretation of the i-mutation as found in Hogg (EpGl: *a/ɑ > æ, but *æ is stem weight sensitive; ErfGl: e throughout) performs equally well as the categorical rules, based on the probabilistic cline, as they were applied in sections 4.2.1, (2) and 4.2.2., (2). Inevitably 7 errors are always found in both theories in the context _/g,k, where the distribution in EpGl is exactly 7 <ae> vs 7 <e>. A more reluctant implementation of i-mutation before a velar consonant is definitely in the spirit of the A-F i-mutation cline. As a purely probabilistic pattern, the outcome of i-mutation can be predicted with a higher accuracy by the Anglo-Frisian hypothesis (sections 3.5 and 4.2).

There are 14 instances in table 8 where the ST makes a correct prediction for EpGl and where the A-F hypothesis fails. All these 14 instances involve the application of i-mutation. For ErfGl this proportion is 8 out of 11. Using the probabilistic cline of the A-F hypothesis adds 12 more correct predictions for words with i-mutation for EpGl, which neutralises a large part of the ‘advantage’ of the Standard Theory.

5.4 Breaking and smoothing

Early implementation of breaking is pivotal for the phonological history of pre-Old English in the approach of the Standard Theory. This comes with various problematic side-effects, namely:

1. 1. The assumed early widespread breaking in words that never appear with broken vowel in later attestations;

2. 2. The fairly uncommon effect of both breaking and smoothing by the same *x;

3. 3. The atypical monophthongisation effect of i-mutation;

4. 4. The Anglian Smoothing, a process ‘difficult for scholars to adequately explain in either theoretical or phonetic terms’ (Howell & Somers Reference Howell and Somers2008: 187).

The Anglo-Frisian hypothesis posits the beginning of breaking in the late seventh century and assumes that it gradually spread throughout Old English, both in terms of dialects and in terms of phonological contexts. The performance of the ST and the A-F chronology in the prediction of broken forms is, however, not very different, and the A-F chronology performs in fact even slightly better for both manuscripts, as illustrated in table 10.

Table 10.

Performance of the theories with respect to breaking

The first category (‘predicted and not attested’) is more numerous in ErfGl, where it includes four words with ‑rn. Footnote ³¹ Apart from that, breaking in ErfGl is not widespread when the consonant following the ‑r is velar or voiceless. The spread of broken vowels before rC beyond the contexts of voiced alveolar stops, in particular in EpGl, may be interpreted as an instance of lexical diffusion, defined by Kiparsky (Reference Kiparsky, Joseph and Janda2003: 314) as ‘the analogical generalization of lexical phonological rules’, with expanding underspecification of the phonetic constraints to the position before any rC.

The ‘attested and not predicted’ forms in EpGl with ea in both approaches are fealga, leax, heamol, sceadu and ‑sceadan. The last three have been identified as early instances of back mutation (section 5.2). The forms fealga ‘harrows’ and leax ‘salmon’ are expected to appear as *falga and *læx under both theories. Both words have phonological parallels in EpGl with <e>: sadulfelgae ‘felly’ (ErfGl idem) and fex ‘kelp’; ErfGl has also lex ‘salmon’. In the Standard Theory the broken forms ‘fail’, because here Anglian Smoothing would be in place. Anglian Smoothing is also absent in mear‑/mearc- and uuearg- (see the discussion in section 3.4). Breaking of *a before *xs can be linked to the lengthening hypothesis: diphthongisation to ea, following the lengthening of earlier *a before /ks/, as can be found in Modern West Frisian flaaks ‘flax’, waakse ‘grow’ < OFri. flax, waxa. On the whole, the patterns in EpGl show more similarity to those in West Saxon than ErfGl. The breaking in fealga and leax may be an expression of that nearness. Other parallelisms between the language of EpGl and West Saxon are: (i) the wider impact of breaking of short vowels before -rC than in ErfGl; (ii) no blocking of breaking in the context *w_rC (a strong factor in Early Northumbrian); (iii) the appearance of <ae> for PGmc *ē ¹ (alongside dominant <e>); (iv) the consistent spelling of <a> for PGmc *a before nasals, against frequent <o>-spellings in ErfGl.

