1 Background
The list of scholars who have written on Old English fricative voicing is an honor roll of English historical phonologists. Few members of this admittedly small and shrinking group have never expressed an opinion on the topic in print. As someone without a parti-pris, either on the Celtic hypothesis in general, or on the dating and motivation of the establishment of an underlying contrast between /f/–/v/, /s/–/z/, /θ/–/ð/, I was intrigued to read Stephen Laker's contribution ‘An explanation for the early phonemicization of a voice contrast in English fricatives’ in ELL (Reference Laker2009), which is one of two phonological papers in the journal's special issue on re-evaluating the Celtic hypothesis. Predictably in this context, Laker argues that the phonemicization of the fricatives was a development induced by contact with Celtic.Footnote 2 This piece started as a brief commentary to that paper on the basis of some alliterative matching data. As usual, the topic turned out to be a densely tangled web of suppositions and inferences which seemed in need of some additional data-links and clarifications.
The distribution of voiced–voiceless fricatives in OE can be presented in an enticingly simple manner: the ‘unmarked’ value of the fricatives was voiceless, and voiced allophones occurred only between voiced sounds, either stem-internally, or across an inflectional boundary. (1) is a reminder of how the familiar textbook descriptions match sounds and spellings:Footnote 3
(1) OE voicing of fricatives:
Footnote 4 The title, ‘An explanation for the early phonemicization of a voice contrast. . .’ (Laker Reference Laker2009), presupposes that the early phonemicization of the voice contrast has been established with certainty and we need an explanation for its origin. For this traditionalist the reasons for concluding that early phonemicization is a well-substantiated hypothesis were unclear; this article is intended as a commentary and an invitation for further discussion. I am concerned both with the fair use and transmission of scholarly material, and with exploring further the empirical and theoretical grounding of specific proposals.
I start with a survey of the forms on which the hypothesis of OE phonemicization is based, followed by a discussion of alternative interpretations of the observed patterns and irregularities. The empirical core of the paper presents the patterns of alliteration in Old and Middle English; this kind of evidence has not been previously considered in evaluating the progress of the change. The analytical core of the paper is dedicated to the dynamics of categorization: edge vs domain-internal contrasts, the relative strength of the voicing environments, and the uneven rate of fricative voicing depending on place of articulation. The paper ends with an emphasis on the continuity of the process and further thoughts on the documented interplay of external vs internal factors in the evolution of the voicing contrast.
2 Some unexpected facts about fricative realization in OE
Laker (Reference Laker2009) attributes the assumption of early phonemicization of the voicing contrast in fricatives to Bammesberger (Reference Bammesberger, Kastovsky and Bauer1988) and Fulk (Reference Fulk2001, Reference Fulk2002). Upon re-reading Bammesberger, one finds the suggestion worded extremely cautiously. Indeed, all he claims with assurance, I believe rightly (see section 3 below), is that in OE ‘position can no longer be considered as the sole and ultimate factor governing their [voiced–voiceless] distribution’ (Reference Brunner1988: 122). In his conclusion (ibid. 125) he states: ‘Whether under the circumstances it is justifiable to really ascribe a voice contrast to the Old English system of spirants is a question I would not want to decide.’ The circumstances in question, fully examined and expanded upon in Fulk (Reference Fulk2001, Reference Fulk2002), revolve around the realization of fricatives in medial position, which includes attestations stem-internally, in inflected and derived forms, in quasi-compounds and in compounds. Bammesberger restricts his discussion to /s/–/z/, /θ/–/ð/; he states that medially ‘<f> is always a voiced spirant’ (Reference Krapp and Van Kirk Dobie1988: 122). Fulk treats the <f>/<u> occurrences on a par with the other fricatives in the set.
These important contributions draw attention to unexpected facts in OE that raise questions about the regularity of the correspondences in (1). Following up on Bammesberger's observations, Fulk's two-part study (Reference Fulk2001, Reference Fulk2002) is the most informative record we have of all scholarship on and all aspects of fricative voicing, including the velars.Footnote 5 Without going through all details, the points that lead to skepticism about the allophonic status of the voiced fricatives in OE are shown in an increasing order of relevance in (2)–(6). First, there is the crux of the historical -þu suffix for abstract feminine nouns and the -þa suffix for ordinals, as in PDE strength, health, fourth, seventh, ninth:
(2) The realization of the -þu, -þa suffixes:
Footnote 6 The fricative in the suffix is unvoiced at a morpheme boundary, for which two plausible explanations exist, cited in Laker (Reference Laker2009: 214): fricative voicing was blocked if the flanking vowels were unstressed, e.g. streng-þ(u) <*strangi-þu, originating with Luick (Reference Luick1914–40) and supported by Fulk (Reference Fulk2001), or -þu was exempt from voicing due to the open nature of the juncture between the base and the suffix – open junctures block the phonotactic processes (Dietz Reference Dietz1997). Another non-conforming item, OE hælþ(e) ‘health’, fits the open juncture model too, and as for OE monþe (nom. monaþ), the persistence of the voicelessness of the final fricative from the base into the inflected form should be no surprise.
The derivational suffixes -þu, -þa behave like other productive suffixes in OE: -sum, -ful, -fæst, so the lack of voicing is morphologically motivated. Similarly, Minkova (Reference Minkova2008: 28–9) found that all prefixes constitute an open juncture. Juncture type and its effect on voicing is complicated in suffixation by the fact that the stem-final fricatives to the left of the juncture do get voiced. Fulk (Reference Fulk2002: 95) shows that ‘in prehistoric times in quasi-compounds there occurred regular voicing of fricatives before deuterothemes beginning with a voiced sound’, though the evidence (wisdom, lively, gosling) is hard to date.Footnote 7 Neither Luick's scenario nor an account based on the preservation of affixal voicelessness in derivatives requires a hypothesis of early phonemicization.
