1 Introduction

The debate over whether Old English (OE) had a phonemic contrast between voiced and voiceless obstruents and whether this contrast had something do with the internal processes of the language (loss of stress, syncopation of unstressed vowels) or perhaps areal influence of Celtic¹ still generates interest. The topic was discussed in classical structuralist terms by, among many others, Moulton (1954, 1972), Kurath (1956), Hogg (1992), but also in terms of Government Phonology (Wójcik 2001).

The question whether the topic is still worth pursuing lies in deciding whether earlier analyses dismissed any aspect of analysis (or perhaps did not consider raising it). As the likelihood of finding any new OE manuscripts with patterns of data radically different to those that have already shaped our understanding of the phonology of this language is almost non-existent, we must resort to analysing the existing data from a different perspective.

The shift of analysis will have to involve the reinterpretation of OE along the dimension of laryngeal oppositions. The question of whether OE was a language which contrasted phonologically voiced obstruents with phonologically neutral (= voiceless) ones will be answered in the negative. OE is better analysed as having contrastive fortis (marked) and lenis (unmarked, neutral) stops and only one series of (neutral) fricatives. Unmarked means phonologically unspecified for any laryngeal feature phonologically, marked means being phonologically specified for a particular feature (in OE, the fature of aspiration, spread glottis or GW ‘glottal width (dimension)’ of some analyses). In other words, an analysis based on privative features is adopted here.

Stops must be delimited from the fricatives based on their behaviour. It is well-known that there were two series of stops and one series of fricatives in OE. Judging by what we can gather from the manuscripts, stops that are traditionally taken to show voiced sounds are not normally found devoiced: scandful ‘shameful’ is not normally found as **scantful in spite of being followed by f [f], a phonetically voiceless fricative (in all known accounts on fricatives). It is also commonly accepted that fricatives were voiced between voiced sounds (as in hūsian ‘to house’, which contains a voiced fricative as evidenced by modern English house). The stops are (held to be) voiced in their own right (bīdan ‘wait’, representing [bīdan], phonologically /bīdan/). Yet, there are no examples of regressive voicing (lēohtbǣre ‘luminous’ is not found as **lēohdbǣre/lēogdbǣre).

The contested question of whether a voiced environment on both sides of a fricative was sufficient for a phonetically voiced fricative must also be addressed, and whether it was possible to have a phonetically voiceless (!) fricative when it was surrounded by voiced sounds in a stressed syllable. The most controversial issue of all is whether OE had phonologically contrastive voiceless fricatives, the answer to which will be yes, it did have them, but not as traditionally assumed. We will try to reanalyse some of the views on the laryngeal oppositions in OE and see whether these are better described in terms of lenis-fortis, rather than voiced-voiceless, a claim which has not been made so far for OE (but one that has been proposed for Old High German, and Common Germanic). Before we try to reassess the data, it will be necessary to look at OE from the perspective of laryngeal oppositions. The article is structured as follows: in §2 OE is presented in terms of laryngeal oppositions; in §3 the fricatives are analysed. §4 is devoted to the history of fricatives in OE. In §5 a reanalysis of the stops along the lines of lenis-fortis is attempted. §6 tries to reassess the status of lenis and fortis geminates, and finally, §7 tackles some of the contentious issues of OE fricative (de)voicing.

2 Old English in terms of laryngeal oppositions

The reference accents of Modern English (MoE), like Southern Standard British, General American, or German are phonologically different from the reference dialects of Hungarian, Serbo-Croatian or Dutch. The latter three have a contrast between phonologically voiced (marked) and voiceless (unmarked) obstruents, the former three a phonological opposition between fortis (marked) as opposed to lenis (unmarked) obstruents. The phonological difference between these two types of languages is not that English invariably has phonetic aspiration on /p t k/ in all positions, as opposed to a language like Hungarian, which never does. It is the behaviour of the two classes of obstruents in the two types that is responsible for the classification. There is a long-standing tradition behind the classification of languages into either voice/voiceless or aspirated/non-aspirated, as discussed by Honeybone (2002, 2005). The question of what feature to choose to represent the opposition in a language with a binary set of oppositions in the obstruents is a moot question.

A mere look at the orthography of a language is not revealing: seeing a 〈b〉 (as in English book, Serbo-Croatian braniti ‘defend’) (orthographically) opposed to a 〈p〉 (pack, praviti ‘do’) does not say much about which feature of the opposition is relevant phonologically. Phonology does, however, show which feature is active: in English there is no regressive voicing assimilation in, for example, backbone ([bakbəwn], **[baɡbəwn]), just as there is no regressive assimilation of voicelessness in bagpipe ([baɡpajp], **[bakpajp]). In Serbo-Croatian, however, there is such an assimilation: the prefix od- ‘away from’ is found as either [ot] or [od] depending on the voicing specifications of the first obstruent of the verbal stem (odbraniti ‘defend against’, otpraviti ‘send away’).² The question of why there is no voicing assimilation in either backbone or bagpipe can be handled in two ways: under the assumption that English and Serbo-Croatian have the same set of underlying voiced (/b/) vs voiceless (/p/) phonemes, the difference can be handled as deriving from the absence of some kind of assimilation rule (in English). This tradition (called tradition (i) in Honeybone 2005, 327) views English and Serbo-Croatian as being underlyingly identical in their set of oppositions in the obstruents (cf. Lisker & Abramson 1964; Lombardi 1991, 1994; Wiese 1996; Hammond 1999, and many others). T.A. Hall (2001, 32) refer to this as the ‘broad’ interpretation of the feature [voice]. In the ‘broad’ interpretation of [voice] originating in Lisker & Abramson (1964, 389), who emply binary features, oppositions between plosives at the phonological level may be implemented by any two of the following three phonetic categories: negative voice onset time (as in Serbo-Croatian braniti), short-lag voice onset time (as in standard English bone) and long-lag voice onset time (as in English pony).

The other approach (referred to as tradition (ii) in Honeybone 2005, 328, something already present in Jakobson 1949, 423–424) has a ‘narrow’ interpretation of the feature [voice] with ‘voiced’ applied to only those phonetically voiced sounds with actual vocal fold vibration that disclose phonologically testable voicing (as in regressive voicing assimilation) as opposed to merely phonetic/predictable/passive voicing (cf. the first clear proposal for the ‘narrow’ interpretation of voice by Rooy & Wissing 2001).³ In terms of the ‘broad’ interpretation (of Lisker & Abramson 1964) the [voiced] member of a pair of opposites must have negative voice onset time phonetically (as in Serbo-Croatian braniti) to be relevant phonologically. The ‘narrow’ interpretation holds that Serbo-Croatian and English embody two different types of languages: the former has a phonologically marked/voiced series of obstruents (whatever the graphical exponent is) opposed to an unmarked/neutral series (/b/ braniti vs /p/ praviti), the latter has a marked/aspirated series opposed to an unmarked/neutral one (/p^h/ pack vs /p/ back). The orthographic coding of these differences is immaterial phonologically: English /b/ may be spelled 〈b〉, as in cab or asbestos, or 〈p〉, as in apt (cf. Szigetvári 2020). English [p^h], however, is never spelled 〈b〉. How close the relationship is between orthography and the underlying phonological reality is again a matter of tradition, the cultural background behind the tradition of literacy, the conservatism of a spelling system, etc.⁴

This ‘Germanic’ tradition refers to /p/ as lenis, /p^h/ as fortis. This tradition can be traced back at least to Sievers 1876, Winteler 1876, Jacobson 1949, Jacobson, Carl Gunnar Michael Fant & Halle 1952, Catford 1977, Kohler 1984, Iverson & Salmons 1995, 1999, 2003, Petrova 2011, Stahlke 2003, Honeybone 2005, Szigetvári 2020, and many others use phonological arguments to support this ‘Germanic’ view of the nature of oppositions in the (Germanic) obstruents. In Honeybone (2002, 2005) this view is termed ‘laryngeal realism’, describing a set of assumptions rooted in the phonological reality of the laryngeal features (as measured against the set of phonological processes likely to result from these). The features used to embody the fortis-lenis opposition are many and varied, but the underlying assumption remains the same: lenis obstruents are unmarked/neutral/possessing no phonologically encoded feature, fortis ones are marked by possessing a [spread glottis] (Iverson & Salmons 1995), [+tense] (Jessen 1999), H (Harris 1994), [aspiration] (Lombardi 1991, 1994) feature or a GW ‘glottal width’ dimension (Avery & Idsardi 2001). Strictly speaking, which mnemonic label is used to encode the opposition is immaterial (cf. also Lisker 1986, 4). The basic tenet, however, remains that in laryngeal realism only privative features can be used and consequently oppositions can only be privative.

