06st,   PDF
Attila Starčević

Middle English Quantity Changes – Further Squibs

0 Introduction The article is not envisaged to provide a comprehensive coverage of the literature written in the field, nor does it purport to give an answer to some of the continuing problems in assessing the full range of Middle English (ME) quantitative changes or investigate the range of implications suggested by some of its conclusions. The intentions are, however, to discuss some of the overlooked or rarely mentioned (and, what is more, entirely disregarded) regularities of early ME quantity changes. The aim of the article is to discuss the following issues: why cannot the regularity of open syllable lengthening be claimed to have been a general principle of ME phonology (it seems to have been blocked in many instances where its operation is expected: e.g. in the case of trisyllabic words or in words whose second vowel was not reduced), the issue of whether trisyllabic laxing was a phonological phenomenon of the language and whether it is possible that ME was a templatic language (and if so, in what sense). The interest in this part of ME phonology goes back at least to Luick (1914), a 19th century Neogrammarian, whose ideas on lengthening and shortening processes in late Old English (OE) and ME were recently taken up by Ritt (1994) offering an apologetic discussion of Luick’s ideas in the framework of Natural Phonology. ME open syllable lengthening (MEOSL) is closely intertwined with the process of trisyllabic shortening (TRISH, to borrow a convenient catchword from Ritt 1994), both in the past and the present (Lahiri & Fikkert 1999). The issue of MEOSL has also been scrutinised from the point of view of syllable cut prosody (Murray 2000, echoing the ideas first proposed by Trubetzkoy 1938). A critical survey of these and similar ideas is also not taken up in this article.

1 MEOSL and TRISH step by step: problem and background In what follows a number of phonological changes will be touched upon as we proceed towards our interpretation of MEOSL and related problems. Their description follows their chronological ‘implementation’ in the history of the language. However, before we do this, a few remarks are necessary on TRISH and other shortening processes. The Even Yearbook 7 (2006), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218–8808, http://seas3.elte.hu/delg/publications/even, © 2006, Attila Starčević

Middle English Quantity Changes – Further Squibs 2

1.1 TRISH and related processes 1.1.1 The ‘newer’ stages Some of the processes, most notably TRISH and shortening before consonant clusters (SHOCC, cf. Ritt 1994), however, do not seem to be specifically bound to any of the changes outlined below and seem to have (re)appeared in the language either in the form of static diachronic residue (e.g. MoE herring, originally containing a long vowel in OE, assumed by analogy a short vowel from its trisyllabic counterpart: ME hēring ‘sing’ ~ hěringes ‘pl’, a fact known only to a historical linguist) or in accounts of Modern English (MoE) morpho-phonology, mostly in the form of ‘derivational’ rules now predominantly restricted to (some layer of) the lexicon (see the MoE dream ~ dreamt pair of related items for which one is forced to accept at least some sort of lexicalisation, cf. Kaye (1995) and especially Kristó (2005) for a recent discussion of some of these issues). TRISH has also commonly been invoked to account for various morpho-phonoloical relationships holding between related items of the vain ~ vanity type. Although no readily available (phonetico-)phonological explanation is at hand for the alternation between // and // (unless one takes a step back in the history of the language in the form of transformational rules and assumes that, for instance, the ‘surface’ vowel of vain is actually // which undergoes laxing/shortening in a trisyllabic environment (vanity) and which, once it has been subjected to other phonological rules in a nontrisyllabic environment (e.g. raising to // and diphthongisation), surfaces as // in vain, as was variously assumed in Chomsky & Halle (1968) and a host of other transformationalist accounts), the morphological relationship between vain and vanity seems more straightforward. The suffix -ity (one of the Level I affixes of Lexical Phonology, cf. Kiparsky 1982, 1985, etc.) is attached to the base vain yielding vanity. In addition to this, even the semantic relationship is transparent: vanity means ‘the state of being vain’ (whether one can glean anything from the dimension of meaning and whether it is relevant at all in a transformational grammar affecting the structure of words was taken up by Aronoff (1976), for example). The issue of whether vain/vanity (and a host of other traditionally quoted examples like nature/natural, Bible/Biblical, tone/tonic, etc.) are indeed related in the sense of ‘B is derived from A at any stage of the language’ has been questioned by Lahiri & Fikkert (1999). They claim that words like obscene/obscenity were borrowed independently at different stages of the language (sometimes the derived word The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 3