Another possible challenge to the allophonic status of the fricatives is identified in some metathesized forms: ‘a phonemic distinction /s/ vs /z/ has also been argued for on the basis of forms such as cærse (<*cræsse) ‘watercress’, hyrse (<*hruss(i)j-) ‘mare’ and Horsa (<*hrossa = a personal name), which show simplification of -ss- following r-metathesis but, apparently, no subsequent voicing of s’ (Laker Reference Laker2009: 214).Footnote 8
(3) Metathesis and voicing:Footnote 9
Two considerations cast doubt on the relevance of these items as evidence for a (Celtic) contact-induced early phonemicization: metathesized forms are a special case; they are in free variation, which can block the normal rule application. Compare the history of OE <asc-, acs-> in ascian ‘ask’, producing both ME south-west and west Midl. <ash, esh> and northern <ask>.Footnote 10 Secondly, as Laker points out (Reference Laker2009: 220–1), /z/ cannot be reconstructed for the substratum either phonetically or phonologically. Hence a big question mark hangs over the validity of these three words as evidence for phonemicization. To this mini-set Laker adds Late West Saxon blosm ‘blossom’ < blostm, where <s> should have undergone voicing after <-t> was dropped. The data on this item are shown in (4):
(4) Lexical idiosyncrasy: blosm ‘blossom’ < blostm
blost* × 168, vs blosm* (× 34) (DOE)
<-stm>(x23) vs <-sm> (× 13) (LAEME)
The attestations of the original base <blostm> in the OE corpus, as a search of the The Dictionary of Old English (DOE) shows, are more frequent at a ratio of 5:1. The MED shows only 2 pre-1300 occurrences of <‑sm>, and a continuing preponderance of <stm> forms in LAEME's database (www.lel.ed.ac.uk/ihd/laeme1/laeme1.html) makes this item irrelevant.
Another lexically idiosyncratic item, cited by proponents of early phonemicization, is the adjective wyrsa ‘worse’. The adjectival form coexists with the adverb <wyrs, wiers>. The form with a final orthographic <-a/o> is a minority spelling in the DOE corpus:
(5) Lexical idiosyncrasy: OE wyrsa ‘worse’
<wyrse, wierse, wirse, wærse, wirse, wuyrse, wyrsse> (× 136)
<wyrsa, wirsa, wærsa> (× 29)
<wyrso, wuyrso> (× 2)
<wyrs, wirs, wiers> (× 48)
The count includes only the citation forms, not inflected forms, but it is still indicative of three circumstances which would give us at least a clue to the blocking of voicing in LOE and later. First, there is the strong presence in the ambient language of the vowel-less adverbial form (× 48).Footnote 11 Second, it is probable that the preservation of the final vowel in the adjectival ending may be at least partly orthographic: we expect spelling to be conservative; nevertheless the instability of the realization/the reduction of the etymological final vowel is indicated by the choice of <-e> (× 136) over <a/o> (× 31). Third, the superlative forms of the adjective <wyrrest(a), wyrst(a), wierresta, werresta, wersta>, occurring quite frequently in the DOE corpus (× 74), should also be taken as part of the overall picture. It would be hard to claim that the lexical storage of adverbs and superlatives will not affect the PDE outcome of worse.
The time-honored procedure for establishing phonemic status is the search for minimal pairs, or near-minimal pairs. None of the studies in question claims the existence of such minimal pairs for /f/–/v/, /s/–/z/, /θ/–/ð/, but Bammesberger (Reference Bammesberger, Kastovsky and Bauer1988) comes close to showing a minimal pair with the pertinent observation that the reflexes OE clænsian ‘cleanse’ (also perhaps mærsian ‘glorify’) and OE cursian ‘curse’ appear to challenge the basic principle of OE voicing; voicing occurs if the fricative is flanked by voiced segments, so curse should have [-rz].
(6) The minimal pair test:
(a) OE clænsian ‘cleanse’ vs OE cursian ‘curse’
(b) Like cursian (from Fulk Reference Fulk2001: 72 n. 5): burse ‘purse’, Perse ‘Persian’, persoc ‘peach’ (Latin), gærsum ‘jewel’ (ON)
Again, a look at the lexical relations of the two items makes the proposed pairing irrelevant. The -sian in clænsian ‘cleanse’ is a famous crux (see Fulk Reference Fulk2001 and the references therein); after covering a wide range of comparable data and earlier explanations, Fulk (Reference Fulk2001: 68) states that ‘Where the voicing in cleanse originated remains a mystery.’Footnote 12 It is, indeed, a one-off, since the sibilant is voiceless and shares the voicelessness with the preceding consonants in other derivatives of the weak-verb suffix – *sojan: blissian ‘bless’, blessian ‘sprinkle with blood’, miltsian ‘pity’; the <t> in glisten < OE glisian, glisnian appears c. 1225 and hypothetical OE *glimsian ‘glimpse’ is ME (c. 1400).
As shown in (6b), cursian belongs to a set of foreign borrowings: burse ‘purse’, Perse ‘Persian’, persoc ‘peach’, all from Latin; gærsum ‘jewel’ from Old Norse; they ‘clearly shed no light on the problem’ (Fulk Reference Fulk2001: 72 n. 5). I agree. Foreign borrowings can be exempt from the native phonotactic rules; the foreignness of the word in this case is the defining trigger of a paradigm uniformity effect which is missing from the native lexicon, where the surface alternations remain predictable, as in hūs [-s] ‘house’ n. vs hūsian [-z] ‘to house’. Moreover, the (possibly Irish) word curse (OE curs) is a frequent item in the OE lexicon; there is no corresponding base *clens. The higher density of the noun curs (× 44) vs the derived verb forms (× 9) in the DOE database provides a further argument that spread of the unvoiced realization in the underived form is the most likely cause of the apparent irregularity of [-rs] in curse v.
An interim summary: none of the cases discussed in (2)–(6) can be taken as unambiguous evidence of irreducible lexical contrast between voiced and voiceless fricatives at least until the end of OE.
3 Paradigm uniformity and allophonic distribution
In collecting and analyzing the information about fricative voicing in OE, Fulk (Reference Fulk2002: 87) states that ‘the fricative voicing rule . . . appears to have ceased to be operative on a phonological basis already in the prehistoric period’ (emphasis DM). I think that this is exactly right with one small addition: the voicing ceased to be operative solely on a phonological basis. In the narrower view, phonological factors operate independently of morphological factors, a division of labor for the components of the grammar that can be challenged. It is indeed a truism that despite the analytical usefulness of identifying levels within a linguistic system, functionally these levels are nested within each other in a way which allows for multiple interactions.