The question now is what characteristics fortis-lenis languages such as English exhibit phonetically. They will:

1. (i)typically have aspiration on the phonetically voiceless obstruents (in English, a fortis obstruent is strongly aspirated in word-initial position and/or when followed by a stressed vowel),
2. (ii)lenis series will be passively voiced between sonorants (in English, a lenis obstruent is phonetically voiced when both followed and preceded by a sonorant or other lenis obstruents which are in turn preceded/followed by other sonorants or lenis obstruents, cf. Stahlke (2003, 200), Cruttenden (2014, 164), Szigetvári (2020) another),
3. (iii)there will be instances of assimilation to voicelessness, rather than voicedness, so post-fortis sonorants are often phonetically devoiced (in English, prime, queen, tune all have phonetically voiceless/aspirated sonorants: r̥ w̥ j̥). In English, lenis obstruents word-initially and finally are phonetically voiceless, but the cues for recovering a phonetically voiceless (lenis) obstruent after a stressed vowel are not lost: stressed vowels are longer before lenes than they are before fortes (ridge vs rich), cf. Stahlke (2003), Cruttenden (2014), Kaye (2014), Wells (2008), Szigetvári (2020). Quite to the point, Lisker (1986) lists 16 properties associated with voicing in English and expressly states that the list is incomplete.
4. (v)A lenis obstruent in a language like English is not strong enough to carry phonetic voice on its own, but this does not preclude it from being passively voiced phonetically (as in abbot, where /b/ is lenis phonologically and passively voiced phonetically, i.e. [b], in want of a better symbol)⁵
5. (vi)A fortis obstruent can never be phonetically voiced, but it can appear without aspiration (in bat, the /t/ is fortis phonologically and not aspirated phonetically). It must be noted that a feature like H ‘aspiration’ carries intuitively a rather well-defined phonetic interpretation (‘a puff of air’), but this feature (or the one used in its stead, ‘h’, here) is more abstract than this: in some languages it can be realised as pre-aspiration (as in Icelandic under certain conditions, e.g. epli ‘apple’ [ɛhplɪ], opna ‘open’ [ohpna], cf Árnason 2011), its presence in a fortis fricative does not necessarily imply an aspirated fricative, given that there is no audible release phase in a fricative, but rather an articulatory gesture that can be timed to be sufficiently different from that of a lenis fricative (cf. Vaux (1998) who argues that voiceless fricatives in Julfa Armenian are [spread] glottis). (Phonetically) Aspirated fricatives, however, are not an impossibility, as shown by Burmese (Ladefoged & Maddieson 1996).
6. (vii)Stops that are laryngeally marked for ‘h’ may not be aspirated at all: MoE hit, with /t^h/, has no aspiration on [t], but it can always be distinguished from /d/ (phonetically [t]) in hid by virtue of other acoustic cues (a shorter vowel before [t] or pre-glottalisation, or a combination of the two, not found for /d/ in hid). This is why the term fortis (vs lenis) is more neutral with respect to the acoustic correlates that the feature ‘h’ stands for.⁶ No analyst working with privative features will deny that while phonology is stripped to the barest relevant minimum, phonetics is overspecified and brimming with equipollent oppositions (cf. Avery & Idsardi's (2001) analysis of English and Korean): in atop vs adopt the phonological opposition between fortis /t/ and lenis /t/ is not only characterised phonetically by aspiration but also (passive) voicing ([t^h ] vs [d]), to mention just two of the commonly assumed differences.

After this short detour into modern English, let us see what we can glean from OE. Quite understandably, in the major works on OE phonology, like Hogg (2011) or, more recently, Ringe & Taylor (2014), the terms ‘aspirated’ or ‘lenis’ are not even mentioned. Hogg (1992, 90), however, says “there is no way of telling whether these stops [/p t k/] would have aspiration in initial position in the syllable, a characteristic feature of present-day English which distinguishes it from, say, Dutch.” It is ‘voice’ and ‘voiceless’ that are found to the exclusion of ‘fortis’ and ‘lenis’. But then, the justification for the use of exactly these features (and their ramifications) is never explained: does ‘voiced’ in any description of OE mean phonetically voiced or phonologically voiced? Naturally, the effect of phonological voicing would be the only feature that is relevant for a historical analysis and expected to show up as such.⁷ The lack of reference to fortis/lenis or, indeed, the unquestioned acceptance of a phonological system rooted in a voiced-voiceless set of oppositions may have its origin in the Latin-based spelling system of OE, (blindly) applied to the phonology of OE. After all, a 〈b〉 in bone ought to be a /b/, which is phonologically voiced. What's more, what does a change from 〈b〉 in describe to 〈p〉 in description show, if not devoicing (and what distinguishes it from the apparent lack of such a change witnessed by 〈b〉 in sub in subscribe/subscription)? Almost all technical textbooks (e.g. Hogg 2011, 246) contend that pre-OE /b/, for example, was realised as [b], as in *bindan ‘bind’, *habbjan ‘raise’, etc. Little discussion is offered on the correspondence between recorded OE 〈b〉 and the supposed phonological representation (/b/), whose pronunciation was [b] word-initially (bindan). We will see that we have reasons to suppose that 〈b〉 was in fact [p], and 〈p〉 (at least) word-initially was [p^h]. Whatever the phonetic difference was between the sounds represented by 〈b〉 and 〈p〉, the opposition must have been a real one in phonological terms.

A notable difference is Minkova (2014, 27), who devotes a short discussion to the degree of glottal opening viewed from the history of the language: (i) the non-cooccurrence of /h/ on aspirated stops (**hp, **ht, **hk, cf. hn, hl, hr, hw) and (ii) the distributional facts (aspirated stops and /h/ are found in identical environments: betoken – behave) suggest that aspiration may have been phonologically relevant. However, Minkova chooses to follow the tradition on ‘voiced’ vs ‘voiceless’ in her account of (O)E phonology, noting though that it may have its flaws.

We should find those points of reference which could help determine whether OE was a fortis-lenis or voiced-voiceless type of language in its system of obstruents. One piece of evidence could come from voicing assimilation across morphemes. The evidence we can gather can only come from the spelling of the stops from, for example, 〈d〉 for expected 〈t〉 or 〈t〉 for 〈d〉, not the fricatives (discounting certain examples of ‘hardening’ from, /θ/ to /t/, for example, which may or may not show a change in the laryngeal specification), whose phonetic voicing depended on context (or at least so it seemed to OE scribes), so their differentiation was never attempted by any scriptorium.

Theoretically, we could expect to find signs of voicing assimilation across any free (root) morpheme. This, however, is not a characteristic feature of OE: in a sequence like eft byreð ‘bears again’, post-lexical voicing (if any) of 〈t〉 to 〈d〉 is not shown (**efd), neither is post-lexical devoicing shown in a sequence like friðusibb folca ‘peaceful tie of nations’ (**-sip(p)). The data in (1) shows the principal derivative suffixes (morphemes that can be thought of as forming a close-knit unit with the preceding word) beginning with a letter not showing a sonorant sound or a vowel.

The expectation in a voiced type of language could be to find 〈b〉 regressively voicing the preceding non-sonorant sound, or 〈c〉 /k, c/ regressively devoicing the preceding non-sonorant sound (as it would be the case in Serbo-Croatian or Hungarian). As Fulk (2002, 2.1) remarks: “certainly assimilation was not an absolute requirement in quasi-compounds.” Using Fulk's examples, see (2) below for something we never find in OE.