The word feormat (not included in the tally in section 4.2 because of the irregular <eo>) can be interpreted as support for the A-F chronology. Being derived from *farmjan- ‘to supply’, the word stem is expected to appear as *ferm- both under the Standard Theory and in the Anglo-Frisian chronology, with initial fronting before palatalised /rm/ and further raising through ‘j-mutation’ (probability for <e> in this form is 0.55 in the statistical model). The sequence -erm- is a valid input for breaking of e to eo before ‑rm (cf. geormantlab ‘mallow’) in the Anglo-Frisian chronology, where breaking follows i-mutation. The other six instances with e < *a + i-mutation before rC in EpGl do not show breaking; they stand before rk or rw, i.e. contexts where breaking of PGmc *e is not found in the glossary (cf. uuerci, smeruui, teru). The attested form feormat supports the hypothesis that breaking was later than i-mutation, because a mutated vowel could apparently still qualify for breaking on the basis of its newly acquired quality.

Although a treatment of PGmc *e and a systematic comparison with the Corpus Glossary (CpGl) lie beyond the scope of this article, there is one form that should be mentioned here, as it supports the interpretation given to feormat in the previous paragraph and provides compelling evidence for a later date of breaking than in the Standard Theory. EpGl contains feruuitgeornnis and ErfGl feruitgernis ‘curiosity’ (Pheifer no. 208). The word is a compound of words that appear in West Saxon as fyrwit ‘curiosity’, lit. ‘pre-knowledge’ and geornness ‘eagerness, desire’. The first part has cognates in Icelandic forvitni and Old High German firwizzi ‘curiosity’ and a Gothic verb fariweitjan ‘to investigate’. The Gothic, Old High German and Old English glossary forms of the compound all point to a PGmc prefix *fer- (IOB). This implies a morpheme boundary between fer- and ‑wit in pre-Old English. The vowel *e stood before single *‑r and would not qualify for breaking, which is confirmed by the forms in EpGl and ErfGl. The corresponding form in the Corpus Glossary is feorwitgeornis, with feorwit- < ferwit‑ as the first element of the compound. It seems conceivable that the morpheme structure was no longer recognisable at the time when the Corpus Glossary was written and the root vowel developed as with any other short e before ‑rC, such as in CpGl smeoruue ‘grease’, EpGl s[m]erwi (ms. sperwi Pheifer no. 769, cf. no. 944 smeruui). This could only happen at a relatively late date, when the morpheme structure of the word had become opaque to speakers of Early Old English. This testifies to breaking as an active process at the time of compilation of the Corpus Glossary. The dates of compilation of EpGl (c. 700) and CpGl (ninth century) therefore provide roughly the chronological frame for the historical period of productivity of breaking. An alternative interpretation would be to assume an error in CpGl in the first syllable, triggered by the <eo> in geornes or influence from the existing prefix feor- < PGmc *ferra.

In view of the second hypothesis within the Anglo-Frisian chronology, with early fronting before *rC, we also have to consider instances of unbroken æ or e before rC. There are four such instances in EpGl with <e> after i-mutation, which appear in breaking-unfriendly contexts (see the earlier discussion about feormat). Three words developed <ae> before ‑rC. The form auuaerdid ‘turned’ may show lack of breaking after /w/. The word baers ‘perch’ is expected to show breaking under each approach, whereas faerh ‘farrow’ is expected to show breaking in the ‘early fronting’ theory (as part of the A-F hypothesis) but not in the Standard Theory. Both baers and faerh have a voiceless second consonant in the cluster, a criterion that has been identified to inhibit breaking in ErfGl. The lack of breaking in words with a fronted vowel before ‑rC, which goes against the ‘early fronting’ hypothesis, can be interpreted as a result of the late timing of the process: the words with unbroken vowels fit specific breaking-unfriendly contexts and had not been broken yet by the time of attestation.

5.5 Conclusions regarding the performance of both theories

The Standard Theory and the Anglo-Frisian chronology both perform well and are confronted by roughly the same stock of ‘problematic’ tokens. Under the applied categorical implementation of i-mutation, the Standard Theory performs slightly better for EpGl, but there are good reasons to consider this process as a probabilistic event, governed by a series of phonological factors. This approach to i-mutation eliminates almost all instances where the Standard Theory makes better predictions than the Anglo-Frisian chronology.Footnote ³²

Exactly in contexts where the Standard Theory has its most specific implementation, namely early breaking, monophthongisation-through-i-mutation and Anglian Smoothing, the Anglo-Frisian chronology performs slightly better (see tables 2 and 10). When breaking is understood as a relatively late process, following i-mutation and particularly productive in contexts with /r/ + voiced, non-velar consonants, preferably stops, most of the observed variation in both glossaries can be explained.