Paradigmatic analogy is a frequent recourse in the face of unexpected forms in the traditional accounts of many phonologically conditioned changes, and it has to be admitted that, under the rubric of ‘analogy’, it has been used as a convenient catch-all for all irregularities that don't fit the strong neo-grammarian assumption of the regularity of sound change. The question is whether some surface evidence of contrasting realizations in identical environments is sufficient to overthrow the analysis of allophonic distribution of the fricatives in OE. This question takes us to the most clearly stated structuralist argument against the allophonic status of the OE fricatives:
(7) ‘This is an axiom of phonemic theory: allophonic variation, being subphonemic, is produced unconsciously. Native speakers do not notice it, and therefore it cannot be subject to analogical extension and leveling. Analogical replacement of one allophone by another contradicts the very concept of the phoneme, which dictates that allophones are in complementary distribution.’ (Fulk Reference Fulk2001: 60–1)
Do we abandon the generalizations in (1)? They do not appear accidental, yet that's what the theoretical premise of (7) dictates. Without launching a discussion of phonemic theory and its incorporation into alternative linguistic models – my interest is in clarifying the record on the allophonic vs contrastive distribution of fricatives in the history of English – what I want to defend and illustrate is that the results of allophonic variation can be obscured by analogical extension. Two examples from analyses of PDE data should make us aware of the interaction of allophony and paradigm uniformity. The first is a widely cited example from PDE truncation from Kahn (Reference Kahn1976)Footnote 13 which shows the realization of tautosyllabic [ær] in derived/truncated names in rhotic New York English, a dialect which otherwise mandates the realization of the low back allophone [ɑ] in this environment:
(8) Allophonic [æ]/[ɑ] in NY English
Alternations of this type, which served as the basis for Correspondence Theory (Benua Reference Benua and McCarthy1997/2004), show that maintaining identity with the features of the base, a faithfulness constraint, can dominate the requirement for complementarity of the [æ]/[ɑ] allophones in English. The new forms [lær], [sær], which maximize identity with the base, are marked because they violate the expected phonetic adjustments; this is a case of ‘underapplication’ of the allophonic distribution of the low vowel before [-r]. The theoretical premise here is that ‘The base of an O(utput)O(utput)-correspondence relation can show non-canonical phonology if it is itself morphologically complex, and its deviant features increase identity with its output base’ (Benua Reference Benua and McCarthy1997/2004: 435, n. 31).Footnote 14
Another instance of unpredictable allophonic realization is provided by the distribution of ‘light’ [l] vs ‘dark’ [ɫ] in American English. Although the realization of /l/ is an important dialect criterion, e.g. Welsh and Irish English have clear [l] in all environments, there is no variety of English where [l] and [ɫ] are separate phonemes. For American English the unambiguous allophonic contexts, here cited from Hayes (Reference Hayes, Dekkers, van der Leeuw and van de Weijer2000: 94–5), are shown in (9):
(9) Light [l] vs dark [ɫ] in American English:
In addition to these clear allophonic choices, there are environments in which the phonological conditioning is obscured and the results are scalar. As Hayes’ (Reference Hayes, Dekkers, van der Leeuw and van de Weijer2000) study of acceptability judgements shows, there is essentially free variation between [l]~[ɫ] in Mailer, Hayley, Greeley, Daley, while Paradigm Uniformity effects are considerable in cases such as (touchy)-feely, mailer, mail it, all allowing [ɫ]; the effect is stronger in the clitic phrase than in the derived forms -feely, mailer.
Whether one treats the problems in (8) and (9) in the parallel framework of competing constraints without stratification, or in the cyclic framework of Lexical Phonology, with an intermediate level between the underlying representation and the final output, or in Stratal OT, the basic point remains: the allophonic alternations on the surface interact with and can be overridden by non-phonological factors.Footnote 15 This morphologically driven opacity of the phonological rules can be projected backwards, making our decisions about phonemic vs allophonic status in a dead language more problematic. The [-ær] ~[-ɑr] and [l]~[ɫ] complementarity in PDE parallels the complementarity of the voiced ~ voiceless fricatives in OE.
4 The evidence from alliteration
This section addresses one aspect of the empirical record on phonemicization that to the best of my knowledge has not been used in any studies of the process: the use of the fricatives in alliteration.
4.1 Alliteration on <f, þ, ð, s> in OE
Starting from the axiom that within a tradition of oral composition and pre-silent reading alliteration is based on phonemic distinctions, we can look at how the OE scops matched phonemically identical onsets. Since at least Roman Jakobson's statement that ‘The phonemic equivalence of rhyming words is compulsory’ (Reference Jakobson1966: 426–7), reference to rhyme and alliteration for establishing underlying phonemic status has been the cornerstone of studies of phonological identity and contrast.Footnote 16 Indeed, the generality of the principle makes irrelevant the issue of how ‘literate’ a particular verse tradition is; the fact that scribes and scops followed a stylized template has no effect on the systematic patterns of identity found in the verse. Arguably, OE alliteration provides the best documentation of the evolution of palatalization of the voiceless velar stops in OE, where in spite of the rarefied nature of the genre the practice in the late verse differs from the earlier verse. The alliterative matching maps out the categorical identity of the velars where there is no corresponding scribal evidence, as shown in Minkova (Reference Minkova2003); other instances where alliteration can be used to recover the phonological history of English segments are also addressed there. In spite of the potential for confusion, e.g. between [f] [θ], and [s], or between [d] and [t], there are no such alliterative mismatches in OE. Therefore the alliterative evidence for the fricatives should also be explored.
The interdentals [θ] ~ [ð] provide the richest information on alliteration. As is well known, the initial voicing in the relevant function and pronominal words – thou, that, these, then, than, there, thus etc. – must have been under way at least since the beginning of the fourteenth century. It is attested by rhymes in Chaucer (see also section 5.2 below), but its dating in the north-east Midlands and the north remains unclear. Since this is a transparent case of prosodically induced change, one could speculate that these words were initially voiced already in Old English.Footnote 17 However, what we find in the poetic corpus points to stable initial voicelessness in these words too:
(10) OE alliteration of initial dental fricatives:
As shown in (10), the pronominal words are matched alliteratively to fully stressed words in [θ-]; this would obviate any case for initial voicing. It could be objected that if these words were used in alliterating metrical positions, they would be considered prosodically strong and therefore the conditions for voicing would not be present, but this would be a spurious objection. Recourse to prosodically weak syllables/words in metrically strong positions in satisfaction of the parametrical rules of alliteration and rhyme is a common metrical device; it does not tell us more about the prosodic prominence of ðys(ne) or þu in the speech of the OE scop than Chaucer's rhymes such as felawe: awe, biddyng: thing tell us about the stress of felawe or biddyng in his speech. The metrical manipulation stops at the prosodic level, however; phonemic mismatches in OE verse are virtually unknown and need special pleading (Minkova Reference Minkova2003: 78–87).
If one accepts the likelihood suggested by Bennett (Reference Bennett1955) and Lass (Reference Lass1991–3: 41) that /θ-/ was categorically voiced in initial position in Old English (West Saxon, Kentish, or West Mercian), then the alliterative matching in (10) could be on [ð-]. That scenario entails a later reversal of the initial voicing in major class words, presumably under dialectal influence from the northern areas where the voicing did not occur. This is not impossible, but it does raise the question of orthographic differentiation: why did it lag so far behind?Footnote 18 In either case, however, the type of usage shown in (10) remains evidence against the phonemicization of voicing in the dental fricatives in OE.