All this despite the ease with which consonants seem to assimilate in voicelessness across clusters (to which we return in 7.7): *blēdsian > bledsian/bletsian/blitsian ‘bless’, *bīdiθi >*bīdθ > bītt/bīt ‘waits’, etc. If OE stops had been phonologically voiced, one would expect to find more (haphazard) examples of regressive voicing or at least some regularity applying to some of the places of articulation (the velar as opposed to coronal or labial, for example). One could say that in words formed with the suffixes in (1) the spelling of the derived words with the voiced or voiceless sounds was reinforced by the simplices, but then the question of (regular) forms like bītt/bīt (from bīdan) remains obscure. We can also see that whatever feature was responsible for the differentiation of 〈b〉 and 〈p〉, 〈d〉 and 〈t〉, 〈g〉 and 〈c〉, 〈cg〉 /ɟ/ or /dʒ/ and 〈ć〉 /c/ or /tʃ/, this feature was salient enough phonologically to cue the laryngeal contrast in phonologically weak positions, too, i.e. word-finally and intervocalically. Although not every pair of laryngeally differentiated stops could occur in all positions for diachronic reasons (〈cg〉 /dʒ/, for example, was not found word-initially, or /ɡ/, which did not occur word-finally and intervocalically), the contrast was not generally neutralised, see (3), which is not an exhaustive account of the positions in which non-sonorant stops are found.

(3)

Non-sonorant stops in the weak positions

All in all, there are no signs for any lenition in the two weak positions displayed in (3): no regular devoicing word-finally,⁹ no voicing intervocalically. Of course, the word-initial position remains ideal for cuing all the oppositions, but even that position will lack some of the theoretically possible oppositions with a following vowel (/ɡ/, for example, will not be found word-initially before a front vowel in recorded OE, having been palatalised). OE is thus unlike Danish, in which only the word-initial position can cue the laryngeal differences in the stops. In the weak positions, the Danish stops are all voiceless lenes for all places of articulation, cf. Haugen (1982, 81), Page (1997): tyk ‘thick’, lække ‘leak’, and tyg ‘chew’, lægge ‘lay’ all have [g̊ß]/[k]). We will return to the question of whether OE had lenis stops. The spelling itself is no evidence, but the absence of regressive assimilation across morpheme boundaries may point towards lenes. We will see more data as we proceed.

3 The fricatives of Old English

The obstruental stops of OE showed no signs of lenition in phonologically weak positions. Generally, all the stops could be found in all the positions, weak and strong. The fricatives of OE are altogether more interesting. The general consent is that there was just a single series of fricatives phonologically in recorded OE not distinguished by opposing laryngeal specifications, just place of articulation. Phonetically, there were voiced and voiceless allophones of each (voiceless) phoneme. The standard literature (e.g. Campbell 1959; Hogg 1992; Lass 1994; Ringe & Taylor 2014) does not discuss the potentially interesting ramifications of the claim that the [+/−voice] feature was just a surface phenomenon. The discussions all converge on the claim that voicelessness (i.e. [−voice]) is the default value of a fricative. The feature [−voice] is changed into [+voice] in a voiced environment: voicing can come from both voiced obstruents and vowels/sonorants, which means the voicing value of a fricative depends on the environment, i.e. they are not strong enough to license their own [+voice] feature, even in a phonologically strong position (like the initial). Ringe & Taylor (2014, 6.7.2), for example, claim that the anterior fricatives ([f, θ, s]) were voiced ([v, ð, z]) in fully voiced environments if the preceding syllabic nucleus was stressed. In other words, fricatives were not voiced before a stressed vowel, even word-initially, or after an unstressed vowel: in faran ‘go’, befáran ‘go round’, sendan ‘send’, aséndan ‘send forth’, þýncan ‘seem’, geþánc ‘mind’ we have no reason not to suppose that the consonants in bold were not phonetically voiceless. Hogg (2011, §7.54) formulates the rule (somewhat incompletely) as shown in (4) below.

Hogg (2011) makes no reference to the stressed syllable as a conditioning factor, he simply contends that a fricative is not voiced if there is an unstressed syllable immediately preceding it (as in befaran). The rule in (3) is followed by 4 explanatory notes, which take care of some of the contentious issues. One of the two relevant notes for the discussion is note no. 2, which tackles the absence of voicing in geminate fricatives and */sk/ with the palatalised reflex /ʃʃ/. They regularly fail to abide by the rule in (3), but no argumentation is given for this. If a geminate is simply a sequence of two voiceless fricatives, why aren't they voiced given the right environment? They obviously were not; not a single example survives with a voiced fricative originating in a voiceless geminate, which must be more than just a coincidence. Note no. 4 concerns the (apparently) many exceptions to the rule in (3). Fricatives in an unstressed syllable in early (or pre-)OE were not voiced. This explanation is due to Luick (1914–40, §639.2) and concerns some of the suffixes: the abstract suffix *-iθo and the ordinal suffix *-oθa, for example (there are no modern reflexes that show a voiced fricative: mirth, seventh, from OE mirð, seofoða), as well as the suffix *-isōjan, found in MoE cleanse (with /z/, from OE clǣnsian), and bless (with /s/, from OE bletsian, bledsian < *blōdisōjan). Also problematic is MoE adze (with /z/, from OE adesa < *adesan-, which shows that the fricative was originally in an unstressed syllable, so here a voiced fricative is unexpected), etc. We will return to these in Section 7. Perhaps not every voiced (or voiceless) fricative of MoE can be regarded as showing a direct continuation of the OE voiced (or voiceless) fricative. They are not odd, in any way, from the point of view of MoE phonology (cleanse is like lens).

There existed only the graphemes 〈f〉 (in early manuscripts also 〈b〉), 〈s〉, 〈þ, ð〉, 〈h〉 (also 〈g〉), 〈sc〉 to represent the fricatives of OE. No attempt was ever made to systematically distinguish 〈þ〉 from 〈ð〉, which points to an allophonic distribution of the voiced as opposed to the voiceless pronunciation: word-finally they are voiceless, word-internally in a stressed syllable they are voiced (some often-cited examples follow in (5)).

According to Luick (1914–40, §639.2) there was no voicing of fricatives in unstressed syllables, as in the ones shown in (6).

If the distribution of the allophones in (5) and (6) is correct, the rule in (4) is obviously inadequate to capture this complexity. The question is why there is no voicing of the underlined fricatives in a voiced environment in (6). Note also that the rule in (4) makes no reference to whether the presence of a sonorant consonant on either side of a fricative made any difference to the allophonic rule (we will see later that it did).

There is a further problem, one with the importance (or not) of the notion of syllable. Even those who believe that no OE stressed syllable could be light (having a short vowel only)¹⁰ in a polysyllabic word, would not find it convincing that a word like wulfas [wulvas] or earfoþu [earvoθu] should be syllabified as wulf.as or earf.oþu just to ensure that /f/ finds itself forced into a stressed syllable. Wul.fas and ear.foþu would make the first syllable heavy, no further consonants need to be captured by the stressed vowel.¹¹ Appealing to further ‘captured consonants’ is theoretically suspicious.

However, if this is so, even the often-cited example of wulfas is not straightforwardly transparent anymore: 〈f〉 /f/ is in the onset of an unstressed syllable now, so it should be [f] (as in faran). If one still argued that in wesan the syllabification was wes.an to make sure that no stressed syllable of OE was light through coda-capture,¹² we would now be faced with an unrecorded and unwarranted bifurcation of voiced and voiceless allophones: voiceless after heavy syllables (wulfas, earfoþu), voiced after heavy syllables that show the result of coda-capture (wesan). Furthermore, if wesan is syllabified as wes.an, the next step is to admit that /s/ is now in the coda (the end of the syllable), where no voicing is expected otherwise (pæþ, wulf). One would certainly have to make stipulations about the difference between word-internal and word-final codas. The syllable as a theoretical notion does not satisfactorily explain why wulfas and wesan both have voiced fricatives. The question of whether the feature [−voice] is theoretically justified must also be answered in the negative. If a fricative is voiceless, it may merely be phonetically voiceless and, simply, unmarked/neutral phonologically. Its phonetic realisation is either voiced or voiceless depending on the environment, phonologically, however, it is not marked for any laryngeal feature. Such an obstruent is a lenis one phonologically, it does not bear the feature [voice], it is neutral. Now it is time to see how this state of affairs came about.