appears earlier in the language than its supposed base and the temporal gap between them can be a few hundred years). Lahiri & Fikkert (1999) claim that it was only after the Romance suffixes (e.g. -ity) turned productive (in around the 17th century) that unrelated ‘base’ and ‘derived word’ entered into a morphological relationship, i.e. a ‘post hoc’ derivational relationship was established between them. 1.1.2 The ‘older’ stages TRISH and SHOCC have been claimed now and again to have been operative in the older stages of English, too. According to Luick (1914), SHOCC first applied to long vowels followed by three consonants. His well-known example is brēmblas ‘brambles’ which shows up as brĕmblas in late OE. According to the ODEE, bramble ‘blackberry bush’ (the singular of the form quoted by Luick) derives from the base brōm ‘broom’ and the diminutive suffix -el (< *-il),1 the older form being brēmel (also: brǣmel). A similar example is thimble ‘little thumb’: OE þymel (with þ/ð = [] or [] depending on the environment), the base being þūma ‘thumb’ followed by -el. This ‘triconsonantal’ shortening was followed by ‘bi-consonantal’ shortening, i.e. an originally long vowel was shortened before a sequence of two consonants: e.g. late-OE ĕnlefen from ǣnlefen ‘eleven’. These shortening processes are claimed to have different causes. Yet, they both seem to stem from the fact that the original long vowel was followed by a coda-onset cluster (i.e., in traditional terms, this is due to syllabic organisation: the coda consonant closes the first syllable and is followed by an onset ). It seems then that the two processes are not different at all: the shortening boils down to the first consonant’s syllabic affiliation (the number of consonants that follow seem not to matter). Another comment is in order here in connection with OE brēmel. Luick (1914) gives the plural form brēmblas for a reason: the ‘unetymological’ -b- (and similar stop consonants like -d- and -g-) is known as ‘Sproßkonsonant’ (intrusive consonant, literally: sprouting consonant) and appears wedged between a sequence of a non-continuant and a continuant consonant (similarly to the MoE pronunciation of prince being identical with prints). Luick quotes brēmblas to make the three consonants appear ‘even’ closer to each other. Yet, there seems to be no reason why this intrusive consonant should not have appeared in the singular brēmel (> brēmbel) apart from the fact that the spelling suggests a vowel between -m- and -l-. This is 1 Since the data cited in this article are predominantly historical in character, the usual notational conventions apply: a single asterisk (*) for reconstructed data and a double asterisk () for historically ungrammatical/impossible formations. The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 4

certainly not decisive as this -l- was probably a syllabic consonant (and as such it was close enough to -m- for them to ‘produce’ an intrusive -b-). Actually, brēmble is the (early) form given in the ODEE. So, it seems, reference to SHOCC has been claimed a number of times in the history of the language. Note that once shortening happens, bramble has the same structure as dialectal shambles ‘slaughter house’ from OE sc(e)amul ‘stool’ which has never had a long vowel. TRISH, according to Luick, also affected late OE. Late OE TRISH was essentially identical to modern TRISH: in a trisyllabic word (inflected or derived) with no secondary stresses the stressed vowel is shortened. Some examples follow: (1) TRISH in late OE Singular cīcen hēafod ǣnig clōfer hǣring

Plural cĭcenu hěafodu nige clăfere hringas

‘chicken’ ‘head’ ‘any’ ‘clover’ ‘herring’

Luick (1914) assumes that late OE TRISH can only apply in words that have no secondary stress, i.e. the long vowels in trisyllabic words were only shortened after secondary stress was lost. The place of secondary stress in the words below depends on the notion of the Germanic Foot (Dresher & Lahiri 1991), which is not immediately relevant now (examples from Lahiri & Fikkert 1999). (2) Loss of secondary stress and TRISH in late OE ǣ́rende sū́þerne lā́ferce dḗorlìngas frḗondscìpe

ě́rende sŭ́þerne lăferce dě́orlingas frě́ondscipe

‘errand’ ‘southern’ ‘lark’ ‘darling, pl.’ ‘friendship’

So, TRISH in these words occurs only after secondary stress was lost. In words like holiday (< hāligdæg) one has to assume that morphological structure became non-analytic, i.e. the internal structure of the word was no longer transparent (in this case it was no longer a compound). The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 5