For the labiodentals a check of the behavior of Latin borrowings in OE verse would be the only test. The very early loans, prior to the Anglo-Saxon settlement, predictably reflect the approximant pronunciation [w], so wall < Lat. vallum, win < Lat. vinum etc. Unfortunately, there are few later loans that can be tested in the poetic corpus: fann ‘fan’ < Lat. vannu, fiddle < Lat. vidula are not in the poetic records, and the single attestation in verse of fers ‘verse’ < Lat. versus is in a non-alliterating second lift in the on-verse in JDay II: 11.Footnote 19 Some support for Campbell's (Reference Campbell1959: §539) non-verse-based assertion that the spelling with <f> in such items is ‘in accordance with the OE sound-system’ is found in (11):
(11) OE alliteration of initial labiodental fricatives:
Footnote 20 In this case, as in (10) above, even if we allow that the alliteration of Firgilius < Lat. Virgilius: freond is on [v-], it is still uninformative about phonemicization.
Although elsewhere <z> could be used for [ts], the only <z-> initial items in the pre-Conquest loanword sets in the ASPR are names: Zeb, Zebee, Zebed(es), Zefferus. They do show seemingly last-resort alliteration on <s->:
(12) OE alliteration of initial sibilants:
Footnote 21Footnote 22Footnote 23 It would be odd to try and push an argument parallel to the dentals and labiodentals, namely that these alliterations were on [z-]; assuming that the <s> may have represented [z-] forces us into allowing a phonotactically unlikely onset cluster [zw-].Footnote 24 Moreover, the lexical oddity of <z-> initial names places this set outside the range of reliable evidence.
In view of the chronological and dialect diversity of the alliterative matching in (10)–(12), the assumption of adaptation of the loanword phonology to a native pattern of stem-initial voiceless fricatives is the option requiring the fewest additional untestable suppositions. This would be identical to Campbell's (Reference Campbell1959: §539) conclusion regarding OE spellings such as <Firgilius, Fulcania>, i.e. they represent [f-].
Can we assume ‘impure’ alliteration in (10)–(12)? This is really untestable in the case of <z->: <s->. For the dentals and the labiovelars the strongest defense of the axiom that identity of the voice feature is not violated in OE alliteration is the editorial consensus on the issue: voiced and voiceless stops do not alliterate with each other anywhere else in the corpus. Unlike missing alliteration, and lexical, semantic and syntactic editorial imperfections, where the emendations can be controversial, identity of voicing for the stops is always assumed by the editors. This takes us back to the oral nature of the composition of OE verse and the belief that the scops, scribes and copyists shared the same understanding of the phonological categories provided by the ambient language. The potential of lenited allophones in initial position cannot be excluded, but it is purely speculative, while their categorization as voiceless is substantiated by the verse data.
4.2 Alliteration on <f-, v-, th-, s-, z-> in ME
Turning to ME for comparison, we find that the fourteenth-century alliterative corpus does show occasional alliteration of <f> and <v>. The two texts in which the labial fricatives have been claimed to alliterate with some frequency are Piers Plowman (PP) and Richard the Redeless (RR).Footnote 25
(13) Middle English alliteration on <f>:<v>:
Such attestations are of interest only in so far as the evidence of <f->:<v-> alliteration is confined to a particular manuscript and dialect, outside of which the voiced–voiceless contrast is unattested. The intended matching is most likely on [v-].Footnote 26 Crucially, initial voicing in the south cannot be interpreted as triggering phonemicization; it is a separate development, as has been often asserted on different grounds (see Nielsen Reference Nielsen, Laing and Williamson1994; Laker Reference Laker2009: 216, and the references therein).
I am not aware of unambiguous alliteration of a dental fricative-initial function word with a major-class word in the ME corpus. The examples I have been able to identify are all lines defective in some (other) way:
(14) Alliteration of function words in <th-/þ->:
Footnote 27 In the first example from The Alliterative Morte Arthure (AllitMA), the alliteration could not be on [d]:[ð] (Daye:they); such matching is not found in the corpus. In the second example þow, thy have no mate in the b-verse. The other example is most likely a self-alliterating b-verse and therefore uninformative.Footnote 28
The absence of alliteration in this case is significant because it occurs within a verse tradition which is much more forgiving in terms of metrical promotions, and which does allow other function words – prepositions, auxiliaries – to alliterate with fully stressed words (see Minkova Reference Minkova2003: 58–9). The absence of alliterative pairing in the subset of function words with initial dental fricatives suggests strongly that it is the voicing in this, thy, those, thou, etc. that blocks the pairing with [θ-] words in these compositions.Footnote 29 The contrastiveness of the dentals in initial position, coupled with the loss of final -e, would be sufficient for the reanalysis of the earlier allophonic distribution: adding alliterative evidence to the other evidence used so far allows us to state with greater confidence that the phonemicization of the dentals must have been accomplished by the first half of the fourteenth century, which is roughly the beginning of the period known as the ME alliterative revival. Regionally, the texts examined are quite diverse, but they exclude Kent where the ME voicing is best attested.
Finally, the evidence for <s>: <z> alliteration is as ambiguous in ME as it is in OE:
(15) Middle English alliteration on <s>:<z>:
Footnote 30 Again, the lexical idiosyncrasy of the borrowed names makes judgements on the realization of the consonants in such lines unreliable.
5 The dynamics of categorization: the gradience of lexicalization
Opinions on when the voicing contrast of fricatives was reanalyzed as underlying vary, yet there is consensus that context-sensitive voicing of /f, s, θ/ (or final and pre-consonantal de-voicing of /v, z, ð/; see Bennett Reference Bennett1955, Lass Reference Lass1991–3) is an important aspect of the phonology of OE. Jumping 1,500 years ahead, /f/-/v/, /s/-/z/, /θ/-/ð/ are separate phonemes in all dialects of PDE, though some traces of the old alternations survive. Fricative voicing in PDE is a textbook case of a rule whose productivity has decreased with time (Hayes Reference Hayes2009: 193–5, 198–9, 231–2); it has been restructured from a phonologically conditioned alternation to a rule recognizable only as /f/-Voicing in Plurals, and possibly also as /f/-Voicing in Verbs. Hayes’ arguments in favor of the existence of the rule of /f/-Voicing in Plurals are new forms: dwarfs~dwarves, acceptability of alternants such as gulf~gulves, chief~chieves, some environmental restrictions on voicing,Footnote 31 and a reported productivity experiment (wug test), where the plural of [hif] is [hifs] 58 vs 42 percent of [hivz]. How exactly the regular allomorphic alternation in OE, as in (1) above, became narrowed down to an alternation productive only for the labial fricative in specific morphological environments is a monograph-length study; the goal of this section is only to identify some additional factors that may have been ignored in earlier descriptions of the process.Footnote 32
Bermúdez-Otero (Reference Bermúdez-Otero and de Lacy2007) reminds us that the implementation of phonological change is never completely abrupt and that gradualness can be observed in one or more of the following dimensions: sociolinguistic, phonetic and lexical.