4 The history of Old English fricatives (noch einmal)

Moulton (1954, 1972) gives a structuralist account of the consonants of Common Germanic (CG) from the point of view of the developments of some of the Germanic daughter languages, including OE. (7) shows the CG heritage in West Germanic.

(7)

Common Germanic heritage in the (labial) obstruents of West Germanic

(7) shows the distribution of the labial obstruents with Y showing possible and N showing impossible instances. As can be seen, and as traditionally assumed, [b] and [v/β] are the allophones of the same phoneme (/b/ in all likelihood): wherever [b] is found [v/β] cannot (word-initially and after a nasal, the stop allophone is found, elsewhere the continuant). The West Germanic innovations are shown in (8).

(10)

West Germanic intervocalic contrasts for the fricatives after the changes in (8)

Change 1. in (8) increased the number of occurrences of /d/ in the language (previously found word-initially and after a nasal). It did not result in a new phoneme, as Ringe & Taylor (2014, 42) say: “In PWGmc the non-coronal voiced obstruents continued to exhibit [the continuant-stop] allophony, but *d [meaning /d/ which is [ð] intervocalically and word-finally] became a stop in all positions,” an example for which is *dēðiz ‘deed’ (OE dǣd). Change 2. in (8) merged all instances of /z/ with an already existing /r/ (*maize- > OE māra ‘more’, *gaiza- > OE gār ‘spear’). Whatever phoneme one reconstructs on structuralist grounds to underlie the stop ∼ continuant alternation of [d] and [ð], the fact remains that (phonetically) now [d] is found in those positions in which it did not occur previously (or alternatively, [ð] was removed from positons in which it stood originally in contrast with [θ]): *dauθu- (OE dēaþ) vs. *dau[d]a- (OE dēad). /z/ and /r/ had been phonemes since Germanic times, but this change deleted all phonetic instances of [z] which could earlier be contrasted with [s] in similar environments: *kuzun (OE curon ‘they chose’) vs *nusō (OE nosu ‘nose’, cf. Hogg 2011, §3.10, note 2). For a speaker acquiring a form of post-West Germanic (leading towards OE), crucial contrasts in the fricatives are lost. There is no change disturbing the distribution of [f] and [v] yet, nor is there a change affecting the distribution of [x] and [ɣ]. As we can see, in proto-West Germanic there was no regularity comparable to the intervocalic voicing of OE: intervocalically [f] was contrasted with [v], [θ] with [ð], [s] with [z] and [x] with [ɣ]. The laryngeal differences (whatever they were) could be cued phonetically in non-initial positions in proto-West Germanic. After the change in (8), (10) ensues in (later) West Germanic, showing ‘unpaired’ [s] and [θ].

OE had its own innovations (see (11) below). One of these was the debuccalisation of [x] to [h] in onset position, followed by the loss of [h] in intersonorant position after a stressed vowel (see (12) below). Word-finally and before obstruents (including itself) [x] was retained, as in hēah ‘high’, niht ‘night’, hweohhol ‘wheel’. The distribution of the fricatives can be seen in (13) following the early OE innovations in (11).

(13)

Distribution of the fricatives after the changes in (11) (intervocalic and word-final oppositions shaded for identical places of articulation)

The distribution in (13) shows that intervocalically the following singleton fricatives remain phonetically: [f v/β s θ ɣ]. The coronal ones have no voiced congeners, out of the two non-coronals, the velar is always voiced, the labial ones seem to qualify for the only voiceless-voiced pair, although there may have been a difference in the place of articulation (labial vs labio-dental, which would have made this pair rest on a very tenuous difference, a feature soon lost).

Word-finally only the non-coronals were found in contrastive pairs: [f v/β] and [x ɣ]. That there must have been some difference between the two labials is supported by early spelling practices (Hogg 2011, §2.58; Campbell 1959, 197), which have 〈b〉 for [v/β] and 〈f〉 for [f]: scēabas ‘sheaves’, salb ‘ointment’, geab ‘gave’, Albred ‘Alfred’, but wolf ‘wolf’, ofen ‘oven’ (in German the historical difference is still seen in gab vs Ofen). At this early time (before the onset of ‘intervocalic voicing’), the labial place of articulation was the only one that had a (phonologically) contrastive pair of fricatives in intervocalic and word-final positions.

A system like this is badly in need of reorganisation. The change occurred in the form of loss of whatever laryngeal feature [v] had to distinguish it from [f]. This brought it in line with the rest of the fricatives: now there was just one neutral/lenis labial fricative (/f/) whose phonetic voicing hinged on the environment: whenever it was next to a sonorant or a lenis stop which itself was in contact with a sonorant or a lenis stop, it was voiced. This loss of phonemic opposition between /f/ and /v/ in favour of a lenis /f/ brought with itself another change: a lenis fricative word-finally (i.e. when not supported by sonorants/lenis stops) could only be rendered phonetically voiceless. Word-final /v/ was lost when intervocalic lenis /f/ was born, and now the only fricative to be brought in line with this massive reorganisation was the velar fricative. It was phonetically devoiced word-finally following its reinterpretation as a lenis fricative /x/, one with a defective distribution though: it was found word finally as a phonetically voiceless [x] (collapsing with the voiceless [x] allophone of the historical /x/ now), and as [ɣ] when surrounded by contunuant sonorants/lenis stops. Word-initially (and word-internally before a stressed vowel), historical */x/ came to be reinterpreted as [h], phonologically /h/ (its presence was phonologically relevant, cf. healdan ‘keep’ vs ealdan ‘old ASg’).

A new system of lenis fricatives was born, whose phonetic voicing was regulated by the environment. The once contrastive system of fricatives gave way to a system of non-contrastive /f/, /θ/, /s/, /x/, /h/, which may be shown as /f⁰/, /θ⁰/, /s⁰/, /x⁰/, /h⁰/ (cf. Honeybone 2005, 332). The last member of the family was no longer a fricative, it came to be reinterpreted as voicelessness or aspiration (/h⁰/). The palatal fricative historically comes from /sk/ and can be assumed to have been a geminate (cf. Hogg 2011, §7.37), which was degeminated word-initially (and probably word-finally, too, after the degemination that affected all the geminates, like webb ‘web’ > web, cf. Hogg 2011, §2.53, §7.80). There is certainly no example for (MoE) /ʒ/ from the intervocalic voicing of OE /(ʃ)ʃ/. The system of singleton lenis fricatives resulting from the early reorganisation of inherited OE fricatives can be summarised in (14).

(14)

The fricatives of Old English

The advantage of a system of fricatives in (14) is that the notion of phonetic voice (or voicelessness) has been disentangled from the notion of phonological unmarkedness of the laryngeal specification for this class of OE obstruents. We can assume that intervocalic voicing as a phonetic process still exists: lenis fricatives were passively voiced between phonetically voiced sounds (as in wesan, wulfas). Word-final and word-initial devoicing is simply the correlate of the neutral laryngeal specification of these fricatives: they cannot be passively voiced because they are not couched between phonetically voiced sounds (wæs, wulf). Some of the allophones in (14) have nothing to do with the historical phoneme (or phonemes) to which they originally belonged in pre-OE: [x], historically viewed, is the word-final allophone of both */x/ and */ɣ/, [h] is the word-initial allophone of */x/, [v] may come from either earlier */f/ or */v, β/, [ɣ] comes from */ɣ/ (but never from */x/ of pre-OE), etc. But this is a welcome step because the OE described here is fundamentally different from its predecessor not yet characterised by the changes in (12), let alone the kind of West Germanic from where it originates. We will discuss the issue of geminate fricatives like 〈f〉, 〈ss〉, etc., in Section 6. These appear to have been always phonetically voiceless (see Section 5 below).¹³

5 Old English stops as lenes

In Section 2 we saw that there was no voicing assimilation across morphemes (or at least there are no orthographic signs of it) in either direction (recall lēohtbǣre/**lēohdbǣre ‘luminous’, scandful/**scantful ‘disgraceful’, scyldhǣta/**scylthǣta ‘deputy’). This may simply have been because in OE obstruental stops were not specified for [+voice] (or [−voice]) laryngeally. This did not preclude them from undergoing phonetic voicing, of course, when couched between sonorants/other lenes (as in dǣda ‘deed, NAPlFem’). We have also seen that whatever feature it was that differentiated /t/ from /d/, this feature was well-cued phonetically in all positions, making OE not exhibit cases of lenition affecting the stops (the /t/ in bītaþ ‘bite, Pl’, for example, never shows conflation with /d/ in bīdaþ ‘wait, Pl’).