1.1.3 TRISH as SHOCC? There have been a number of attempts to unify the environments of TRISH and SHOCC responsible for the above processes. Kiparsky (1968) unifies the two environments for both OE and MoE: (3) TRISH and SHOCC in late OE and MoE C V  [-long]/_(C)C …V … V The only difference between late OE and MoE is the presence of the bracketed consonant in (3): in late OE the environment was longer by one consonant, with the condition relaxed in MoE. This unification of early generative grammar can be criticised from a number of perspectives: (i) the three-consonant environment does not seem to be significantly different from the two-consonant environment (cf. brĕmblas vs. ĕnlefen) in view of its effect, (ii) if another consonant should be lost in a further stage of the language, the two environments should have nothing in common at all (_C and _…V … V) and (iii) if one assumes that use of such braces is not only to capture a number of seemingly similar processes but also, and more importantly, to express a ‘natural’ inclination of such environments to produce the same outcome, it is difficult to see what the underlying cause for shortening is (apart from the very suggestive formalisation that there is such a process as the one described by the notation itself) in alternations like sincere ~ sincerity, keep ~ kept (and historical ‘alternations’ such as brēmblas ~ brěmblas or ǣnlefen ~ ĕnlefen). To account for lack of shortening in words like nightingale, Abraham, etc. Kiparsky (1982, 1985) claims that TRISH is a Level I process that applies in derived environments (and, accordingly, fails in non-derived words). Myers (1987) argues for a stress-based resyllabification and syllable extrametricality, i.e. sequences like V.CV are resyllabified as VC.V. Some suffixal syllables (e.g. -ic) are independently assumed not to be extrametrical. This accounts, for example, for tonic (which is ton.ic), sincerity (sin.cer.ity), natural (nat.u<ral>), etc. As opposed to this, nature is analysed as na.<ture> (the first syllable is not closed and thus no shortening takes place). The stressed vowel in na.<ture> has to be specified as tense (or long) underlyingly. One wonders, however, why there is no ‘broadening’ in the stressed closed syllable of sincerity under the influence of coda-r. The answer The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 6

lies probably in assuming ambisyllabicity and a serial derivation which takes away // at the right moment, i.e. after it has closed the syllable but before ‘broadening’ has a chance to apply (i.e. before // followed by a tautosyllabic-r is ‘broadened’ to //). As we can see, TRISH is actually not different from SHOCC. Yip (1987) does not resort to syllable extrametricality but rather to epenthesis. Suffixes like -ity, -ic, etc. are assumed to be consonant-initial underlyingly forcing the preceding consonant to close the syllable: san.ty, ton.c, etc. Later, epenthesis inserts the appropriate vowel: san[i]ty, ton[i]c. Again, TRISH is a form of SHOCC. As opposed to this, Lahiri & Fikkert (1999) are less ambitious and argue that TRISH and SHOCC are not reducible to the same cause: SHOCC is syllable based, whereas TRISH is basically a prosodic phenomenon optimising the foot structure of words allowing only a specified amount of phonological ‘material’ in a window, so to speak. The exact details of this approach need not concern us here. As we can see TRISH and SHOCC have, on various counts and in various forms, been claimed to be part of the phonology of the language. The next phenomenon seems to have played a significant role in the shaping of MoE phonology and was part of ME. 1.2. Homorganic lengthening (HOL) Some of the processes taken for granted in the traditional accounts of late OE and early ME phonology are the lengthening of the five short vowels of late OE (as represented by the West Saxon variety of OE, for example) in closed syllables whose first member is a sonorant (r, l, N = any nasal homorganic with the following obstruent) and the second member a voiced obstruent, basically b, d, z and g (Wright & Wright 1928, Wardale 1936, Brunner 1970, Luick 1914, etc.). The minutiae of this process are complex: see Ritt (1994: 96) for the name of the process (HOL) and a probabilistic ‘natural’ formula with a number of components inversely or directly proportional to vowel height, the weight of the syllable following the original short vowel, the stability of the nucleus of the following syllable, etc. A characteristic sample follows (problematic clusters in -rC- as well as items not found in standard RP have been disregarded):

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 7

(4) late OE HOL resulting in ME long vowels wild

gold

climban > camb

windan > kind

wīld gōld clīmben cōmb wīnden kīnd

‘wild’ ‘gold’ ‘climb’ ‘comb’ ‘wind’ (verb) ‘kind’

HOL results in long vowels followed by a consonant cluster (more precisely, a traditional coda-onset cluster: e.g. -ld-, -nd-, -mb-, etc.), which is a rather marked phonological constellation taken the generally observed restriction of ‘closed syllable shortening’ in the world’s languages (this can be supported by both diachronic and synchronic data:): generally a long vowel is banned from occurring next to a coda-onset cluster. This well-documented constraint has found its way into modern phonological theories (e.g. Government Phonology, Strict CV phonology and VC phonology, for example) in the form of various lateral relationships holding between governor and governee or the sharing of ‘burial’ domains, for example. The fact that such -VC$Csequences do exist (where $ marks a syllable-boundary), however, have been attributed to various factors. The majority of surviving HOL examples consist of long vowels followed by coronals (e.g. wild, child, kind, gold, etc.). The unmarkedness of coronal consonants has been claimed now and again (Paradis & Prunet 1991). Ritt (1994: 90) assumes that in the remaining cases of long vowels followed by non-coronal clusters (e.g. climb, comb) the offending constellation of a super-heavy syllable (//) was remedied by the loss of //. This seems a fair conclusion, but obviously cannot be the whole story because // was, in all likelihood, lost for independent reasons: short vowel plus non-coronal clusters have also been remedied trough the loss of // and // (e.g. lamb, king). In addition, in MoE there are no surviving examples of long vowels plus velar nasal // (< //) sequences. In addition, Ritt 1994 (criticised by Lahiri & Dresher 1999) has assumed that highly sonorous coda-onset clusters weigh 1 mora but since they are usually ambisyllabic (i.e. they belong to two different syllables at the same time), they will lose half of their weight and thus weigh only .5 mora. This assigns the stressed syllable of a typical OE input like bindan the weight of 1.5 moras: bi[nd]an (with square brackets representing ambisyllabic