Restructuring does not happen in a vacuum: everyone who has looked at the history of the fricative voicing contrast in English recognizes the importance of Romance loan vocabulary in the process of phonemicization.Footnote 33 That any language contact can help build up a contrast which is not fully established is beyond question, but identifying the external source(s) instrumental in the history of the fricatives needs empirical support. As noted in section 1, Laker (Reference Laker2009) is an appeal for Celtic influence in the categorization of the fricatives as underlyingly distinctive. This could be a valid argument in principle, but it fails in view of several considerations: recall that Celtic did not have the sibilant voicing contrast;Footnote 34 there is no testable body of OE Celtic loanwords in which the fricatives are used by first-language speakers contrastively; the earliest possible firm indications of non-complementary distribution of the fricatives are post-Conquest and appear to be prosodically conditioned (see section 5.3 below). There is also a problem regarding the dialectal distribution of the earliest instances of the voicing of initial fricatives: while a generalized ‘south-west’ would fit the expectations of Celtic contact-induced transfer of contrastiveness, the concentration of initial voicing throughout Kent and the south-east (Lass Reference Lass1991–3) remains a problem for the Celtic hypothesis. We will return to the empirical basis of the contact dimension of the phonemicization in section 6.
Let us now turn to the phonetic dimension of the change. For obvious reasons, the reconstruction of the status of the OE fricatives has been based on orthographic evidence, historical comparison, place names, evidence from language contact and system-internal structuralist arguments. Some finer phonetic aspects of the process are bundled together: where does phonemicization start? Are all ‘voiced’ environments equally transparent surface triggers of allophony? Are all fricatives equally prone to phonemicization?
Addressing the central phonological concern of ‘contrasts’, Goldsmith (Reference Goldsmith and Goldsmith1995: 9–13) proposes a gradient schema of contrastiveness:
(16) The cline of contrastiveness (Goldsmith Reference Goldsmith and Goldsmith1995: 12):
1. Contrastive segments
2. Modest asymmetry case
3. Not-yet integrated semi-contrastsFootnote 35
4. Just barely contrastive
5. Allophones in complementary distribution
In that schema, in ‘pure’ allophony, the case of (16/5), the specification of a particular feature is completely controlled by the principles of the grammar, but cases (16/4)–(16/2) implement these principles less and less forcefully, until in case (16/1) there is no pressure, synchronic or diachronic, that controls the distribution of the feature. Goldsmith does make the point that if one insists on a binary distinction, only the case in (16/5), ‘allophones in complementary distribution’, would be qualitatively different from the remaining four cases, which present various degrees of ‘lexical contrasts’. Note, however, that the ‘just barely contrastive’ case in (16/4), illustrated with the famous contrast of can [-æ:-], noun, vs can [-æ-], modal verb, in many forms of American English, is ‘just barely’ lexical, i.e. it has the potential to generate a phonemic split and change the complementary distribution of the ‘tense’ and the ‘lax’ realization of the vowel in these dialects.Footnote 36 The important point about the scale in (16) is that the model allows multiple stages of categorization during which the phonemic status of a segment may be indeterminate; this transitional status could be described as quasi-phonemic.Footnote 37
Relegated to a footnote in Laker's paper (Reference Laker2009: 215 n. 7) are two important observations: ‘it is difficult to determine when phonemicization takes place’ and ‘some phonologists prefer to view contrastiveness as a scalar property’. These are related questions: if we approach the process with the assumption that contrastiveness is gradient, the relevance of the dating dissipates. We can start with the attestations and the probabilities of imperfect allophonic distribution of the fricatives from the earliest, and even prehistoric, Old English.Footnote 38 At the other chronological end, the residual /f/-Voicing in Plurals (dwarf–dwarves) suggests that we cannot analyze the /f/–/v/ contrast in PDE as completely context-insensitive. With this in mind, we turn to the more detailed analytic paths and competing phonological and lexical factors of OE fricative voicing.
5.1 OE fricatives at word-edges
It is a phonological truism that segments are realized differently at different positions in the phonological domain. The phonemic status of a segment does not require that this segment appear across the board; recall the case with PDE /h/, /ŋ/ in General American, or /ƿ/ in Scottish English – their distribution is restricted, but their contrastiveness is beyond question. Section 2 addressed the possible contrastive attestations of the OE fricatives in word-medial position and every single case allowed an alternative interpretation, compatible with allophonic distribution. This section turns to the attestations and arguments relating to the realization of the fricatives in word-final and word-initial position.
In the scholarship on OE fricative voicing, opinions converge on the word-final realization of /f, s, θ/; the consensus is that the reconstruction of word-final voiced fricatives has to be dated in relation to the /-ə/-loss in ME, as in drive, nose, bathe. Footnote 39 The loss of the conditioning environment occurs at different rates in the dialects and it is also word-class specific; this is relevant in our context only in so far as the difference between verbs and nouns continues to be productive in ME on the basis of surviving vowel-initial verbal inflections. Most importantly for this account, preservation of the voiced /v, z, ð/ at the right edge of the prosodic word has to be dependent on an inventory of phonemes which includes /v, z, ð/ already, otherwise the phonetic naturalness of word-final devoicing would interfere and possibly reverse the glottal state.
The exact phonetic nature of word-final devoicing of fricatives is related both to the manner of articulation and to their timing at the right edge of the word. Fricative production favors voicelessness more than the production of the corresponding stops (see Lindblom & Maddieson Reference Lindblom, Maddieson, Hyman and Li1988: 67; Ohala Reference Ohala1997). The increased duration in word-final position also correlates with voicelessness (see Lavoie Reference Lavoie and Minkova2009). Somewhat redundantly, we can conclude that the realization of word-final fricatives can be taken out of the discussion of the triggers and even the early stages of implementation of the phonemicization of the voicing contrast.Footnote 40 The qualification ‘word-final’ in this context refers to bare stems or inflected forms ending in <-e>; the special prosodic status of inflectional final <-s> which is the coda of a weak syllable is a separate case – it is instrumental in establishing the contrast for the spirants.