Some further data for the stops is shown in (15). The past tense marker of weak verbs in OE appears as 〈d〉 or 〈t〉 depending on the environment. When this suffix immediately follows a stem-final lenis fricative, the /d/ is unchanged and appears as 〈d〉 (past tense forms like brōhte, þōhte will be discussed in 7.10).

After the deletion of *i after heavy syllables (in Class I weak verbs), *d came to follow the lenis fricative (hogde never contained an *i, but behaves identically to the rest of the fricatives). The 〈d〉 here shows that the preceding lenis fricative was phonetically voiced if the fricative was preceded by a vowel or a sonorant consonant and followed by a lenis stop, followed by a sonorant (similarly to MoE husband). All authorities agree on this, although not along identical lines. One can argue that this may simply show the effects of an underlyingly voiced stop on a preceding fricative (OE would at this point behave like a voicing language). We need to broaden the horizon somewhat and look at the data in (16).

Now we must examine the distribution of the past tense suffix. In (15), we see that the past tense suffix is realised as [d] 〈d〉 after lenis fricatives. After sonorants (16a) it is also [d] 〈d〉, not surprisingly. In (16b), after (geminated) voiceless stops it appears as [t] 〈t〉. In (16c), after voiced stops it is found as [d] 〈d〉. After geminate fricatives (16d) it is found as [t] 〈t〉. Generally, one can say the past tense suffix in this class of verbs shows a simple alternation of voiced/voiceless depending on the last sound of the verbal stem (if it is voiceless, the suffix is [t], if it is voiced, the suffix appears as [d]). From this should follow what traditional wisdom has always argued for: the stops of OE ought to be distinguished with the feature [voice], /ʃ/, for example, being [−voice], /b/ being [+voice]. This seems a straightforward enough distribution, but some doubt must be cast on whether this is the right description. We have already seen that the fricatives are best analysed as lenes. Can the same be done for the stops? The answer is yes, and such an analysis was attempted earlier for Old High German by Moulton (1972, 1.2.7), Penzl (1968a, 147), Goblirsch (1997, 136). Penzl (1971, 150–152) reconstructs a fortis/lenis opposition for Common Germanic.

The expectations in a voicing language is that the feature [voice] should be able to spread regressively onto obstruents (similarly to Hungarian or Serbo-Croatian) given that this is a distinguishing (marked) feature of these consonants. This is, however, not what we see for OE: the past tense suffix [d] is unable to spread voice backwards, which would mean that it had no such laryngeal feature to spread. If, for the sake of argument, we take OE to be a voicing language, the following pattern should emerge instead (see (17) below).

In this type of alternative OE, the past tense suffix spreads its voice feature regressively, leading to unattested forms in (17a) and (17c). In contrast to these, (17b) would not disclose the active laryngeal feature of this alternative OE (we would expect cembde for both this type of OE, and the one attested). We also have to conclude that only looking at the behaviour of the fricatives in contact with the past tense suffix (in (15)) would not disclose what type of language OE was. In this alternative (voicing) OE, the past tense suffix /d/ would be expected to voice the fricatives, leading to identical results with those observed for attested OE ((17d) = (15)). On balance, we have to conclude that OE voiced stops were phonetically voiced (in the right environment), but phonologically not marked for any laryngeal feature. In other words, they were lenis: /b⁰/, /d⁰/, dʒ⁰/, /ɡ⁰/. When preceded by a sonorant/vowel (or lenis fricative) and followed by a sonorant/vowel, they were passively voiced (rǣsde, cȳþde, cem(b)de, gemengde, iernde). Of course, the lenis fricative was also passively voiced under the same circumstances (rǣsde, cȳþde).

In OE it is the voiceless stops and (geminate) fricatives that devoice the past tense suffix (cēpte, pyfte), rather than the past tense suffix that voices the fricatives regressively. Now that OE has proved not to be a voiced language, but one that has lenis stops and fricatives, a new feature must be invoked to explain cēpte and pyfte. In MoE this is conveniently the feature of aspiration, which resides in the marked members of the obstruents (Honeybone 2005; Szigetvári 2020). We may assume this feature was present in what is thought of traditionally as voiceless stops of OE, making OE an aspirating language phonologically: /p^h/, /t^h/, /tʃ^h/ (or /c^h/) and /k^h/. This feature now distinguishes the first stop of cnāwan ‘know’ (/k^h/) from gnagan ‘gnaw’ (/ɡ⁰/). The feature of aspiration is used here as a phonologically distinguishing feature, which may not always have been present phonetically (it may have had other phonetic correlates, like pre-glottalisation, etc). How the feature of aspiration was actually realised phonetically is something that will have to remain a matter of conjecture. There are no proprioceptive descriptions dating from this period, either native or non-native. We know of no OE orthoepist who has an axe to grind with any other scribe or scriptorium over issues pertaining to pronunciation or how pronunciation appears in spelling. Even granting all this, OE stops were distinguished by a laryngeal feature, one which can't have been the feature of [voice], but rather the feature of aspiration, shown as ‘h’ here.

The presence of the feature of aspiration as an underlying distinguishing feature of stops can also be seen at work in the impossibility of these aspirated stops being passively voiced. This is just a reformulation of the fact that OE shows no signs of ‘intervocalic lenition’ of the stops: the distinction between bīdan and bītan is never lost, the two stops never merge.¹⁶ The salience of this position is maintained by the presence of ‘h’ in /t^h/ in bītan. One could argue that this is still the case in MoE, seeing the feature of ‘h’ at work in biding vs biting. What the phonetic correlates are of this ‘h’ is yet another (complex) matter: pre-fortis clipping, pre-glottalisation, etc. that help preserve the opposition between the phonetically voiceless lenis and fortis consonants (Jansen 2004). The categories of /d⁰/ and /t^h/ have not been (generally) merged in this language, neither have those of /b⁰/-/p^h/ or /ɡ⁰/-/k^h/.

If we accept that OE was an aspirating language distinguishing phonologically aspirated stops from lenis/neutral ones, the question of why there is no voicing assimilation in the oft-cited forms swicdōm, lēohtbǣre and scandful receives a straightforward explanation: the lenis stops have no [voice] feature to spread. The question of why devoicing of 〈b〉 and 〈d〉 is not shown in sibsum and rǣdfæst can now be attempted from a totally new perspective: /b⁰/ and /d⁰/ were considered phonetically different enough to preclude them from being categorically merged with /p^h/ and /t^h/ (**sipsum, **rǣtfæst). After all, /b⁰/ and /d⁰/ are followed by another lenis consonant, followed by a vowel. What the pronunciation of 〈bs〉 /b⁰s⁰/ and 〈df〉 /d⁰f⁰/ was ([bz], [dv], [bs], [df], [b̥s], [b̥z̥] or anything else that an analyst feels captures the phonetic reality of the two lenes in this ‘across the word’ environment) is impossible to determine from MoE (but cf. MoE wisdom, which has two passively voiced lenis obstruents probably continued from OE times).

The question of what went on phonetically (and phonologically) in cēpte and cyste can be answered in two ways: either the ‘h’ feature of /p^h/ and /f^h/ spread onto the lenis /d⁰/ making it a fortis /t^h/ (which would qualify this process as an assimilatory one) or /d⁰/ could not be phonetically passively voiced (cf. Cyran 2014, 203) after a fortis consonant (this would make ‘h’ a feature unable to spread). The fact that /d⁰/ in cēpte sounded phonetically as [t] (whatever the exact phonetic correlates were) does not mean we have to analyse it as /t^h/ (= ‘h + d⁰’).

Obviously, OE scribes were not phonologists. They followed their ears when they spelt the phonetic realisation of /k^hēp^hd⁰e/ as 〈cēpte〉. They were trained in a certain tradition (ultimately based on Latin) to record such changes in pronunciation in a manner reflecting the processes of OE phonology mapped onto OE phonetics. As a matter of fact, we will argue that having two fortis consonants one after the other (as in cēpte) is impossible in OE. This is a similar stance taken for MoE by Szigetvári (2020), who argues that there can be no two immediately adjacent fortis consonants in a morpheme (what's more, this feature does not necessarily spread, as in Leipzig [pz], see below).¹⁷ Following his line of argumentation and the notation adopted here, description [pʃ] is not /p^hʃ^h/ phonologically (or /pʃ^h/ sharing the same fortis feature), but /b⁰ʃ^h/, with a lenis /b⁰/, disclosed in describe), showing that (despite spelling) 〈p〉 is /b⁰/ with phonetic cues that disclose its phonological characteristics as a lenis consonant. We have to conclude that description is a Latinate way of representing what could more appropriately be spelt describtion in an aspirating language like MoE. Along these lines, we might suppose that OE scribes were in two minds trying to spell [kēpte] as either 〈cēpde〉 or 〈cēpte〉, with the latter spelling bolstered by a Latinate continental and Irish spelling tradition, ultimately gaining the upper hand.¹⁸ Either way, the phonological reality was /kēp^hd⁰e/.

Szigetvári (2020) argues, pace Harris (1994, 137), that we cannot see the effects of spreading of fortisness in MoE past tense, third person singular (and similar formations) involving the /d⁰/ (D for Szigetvári) and /z⁰/ (Z for Szigetvári) suffixes (in wiped ‘Pt’, wipes ‘3SgPres’, wipes ‘Pl (of the noun wipe)’) because this spread is not categorically found morpheme-internally (as in Leipzig [pz]/(**)[ps] /p^hz⁰/), but it is across strong morpheme boundaries (#wipe#s#), and even across word boundaries (Depp's a pirate from Depp is), making this a very odd assimilation that does not operate morpheme-internally. A phonetically voiceless obstruent in an aspirating language, he concludes, is not necessarily fortis. Accordingly, there is no reason to suppose the fricatives have different laryngeal specifications in Leipzig and wipes, they are both /z⁰/ (/lajpz⁰ig/, /wajpz⁰/). The one in wipes undergoes phonetic devoicing to [s] morpheme-finally (#wipe#s#). Similarly, wiped would be /wajp^hd⁰/ [wajpt] for our purposes in MoE. OE cēpte and mētte are similar: /kēp^hd⁰e/, /mēt^hd⁰e/. Find a summary of what we have found so far in (18).

(18)

The system of Old English singleton stops and fricatives

Note that the use of symbols to represent the laryngeal characteristics of these obstruents is somewhat unbalanced (using f⁰ instead of v⁰ to make it conform to b⁰) may seem at odds with the argumentation, but this is the choice taken here to make sure the underlying representation of these obstruents bears some visual correspondence to OE spelling. This is, of course, merely a ‘visual’ problem, with the tradition winning out in this case.¹⁹

6 Old English lenis and fortis geminates

Old English also had geminate voiced stops. Given that voice does not exist as a laryngeal feature of the obstruents, this translates into geminate lenis stops (as in habban ‘have’, gebedda ‘bedfellow’, frogga ‘frog’). There is nothing surprising about this given that these clusters were passively voiced between sonorant consonants/vowels. Phonologically, they can be characterised as shown in (19).

These clusters pose no special problems, they are not typologically marked in any way. Melodically they simply show laryngeally unmarked stops interpreted in two skeletal positions.

However, we do run into problems in the analysis of voiceless geminate fricatives (as in pyffan, cyssan). In (16d) we saw that the past tense suffix is realised as [t] after voiceless fricatives, suggesting that they were phonetically voiceless. Obviously, we cannot claim that voiceless geminate fricatives were composed of two laryngeally unmarked fricatives, i.e. /ff/ ≠ /f⁰f⁰/. If a geminate fricative was simply a sequence of two lenis fricatives, we would expect passive voicing to be able to spread across them, similarly to a geminate voiced stop (/b⁰b⁰/), i.e. we would want to see pyffan /pyf⁰f⁰an/ as [pyvvan], cyssan /kys⁰s⁰an/ as [kyzzan], etc. A sequence of these fricatives would imply a phonetically voiced past tense, too: **pyfde [pyvde]. This is clearly wrong.

Suppose now that phonetically voiceless geminate fricatives were impervious to passive voicing because they were composed of a fortis fricative followed by a lenis one: 〈pp〉 /p^hp⁰/. We have already seen that in a combination of a fortis obstruent followed by a lenis obstruent, passive voicing is barred from occurring throughout the cluster. This is exactly how we analysed past tense forms like cēpte, which is /p^hd⁰/ phonologically. We find the same constellation in a phonetically voiceless geminate fricative. In a cluster like /s^hs⁰/ (cyssan) it is impossible for the lenis to be anything else but voiceless. If another lenis is added (the past tense), it is also phonetically voiceless (cyssde /s^hs⁰d⁰/ [st]). We will also assume that OE, similarly to MoE (Szigetvári 2020), did not have doubly aspirated/marked consonants (**/p^hp^h/, **/s^hs^h/). We will suppose that such marked consonant clusters did not exist in the language (a heritage continued into MoE).

The supposition that OE had a fortis fricative is unusual in a language in which the usual claim is that fricatives are not distinctive, but the same argument can be levelled against all traditional approaches that simply assert that geminate fricatives were voiceless in environments in which voicing could otherwise be expected: in cyssan the fricatives (or at least the first one) is right-adjacent to a stressed vowel, so voicing could be expected.

The next question is whether we can gather further data for the claim that OE had fortis fricatives (albeit always as the first member of a cluster). What we see in forms like pyfte, cyste, etc. can be termed degemination, not only on account of the spelling (usually) having a single fricative and a 〈t〉, but also as genuine examples of degemination leaving behind a fortis fricative, which is now in contact with the lenis stop, making it phonetically voiceless. In pyfte, therefore, we have /f^hd⁰/ [ft]. One could say that all such instances show lexicalisation of the past tense forms (pyfte would in this case be like hæft ‘prison’). But lexicalisation happens haphazardly to individual lexical items, so we could expect there to be at least some past tense formations with geminate fricatives that show voicing (pyfte being found as pyfde), but this is never the case. This probably shows that pyfte is the result of an active phonological process, probably post-lexical degemination, the phonetic correlate of which is [ft]. Of course, we can also claim that the cluster /f^hf⁰d/ would also necessarily be interpreted as [ft] phonetically (the second, lenis half of the fricative and the lenis /d⁰/ are not couched between voiced sounds, and the loss of the second/lenis ‘f’ is simply a low-level phonetic regularity: [fft] > [ft]).

The other piece of evidence comes from Middle English where the OE geminates are continued as voiceless fricatives, establishing these as phonemic: cyssan > kiss, lǣssa > less, hlihhan > laugh (/f/ or altogether lost), etc. One of the interesting facts about degemination happening in Middle English is the absence of compensatory lengthening (words like kiss, bed, sit, etc. never show a long vowel). Following a suggestion made by de Chene & Anderson (1979, 2.32) we suggest that this degemination involved the loss of the second (lenis) member of the cluster, leaving a fortis fricative behind and no compensatory lengthening given the fact that such consonants were lost in unstressed syllables (or word-finally): cyssan /k^hys^hs⁰an/ > kissen /k^his^hən/, **/k^his^həən/. The same process of degemination affected the OE lenis geminate stops with no compensatory lengthening either, leaving behind a lenis stop: frogga > /f⁰roɡ⁰ɡ⁰a/ > froge /f⁰roɡ⁰ə/, later /f⁰roɡ⁰/.

It can be objected that melodically fully interpreted second members of such clusters were deleted in a phonologically strong position (cf. Scheer 2004): following a consonant, preceding a vowel (sitten /t^hd⁰V/). The solution, however, hinges on the representation of geminates. It can be argued that geminates like [tt], [ss] in OE were made up of a fully specified first member (/t^h/, /s^h/) followed by another consonantal position, which copied the melody of the preceding consonant (save the feature of ‘h’), making the second member of clusters like [tt], [ss] phonologically underspecified. ‘Lenis’ means the absence of a laryngeally specified ‘h’ feature on a consonantal position (/t^hd⁰/), see (20). The feature of aspiration links independently to the first C-slot.