Now we turn to the left edge of the prosodic word, corresponding to word- and root-initial realization of the fricatives. Once again we can appeal to contemporary phonetic studies as possible evidence for historical tendencies. Phonetically, the consonants in stressed syllable onsets are produced with higher intensity (Fougeron & Keating Reference Fougeron and Keating1997; Lavoie Reference Lavoie and Minkova2009).Footnote 41 In her comparison of the intensity of the American consonants of each manner of articulation, Lavoie (Reference Lavoie and Minkova2009: table 2.2) reports an intensity ratio of 1:4 for voiced vs voiceless fricatives. For a Germanic-type language with rightmost root stress, this could be interpreted as stacking the production-perception deck in favor of /f, s, θ/ initially. This phonetic bias does not preclude initial voicing, but the recognition of this factor shifts the focus of causation away from phonetic and system-internal to other factors. As argued in Bennett (Reference Bennett1955), reported in Fisiak (Reference Fisiak1985), and more fully documented and developed in Lass (Reference Lass1991–3; see also the references therein), there may be a case for positing initial voicing in Kentish, West Saxon, and West Mercian through early (fifth-century) Continental migrations and contact.Footnote 42
Spelling is a major obstacle to reconstructing initial voiced fricatives even allophonically, prior to the Conquest. The most celebrated (and isolated) example of orthographically marked voicing to [v-] in OE is uif ‘five’.Footnote 43 Referring to the interpretations of occasional spellings, Bennett (1995: 368) cautions: ‘But even at best, the form uif could show only that initial [f] had been voiced at some time before the year 950, and the form fers could show only that initial f might represent [v] in West Saxon Old English. Neither form reveals when, where, or under what circumstances the voicing took place.’ Lass (Reference Lass1991–3) mounts a much more sophisticated argument about the vagaries of orthography and the geographical continuity of initial voicing in ME, proposing a revision of the conventional account for West Saxon, Kentish and West Mercian in which initial [θ-] is categorically voiced and voicing is ‘the majority allophonic control on /f, s/’ (1991–3: 41).Footnote 44 This hypothesis does not preclude lexical variation or sociolinguistically marked pronunciations. It may be compatible, especially with respect to initial /v-/, with a substratum reinforcement from what Schrijver (Reference Schrijver2009) identifies as Late British Latin in the Lowland Zone, itself possibly influenced by Celtic, between the fifth and the seventh centuries.Footnote 45 Note that this account is still explicitly ‘allophonic’; there are no claims of contrastive distribution, only the density of voiceless vs voiced realizations in different positions is more evenly balanced in the south and the south-west than in the east and the north.Footnote 46 It is therefore not possible to argue that the realization of the fricatives at word-edges in any dialect of Old English was a factor in the initiation or the stabilization of the voicing contrast.
5.2 Strength of voicing environment
If the burden of phonemicization falls on the realization of /f, s, θ/ in word-medial position, the phonological and prosodic contexts are of primary importance. In the voluminous literature on OE fricatives the segmental environment for voiceless realizations word-internally is well defined: fricatives are voiceless in clusters with other (voiceless) obstruents, as in æfter ‘after’, mynster ‘monastery’. Geminates, which result from the gemination of the voiceless allophone or are borrowings, are a subset of the more general condition of voicelessness of stem-internal obstruent clusters, as in offrian ‘offer’, missan ‘to miss’, oððe ‘since’. This also makes phonetic sense: producing geminate voiced fricatives requires that the precise constriction of the muscles be held for some time, which makes them even more effortful than geminate voiced stops (see Kirchner Reference Kirchner1998).Footnote 47
The segmental environment for voicing is defined less precisely. There is no disagreement that voicing, a transparently assimilatory process, occurs in intervocalic position, thus ofer ‘over’, broðor ‘brother’, nosu ‘nose’. The vowel to the right may be inflectional; inflectional/close junctures do not block voicing, thus wifa ‘of wives’, aðas ‘oaths’, husas ‘houses’. One possible question on the strength of the assimilatory process comes from the behavior of fricatives when they abut other voiced consonants. Two options are given in (17):
(17) Conditions for fricative voicing:
In (17a), stated in prose in Campbell (Reference Campbell1959: §50), here from Hogg (Reference Hogg1992: 282), any voiced segment within the (inflected) word, including voiced stops, will induce voicing, so hæfde ‘had’, ræsde ‘rushed’ with [v] and [z] respectively. In these weak dental past tense forms paradigm uniformity will be sufficient to account for the voicing.Footnote 48 Lass's formulation in (17b) excludes consonants from the environment to the left, possibly an oversight, and obstruents from the environment to the right. The latter would be correct for underived words, but it is in need of further detailing since it is an a priori proposition that the degree of coarticulation responsible for the voicing of the fricatives will be stronger for /j, w/ than for /r, l, m, n/, and there could be finer distinctions among the latter with potential consequences for our interpretation of their phonetic properties historically.
The main point here, however, is that all of the various examples challenging the allophonic status of the fricatives discussed in section 2 involve liquids or nasals, and not a straightforward intervocalic environment. In a framework of differently ranked constraints on voicing, therefore, the presence of a liquid or nasal may be violable, competing with morphological considerations, while intervocalic voicing within the prosodic word is inviolable. The conditions for voicing are gradually loosened in accord with the sonority of the trigger: the strongest voicing constraint is Inter-V-Voice (‘Intervocalic consonants are voiced’, Kager Reference Kager1999: 325), followed by [+sonorant, –syllabic] – [+sonorant, –syllabic] and lastly, [–sonorant, + voice] – the latter on either but not both sides of the input fricative.Footnote 49 The hierarchy is as in (18):
(18) Hierarchy of fricative voicing environment in OE:
Any sonorant >> liquids and nasals >> voiced obstruents
Another detail in the search for triggers of voicing refers to the location of the underlying segment in relation to the prosodic domains. Voicelessness is associated with stretching the cartilages at the bottom of the vocal cords. Lack of stress on a syllable or a word can correlate with no stretching, as is the case with PDE of with [-v], or his, was with [-z], which must have started in connected speech.Footnote 50
It has been observed, e.g. Anderson (Reference Anderson, Anderson and Macleod1988: 92), Lass (1992: 41), that OE fricative voicing occurs only in foot-medial position. A further refinement is that intersonorant voicing of fricatives occurred in OE only in the onset of the weak syllable in a trochaic foot (Minkova Reference Minkova2008: 29); the latter formulation avoids the problem of defining the foot-boundaries in words with secondary stress such as únsìbb ‘enmity’, únþèawas ‘vices’, míswèorc ‘evil deed’. The matching of voicing to the prosodic constituents in OE is shown in (19):
(19)
Note, however, that in forms with stacked prefixes such as únforcùþ ‘noble’, únforh 
fedness ‘incontinence’, fórþāùrnen ‘elapsed’, ófāslàgen ‘slain off’, the prosodic conditioning of voicing in únfor-, fórþā-, ófā- is blocked by the prefixal nature of the syllables in the foot – as we saw with the -þu suffixes, also -sum, -ful, -fæst, featural faithfulness of the fricatives in derivational affixes is ranked very high in OE.