(20)

A similar structure to the voiceless geminate fricatives could be seen for OE geminate fortis stops in (20b) (sittan /t^hd⁰/, which behave in the same way: their degemination in Middle English leaves behind a fortis stop with no compensatory lengthening (sittan /t^hd⁰/ > sitten /t^h/> sit /t^h/) another). Similar structures should behave similarly: absence of voicing affecting both the fortis geminate fricatives and stops in OE also supports their identical structure: sittan [tt], pyffan [ff]. There is no passive voicing affecting the phonetically voiceless geminates, just as there is no passive voicing affecting the singleton fortis stops: e.g. gerypon ‘reap, PtPl’ (**gerybon), sǣten ‘sit, PtPl’ (**sǣden), mēćes ‘sword, GSg’ (**mēcges), blacum ‘black, DatPl’ (**blagum). Put simply, fortis obstruents cannot be passively voiced in intersonorant position. The feature ‘h’, as we saw earlier, was not given up in weak positions, ensuring that these two classes are still distinguished in OE in non-initial positions, too.

We find justification for the proposed structure of the geminate fricatives and geminate voiceless stops in Iverson & Salmons' (1995) work. Their claim is that the phonetic implementation of the phonological feature [spread glottis] takes longer than one segment to allow the vocal folds to reconfigure for voicing, which explains why there is no aspiration after s+stop clusters (as in spin, stick, skill). It explains why [p] in spin, for example, can only be lenis (/b/) (despite what 〈p〉 suggests). If it was fortis, there would be aspiration after it (**[sp^hin]), hence spin can only be /s^hb⁰/.²⁰ The same structure is assumed for OE geminate fricatives and voiceless geminate stops (the phonetically bisegmental feature of [spread glottis] also explains what is usually assumed for OE h+sonorant clusters, namely that they had a phonetically devoiced sonorant in them, which was probably also true for s+sonorant clusters: e.g. hlinc ‘ridge’, hnæpp ‘bowl’, hrægl ‘dress’, hwæt ‘what’, slitt ‘fine’, etc.).

Middle English degemination is not the only process that deleted the second member of geminate consonant clusters. There is a similar process of deletion of post-consonantal pre-vocalic sonorants (see (21) below).

All the words in (21) contained an etymological /j/, /w/ or /h/.²³ It seems the most sonorous of the sonorant consonants were lost. Traditional sounding (non-reformed or reborrowed) pronunciations, as in Barraclough /bárəkləf/, Cholmondeley /tʃəmlij/, midwifery /mid(w)ifrij/, midriff /midrif/, chestnut /tʃesnət/, Christmas /krisməs/, ha'penny /hejpnij/, Peterborough /pijtəbrə/, etc., show that less sonorant consonants were not lost in this environment. To this list can be added the loss of underspecified lenis obstruents (21c), which would simply mean the deletion of a ‘C’ position with no compensatory lengthening. The loss of the second members of phonetically voiceless geminates led to the creation of new (non-initial) voiceless stops and, more importantly, a steppingstone for the future development of the laryngeal system of MoE, a class of non-initial phonetically voiceless singleton aspirated fricatives (offer, offing, sith/sithens (< siþþan ‘since’, from *sīþ þan ‘after that’), kiss, fish). If the obstruent was not just a consonantal place onto which melody was copied from the preceding obstruent, but was a melodically fully specified obstruent, deletion did not happen (21d): in waistcoat /k/ was not deleted because it is not the second half of a geminate.

Lastly, except for some of the original Germanic geminates, OE geminates originate in West Germanic gemination (Hogg 2011, §4.11), e.g. *kunjō > cynn ‘kin’, *framjan > fremman ‘perform’, *setjan > sittan ‘sit’, *wadjōjan > weddian ‘wed’, *hyɣjan > hycgan ‘think’. Whatever the steps were from West Germanic to OE, historically every second member of a geminate originates in a sonorant (and perhaps even the Common Germanic geminates have a similar origin). We give a summary of the obstruental system of OE in (22). Fortis geminate obstruents are classified based on the first member of the geminate.

(22)

The obstruents of Old English

7 Some of the (more) contentious issues of OE fricative voicing

7.3 Finding minimally different environments for voicing in the fricatives

If phonetically voiceless fricatives are expected in unstressed syllables in OE, one cannot expect to find an emerging opposition between voiced and voiceless fricatives in that environment, so we can conclude that the phonetically voiced vs voiceless fricatives are still in complementary distribution. There is no change in their phonological status. No environment can be found where they would be different, and this difference mirrored by a phonological development.

That fricatives were immediately voiced when they came to stand next to a stressed vowel (after the deletion of an unstressed vowel) seems plausible. One of the often-presented examples involves the suffix *-isō where /s⁰/ is expected to be phonetically voiceless, as in (23a). After the deletion of */i/, /s⁰/ came to stand (almost) next to a stressed vowel, as in clǣnsian. As discussed earlier, Luick postulated that /s⁰/ was [s] because the rule of fricative voicing did not apply in unstressed syllables and by the time it came to be next to (or inside?) a stressed syllable, the voicing rule had already stopped working, so we expect [klǣnsijan], rather than [klǣnzijan] (cf., however, Ringe & Taylor 2014, 263), and so forth for the rest of the words in (23a). The fact that we have cleanse with /z/ today has been presented as a problem (for Luick, but not Bammeserger, as discussed above, or Ringe & Taylor, who say that there is no evidence to be sure about the late OE pronunciation, which might have involved voicing of the fricative after syncope because these fricatives now followed a stressed syllable).

Words that always had lenis fricatives in a stressed syllable (23b) would never be analysed as having anything else but a phonetically voiced fricative (this conviction, as well as the interpretation of the data, is shared by all). Whether the fricatives in (23c) were voiced in unstressed syllables is also contested, as we have seen: Fulk agrees with Luick, as do Ringe & Taylor, that they were voiceless, but Pope (1981) would have them all voiced whenever they occurred in fully voiced environments. We have analysed them as phonetically voiceless because of the prosodically weak position in which they are found (phonologically they are still lenis).

No objections have been raised against voiced fricatives that came to stand immediately next to a stressed vowel after deletion of the unstressed vowel (23d). This reallocation from an unstressed to a stressed syllable indicates that the ‘voicing rule’ of OE was still active after the deletion of the vowel that separated the fricative from the stressed vowel, as seen in *tigiθōn, *hligisōn²⁷ > tīða [ð], hlīsa [z]. No phonological arguments can be found for the (supposed) difference between tīða and hliðum ‘slope, DatPl’ (in the latter example the /θ⁰/ was next to a stressed vowel since Germanic times).²⁸ The words in (23e) are interesting as they present a complex problem and will be tackled in Section 7.7.

(23)

Examples with fricatives in pre-Old English unstressed syllables

(a) nasals/lateral/rhotic+/s0/, /θ0/ in Old English

*klānisōjan > clǣnsian (with a long vowel, Ringe & Taylor 2014, 263), clænsian ‘cleanse’ (with a short vowel, Ringe & Taylor 2014, 283), *hlǣnisōjan > hlǣnsian ‘make lean’, *ferrisōjan > firsian ‘go beyond’, *grimmisōjan > grimsian ‘rage’, *diurisōjan > diersian ‘extol’, *hwinisōjan > hwinsian ‘whine’, *hālsōjan/hālisōjan > hālsian/hǣlsian ‘prognosticate’, *mǣriθō > mǣrþu/mǣrþ ‘fame’, *mānōθ/mānōθV- > mōnaþ/mōnþV- ‘month’, etc.

(b) /s0/, /θ0/, /x0/ in stressed syllables in Old English

*drūsōjan > drūsian ‘drowse’, *lausjan > līesan ‘set free’, *leusan > for-lēosan ‘lose’, *fraisōjan > frāsian ‘ask’, *niuhsjan > nēosan ‘seek out’, *hriuwisōjan > hrēowsian ‘regret’, *maiθma- > māðm ‘valuable object’, *aiθum > āþum ‘son-in-law’, *daugul > dēagul ‘hidden, secret’, *fleugan > flēogan ‘fly’, etc.

(c) /s0/, /θ0/ in unstressed syllables in Old English

*agisōjan > egesian ‘terrify’, *tamisōjan > temesian/temsian ‘sift’, *aβisōjan > efesian/efsian ‘clip (hair)’, *agisō > egesa/egsa ‘terror’, *arβaiθæ > earfoþe ‘hard work, Dat’, etc.

(d) */s0/, /θ0/ in unstressed syllables in pre-Old English, later next to a stressed vowel in Old English

*hligisōn > hlīsa ‘fame’, *sigiþiz > sigþe/sīþe ‘scythe’, *agiþā > egeþe/egþe ‘rake, harrow’, *tigiθōjan > tīðian ‘tend’, *tigiθōn > tīða ‘receiver’, *teoɣoθōjan > teogoþian/ tēoþian ‘(give as) tithe’, etc.

(e) stop+/s0/ in Old English

*rīkisōjan > rīcsian ‘rule’, *gīdisōjan > gītsian ‘covet’, *blōdisōjan > bletsian/bledsian ‘bless’, etc.

Now if the problem does not reside in fricatives being (or eventually coming to stand) immediately next to a stressed vowel (as in sīþe), but it does seem to hinge on fricatives not being immediately next to a vowel (i.e. being separated from it by a nasal, as in clǣnsian), the shortcoming of the syllable-based accounts is obvious: the fricatives in words like clǣnsian are not inside a stressed syllable, we might argue. Of course, words like wulfas [wulvas] are still unaccounted for, where everyone would like to see a voiced fricative.

Discounting post-nasal fricatives for the time being, we have not been able to find a minimally contrastive environment for a stressed vowel+fricative sequence (i.e. sīþe = tīða). To continue this line of reasoning, we will not be able to find a minimally contrastive environment for nasal+fricative sequences either, but now for a completely different reason, as such sequences were lost: in OE, Old Saxon and Old Frisian, nasals were deleted before fricatives, followed by compensatory lengthening: *fimf > fīf ‘five’, *līnθī > līðe ‘lithe’, *munθ > mūð ‘mouth’, *gans > gōs ‘goose’, *sanθ > sōð ‘true’, etc. There were no nasal+voiced fricative sequences (in these the voiced fricatives are found as stops, as in bindan < IE *bhendh-). So strictly speaking, there is no comparable word to clǣnsian (words like *gans- having been lost earlier) to test the behaviour of fricatives that came to stand next to a nasal (and a stressed vowel) in comparatively recent times.

The environment for testing /lθ/ (as in hǣlð-) that arises by vowel loss, compared to historically older */lθ/ is also missing. In West Germanic, */lθ/ sequences were voiced and occluded in an environment identical to later OE (phonetic) voicing (*/lθV/): fealdan ‘fold’, wilde ‘wild’, wuldor ‘glory’, seld ‘dwelling place’, and feld ‘field’, beald ‘bold’, etc. with the stop extended from the oblique cases (Hogg 2011, §4.18). Again, having a sound historical comparison between hǣlðe (with a secondary /lθ/ sequence) and original /lθ/ is impossible. It seems there was no such occlusion in /rθ/ sequences in West Germanic or pre-OE, so this is the only environment where one could potentially find a difference in treatment between original /rθ/ (*werþai > weorþe ‘it may become’) and newly formed /rθ/ sequences via syncopation (mǣrþu ‘fame’). It may well be that this environment does not lead to fricative voicing (the fricative is not immediately next to a stressed vowel). Note also that West Germanic */rð/ was also eliminated (found as OE /rd/): *swerða- > sweord ‘sword’, *wurðun- > wurdon ‘became, Pl’, etc.²⁹

The */rz/ sequence was also eliminated before OE times (West Germanic */z/ > */r/): *irzijaz > ierre ‘mistaken, wrong’, *θurzi- > þyrre, ‘dry, dried up’, *warza > wearr ‘callous’, etc. It seems we do have original pre-OE /rs/ clusters surviving in OE, as in *wirsizo > wierse ‘worse’ (although here forms with */rz/ must also be posited to explain the superlative wierrest, cf. Ringe & Taylor 2014, 85), but we also have formations like *diurisōjan > diersian with secondary /rs/ clusters (see short summary in (24)).

(24)

nasal/r/l+voiced/voiceless fricatives in West Germanic and Old English

As can be seen in (24), OE has no ‘inherited’ voiced fricatives after a stressed vowel preceded by a nasal or rhotic (no [VnzV], [VrzV], [VrðV], [VlðV], [VlzV]). The only (restricted) environment where such a fricative could be expected was the /VrɣV_[back]/ or /VlɣV_[back]/ environment (e.g. burga, burgum ‘town, GPl, DPl’, also (probably) in forms where the back vowel was neutralised: burgen ‘guard, PPpl’ < *burɣan, folgian < *fulɣōjan ‘follow’, etc.).³⁰³¹³² The fact that /VrɣV_[back]/ or /VlɣV_[back]/ had [ɣ] is also supported by late-OE and ME developments and their MoE continuations like follow, borrow, sorrow.

As can be seen, there were no stressed VSonFric_[voiced]V clusters to form the basis of new voiced ones that had meanwhile arisen through syncopation. And perhaps original ones (as in horses ‘horse, DSg’, teorses ‘peg; penis, GSg’), too, failed to partake in the supposed voicing: the fricative in a cluster like /Vrs⁰V/ simply continued West Germanic phonetic voicelessness, i.e. [VrsV]. After all, the fricatives were not in a stressed syllable in OE (and those who adhere to a syllable-based account would also find it difficult to argue for syllabifications like hors.es). The fricatives were in a voiced environment, no doubt, but there were also sonorants blocking direct contact with the stressed vowel). It seems a historical ‘catalyst’ was needed for voicing to be extended to VSonFri_[voiceless]V clusters.

8 Towards a conclusion

We have tried to reanalyse OE along the lines of laryngeal oppositions rooted in lenis (neutral) vs fortis (marked): OE stops are divided into lenis /b⁰ d⁰ ɡ⁰/ and fortis /p^h t^h k^h/. Similarly to MoE, the lenis stops of OE must have been either phonetically voiceless or voiced depending on the environment. The phonological oppositions between the two classes were not neutralised in the weak positions (word-finally or intervocalically).

The West Germanic opposition in the fricatives collapsed for the coronal fricatives and coupled with the early OE changes (the debuccalisation of */x/ > */h/ and subsequent loss of */h/) a new system ultimately developed that had a neutral series of fricatives only (/f⁰ θ⁰ s⁰ x⁰ h⁰/). These could be phonetically voiced or voiceless depending on the environment.

One of the moot questions has always been the voicing of fricatives in unstressed syllables. Evidence points to these fricatives being voiceless in such a position. The syllable as a theoretical construct is of no use in the explanation of (phonetic) voicing. The question of whether there was voicing in sonorant+fricative sequences seems to have depended on whether OE had inherited such sequences of voiced fricatives after sonorants to act as a catalyst for further cases to be drawn in (found for the labial/velar fricatives in [lv/ɣ] and [rv/ɣ] clusters, but not the rest of the fricatives, which can only be explained as a historical coincidence).

Cases of devoicing (as traditionally understood) were analysed as examples of devoicing in clusters, found in the phonologically weak positions (word-finally, and in unstressed syllables). One of the interesting consequences of the analysis was that fortis fricatives also existed in the language, but were only found as the first member of geminate fricatives (/s^hs⁰/ 〈ss〉), s+stop clusters (/s^hb⁰/ 〈sp〉) and ft/ht-clusters (/f^hd⁰/, /x^hd⁰/). The answer to whether OE had phonologically distinctive fortis fricatives is yes, it did, but only in clusters (as possible remnants of an earlier opposition of pre-OE or West Germanic). It was only in Middle English that a system of oppositions resembling that of West Germanic was to develop again (continued into MoE), based on the proposed structure of OE geminate fricatives. By way of summary, see (33), which shows the OE obstruents with their phonetic rendition, the ‘rule’ for passive voicing of the fricatives (34) and a summary of the environments in which the obstruental stops were in contrast (35). We have to agree with Fulk (2001) and Minkova (2011), on slightly different grounds though, that OE had no demonstrable cases of contrastive fricatives during its 500-year history. If there is a rule, there can be no contrast.

(33)

Old English (fricative) voicing