Traditionally, as noted in section 4.1, the initial voicing in function and pronominal words: this, that, thy, those, thus, is dated to the late fourteenth century on the basis of Chaucerian rhymes soothe: to the, swithe: hy thee, though Jordan (Reference Jordan and Crook1934/1974: 185–7) lists studies of earlier ME manuscripts where the orthographic differentiation of voiced and voiceless fricatives seems to indicate earlier voicing of what he calls ‘the little words’. In principle, given the interchangeability of orthographic <þ> and <ð> in OE, one could declare the initial stage of this process unknowable and posit an early, even pre-c. 1000 initial voicing of [θ-]. As argued in section 4.1, however, the alliterative deployment of the pronominal words in OE gives no indication of initial voicing, so it is reasonable to stay with the textbook dating of the phonemicization of the interdental fricatives.
5.3 Were all fricatives created equal?
This tongue-in-cheeky question must obviously be answered in the negative. For a start, all scholarship on the English fricatives separates the history of the velar/glottal fricatives [ˠ] and [h] from the labio-dental, dental, and alveolar fricatives [f]-[v], [θ]-[ð], [s]-[z]. The question is whether we can refine the picture further: are place distinctions irrelevant? The modern distribution of /v-/, /z-/, /ð-/ is roughly the same (Wakelin Reference Wakelin1972: 90–2; Fisiak Reference Fisiak1985: 13; Lass Reference Lass1991–3: 21–2), which provides the basis for a unified ‘natural class’ account for all three.Footnote 51 Fully aware of the problematic nature of this assumption, Lass describes the various interpretative options as ‘a matter of analytical faith rather than being derivable from textual – or even fieldwork – evidence’ (Reference Lass1991–3: 22).
The issue may be undecidable, but we should try and get a better sense of the different steps needed for the establishment of contrast in each set. The prosodic specificity of the initial voicing of [θ-] > [ð-] makes parallelism with the labials and the sibilants unlikely – the seemingly obvious objection is that if the fricatives behaved always in a coherent manner, initial voicing in unaccented words (from, for, so, such) should have occurred with much greater consistency in ME outside Kentish, and the voicing should have remained as parallel to the voicing in this, there, though, etc. Indeed, surprisingly, in the ‘ongoing change’ described for the Cotton MS of The Owl and the Nightingale by Lass (Reference Lass1991–3: 19–21) we would expect for-, for and from to have had only <v/u> spellings, but this is not the case; all the <v>-only spellings are major-class words: fen, foe, fold, etc. The fact that the phonemicization of voice for the dental fricatives is accomplished almost entirely on the basis of native words also speaks against bundling them together with the other fricatives; no language-contact explanations are needed or even reasonable for the phonemic split of the dentals.Footnote 52
The sibilants are also a special case; acoustically, the contrast between the voiced and voiceless sibilants is masked by the high-amplitude frication noise (see Ladefoged & Maddieson Reference Ladefoged and Maddieson1996: 176); this may add to their confusability. Historically, the voicing of /s/ to /z/ in the south lags behind the voicing of /f/ to /v/ (Lass Reference Lass1991–3: 31). This would be in line with some dialectal and typological findings: initial voicing of /s/ in ME can be established only for Kent (Lass Reference Lass1991–3: 23). There is a parallel with Dutch too, where the phonemicization of the /s/:/z/ contrast ‘did not come into existence until the Modern Dutch period and most probably subsequent to the split between /f/ and /v/’ (Nielsen Reference Nielsen, Laing and Williamson1994: 24). Recall that there was no voiced alveolar /z/ in the Celtic substratum, which weakens a contact-based hypothesis of phonemicization. One should also consider the very different rate of occurrence for initial /v/ and /z/ in the loan vocabulary of early ME: there are only 5 <z>-initial entries in the MED before 1382, and a total of 31 infrequent items by 1500. This is minimal compared to the very robust attestation of /v-/-initial borrowed words (see figure 1).Footnote 53
/v/-initial vs /z/-initial loanwords in ME
Details on the /z-/ and /v-/ items borrowed in ME deserve further attention. The striking tenfold numerical difference in the set of loanwords in /v-/ is a strong confirmation of the hypothesis that the period c. 1300–1500 was crucial for the entrenchment of the voicing contrast for the labials. Of note in these data are also the types of loanwords recorded in ME: /v/-initial loanwords include a predictable ratio of nouns vs adjectives vs verbs, with verbs comprising about 10 percent of the totals, not counting another 6 percent of gerunds and participles.Footnote 54 There is even a /v/-initial interjection, ven ‘come’ (1475), while the /z/-initial words are almost exclusively nouns and adjectives. The lexical assimilation of the /v/-initial words is at an advanced stage, e.g. venim, n. ‘venom’ is part of a rich derivational set including etymological hybrids: venim (adj.), venimehēde (n.), venimen (v.), venimī (adj.), veniminge (ger.), venimlī (adv.), venimnesse (n.), venimoushēde (n.), venimouslī (adv.), venimousnes (n.). Many other /v/- initial roots show high productivity: vil(e)- ‘vile’ (30 headword entries), vari- (12), vers(e)- (13), ver(e)i- (11), victor- (10), while there is only a single participle in /z-/, zōdiaked, ppl. (1500). It is also the latest <z>- initial loanword form in the MED. The distribution is shown in figure 2.
Distribution of /v/-initial and /z/-initial words in the MED
The dialect distribution of the /v/-initial LAEME items is also of interest, though this question cannot be pursued here. The correlation between the density of borrowings in some areas and texts and the nature of the texts themselves warrants a separate study.Footnote 55 In any case, the loanword data are robust and testable and provide a very solid empirical foundation for the claim that the /f/-/v/ contrast initially is driven by the adoption of Romance vocabulary. The data in figures 1 and 2 confirm the soundness of Jespersen's hypothesis (Reference Jespersen1937) that loanwords were crucial in the phonemicization of the voiced labial fricative. The robustness of the exemplars of /v-/ also provides a basis for addressing the question (Laker Reference Laker2009: 222) of why other Germanic languages and dialects did not establish the voice contrast to such a degree and so early as in English.
6 When and why does the voicing contrast of fricatives become context-free?
This study reaffirms the canonical allophonic interpretation of the voicing contrast between [f]–[v], [s]–[z], [θ]–[ð] in Old English. Traces of this allophony persist and are still identifiable in PDE. The shift from a predominantly phonotactically governed to a predominantly lexically stored contrast has been occurring at different rates for a whole millennium. The factors involved in the shift are multiple and complex; moreover, in most cases the orthographic system denies us the confirmation of our hypotheses about OE pronunciation.
Virtually everyone today, irrespective of the particular model of language change they favor, will agree on the need for granularity in the discussion of contrast and categorization. User-based studies of storage and learnability of contrasts tell us that ‘Linguistic categories appear to contain a great deal of phonetic detail. . .Speakers store generalizations involving phonetic detail at multiple intersecting levels of categorization, for example word and segment. . . Speakers’ categories include a record of previously encountered detail, and this detail influences subsequent categorization and production’ (Blevins & Wedel Reference Blevins and Wedel2009: 161). This is compatible with the gradient schema of contrastiveness discussed in section 5. In either case we can identify several foci of phonemicization. In terms of inherent linguistic properties, the most salient factors are position in the onset of a stressed syllable and vowel-consonant-vowel coarticulation. As for the intersecting levels of categorization, our access to the OE system allows us to reconstruct only single underlying fricatives on the segmental level, with some additional categorizations emerging on the word level, i.e. the voicelessness of derivational affixes. The record of previously encountered detail is most clearly seen in the case of metathesis (cærse~cresse ‘cress’), lexical idiosyncrasy (blosm~blostm ‘blossom’), and paradigm uniformity (clænsian ‘cleanse’ vs cursian ‘curse’).
The structural dynamics of the change cannot be separated from the influence of borrowed vocabulary: these ‘new encounters’ are insignificant in our records of Old English, but the adoption of nearly 800 items with initial [v-] in ME confirms the most widely shared account of the phonemicization of /f/–/v/ as triggered by language contact. Already in LAEME, i.e. by 1325, we find many contrasting minimal pairs of the type vile, adj. ‘vile’: file (OE Anglian fil) ‘file’; veyn ‘vein’: fain (OE fæƷen).Footnote 56 In initial position the functional load of the contrast is highest. Reinforcement for the establishment of a contrast came also from word-medial intervocalic [-f], both through degemination, first in the north, and through borrowings such as coffin (1330, L cophinus (<Gr.) & OF cofin, coffin), sacrifice (1250) < F sacrifice, taffeta (1373) < OF taffetas, Med L taffata, sulphur (twelfth century) < AF sulf(e)re. Word-medial assimilations and obscuration of morphemic boundaries in compounds and place names also add to the unpredictability of the distribution of voiced–voiceless realizations.Footnote 57 The most recalcitrant aspect for the labial alternation is morphologically conditioned, both in noun–verb and in singular–plural alternations:
(20) Continuity of [f]:[v] alternations:
Footnote 58Footnote 59Footnote 60 The plural forms are also variable in kerchief (<AF. courchief c. 1300) –kerchiefs/kerchieves, roof (OE hrōf) – roofs/rooves. Some interesting evidence for continuity may be provided by double borrowings from French such as naif, n. adj. (1531, 1598), naive, fem. (1614), digestive, adj. (1532), digestif n. (1908).
For the sibilants and the dentals, the main factor initiating and solidifying the contrast underlyingly must be prosodic. The generalizations are somewhat different for the two places of articulation: for the /θ/:/ð/ contrast it is the placement of the word in the clitic group that initiates initial voicing. Medially, the contrast in PDE is very much what it used to be in OE because of the rarity of geminate [-θθ-] in OE, although the picture is obscured by the frequent alternation between [ð] and [d] in intervoiced position.Footnote 61 The contrast in final position in PDE is based largely on fossilized [-ð] in verbs after the loss of -e: breathe, seethe, bathe, swathe, smooth, loathe, clothe, mouth, soothe, etc. Indeed the only [-ð] items without the possibility of analogy from a verbal form are with, and the adjective lithe, both of which are variable.Footnote 62 This distribution in PDE thus shows the residue of voicelessness in historically final position; note also that [-ð] can appear only after long vowels.
The sibilants owe their contrastiveness to a combination of factors, perhaps the most important one of which is the voicing of [-s] in inflectional endings, which was under way during the fourteenth century. Stress shift in borrowings with intervocalic [s] – passage (1300), vassal (1166), vessel (1330) – provide more evidence for contrastive distribution, though the sibilants continue to show some variability in word-medial position: compare PDE usage, unison, gosling (c.1425, OED), greasy both with [s] and with [z].Footnote 63 The noun–verb pattern must have been productive throughout ME, e.g. peace n. (1160) vs appease (1330). In this set the influence of language contact is minimal word-initially and word-medially, but in final position it must have been reinforced by the loss of final [-ə] in e.g. lease ‘glean’ [-z] < OE lesan vs lease, v. ‘grant’ (1292) <AF lesser; phase, n. ‘Passover’ [-z] (OE) vs face, n. [-s] (1300), and in verbs with the borrowed suffix -ize after the sixteenth century
7 Conclusion
Laker's (Reference Laker2009) hypothesis linking the development of a phonemic contrast between voiced and voiceless fricatives in English to the existence of a Celtic substratum lacks empirical support. The basic test for phonemic contrastiveness is the existence of minimal pairs: no such incontrovertible evidence has been found in OE. While the basic premise that in language-contact situations second-language learners will transfer their underlying system of contrasts to the new language is in principle unobjectionable, voicing contrast of the fricatives in OE, at any point of its 500-year history, cannot be demonstrated. The hypothesis of allophonic distribution in OE is much more solidly supported and the reasons for dating the reanalysis to post-Conquest English are empirically testable.
Categorization is always gradient: the final stage of completely lexicalized fricative voicing has not been reached yet, but for fifteen centuries there has been ‘(Labovian) cumulatively weighted variation in a particular direction’ (the phrase is from Lass Reference Lass1991–3: 22). The direction is from predominantly allophonic to predominantly contrastive distribution. Therefore there is every reason to endorse the traditional position: the phonemic split of the OE allophones occurred after 1100. Unlike the traditional descriptions of the phonemicization, which bundle together all fricatives in all positions, however, we should treat the developments in the three sets separately, identifying different triggers for the different sets. These triggers include: the familiar language-contact influence from Romance loanwords for /f/–/v/, phonetically based voicing in weak prosodic positions for initial /θ-/–/ð-/ and final /-s/–/-z/, the degemination of fricatives word-medially for /f/–/v/ and /-s/–/-z/ and the loss of final -e for all three sets.