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 8

consonants). In Ritt’s (1994) view the ideal weight of ME stresses syllables was 2.5: thus, after HOL, OE bindan (1.5 moras) was transformed into ME binden weighing 2.5 moras with the missing mora supplied through the lengthening of the vowel. A commonly acknowledged source of interaction with HOL is shortening of the long vowels in case the homorganic cluster is followed by another consonant (this seems reminiscent of the triconsonantal shortening of OE): ME chīld ‘child’ vs. chĭldren ‘children’ (childer, presumably with a short vowel, is also recorded, cf. Moore 1929). Such examples are not plentiful in the traditional literature. This may be due to morphological factors: OE plurals in -r, e.g. cild ‘sing.’ ~ cildru ‘pl.’, lamb ~ lambru, are very few in number and this class is non-productive (some members of this class were reanalysed and reclassified as members of other more productive classes even in OE times). Such plurals were probably lexicalised in late OE and/or early ME and this seems to be supported by the MoE reflexes: children is a ‘reinforced’ or ‘double’ plural with two original plural endings, the latter a so-called ‘weak’ plural traditionally, which enjoyed a ME productive revival (Wardale 1958). In MoE, such plurals are also lexicalised. In other cases, the lexicalised plural form was lost altogether and the base was subjected to the more regular process of plural suffixation in MoE: lamb ~ lambs. Another generally accepted view is that HOL was counteracted by TRISH. Thus, the stressed vowel of OE wildernysse is assumed to have had a short vowel in ME because it is in the antepenultimate position. This vowel quantity is continued into MoE wilderness. Even if one tries to subsume the case of wildernysse under the tri-consonantal environment (children) by assuming that // was lost (or apocopated) it seems highly unlikely that // should not have become syllabic (wildrnysse) in this environment. So, it seems TRISH is just another means of shortening a vowel in the HOL environment. 1.3 MEOSL proper 1.3.1 Open syllable lengthening in the West Germanic languages It has been quite an uncontroversial assumption that the West Germanic languages underwent a process of open syllable lengthening (OSL) in the 13th and 14th centuries (Prokosch 1939). This effectively means that sequences of C0V were transformed into C0V. In Lahiri & Dresher (1999) a survey is given

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 9

including Dutch, German and English. Probably, the process is most easily captured in Dutch from a diachronic perspective which still has short-long vowel alternations (although these are regarded as ‘exceptional’ today) for monosyllabic nominal stems with an ending in the plural, the basic pattern for them being C0VC ~ C0VCV… (depending on the type of plural suffix). A few collated examples follow from Lahiri & Dresher (1999) with length marks as they appear in the original. (5) OSL in the Dutch nominal paradigm involving monosyllabic stems Singular bad gat dal weg smid god hof blad

Plural baden gaten dalen wegen smeden goden hoven bladeren

‘bath’ ‘hole’ ‘dale’ ‘road’ ‘smith’ ‘god’ ‘courtyard’ ‘leaf’

Lahiri & Dresher’s (1999) main objective is to account for the paradigmatic alternation of short and long vowels in Dutch, German and English and a possible direction of levelling in the paradigm in either direction (i.e. the generalisation of either the long or the short vowel throughout the system). Their conclusion is that the Dutch system shows what it ought to show, i.e. OSL regardless of the number of syllables that follow the stressed open syllable (in other words, there are no impeding processes such as TRISH: cf. bladeren ‘leafs’ vs. goden ‘gods’ above). In multisyllabic stems, as expected, OSL also took place. A few examples follow (note that the original vowels were all short, as shown by their OE and German counterparts, examples are from Lahiri & Dresher 1999): vogel ‘bird’ (OE fugol), vedel ‘fiddle’ (OE fiele), honing ‘honey’ (OE hunig), weduwe ‘widow’ (OE widewe), etc. A similar effect is observed in the singular of disyllabic nouns with original schwa-final nouns in Middle Dutch: taal (< tale, OE talu) ‘tale’, naam (< name, OE nama) ‘name’, zoon (< sone, OE sunu) ‘son’, haak (< haka, OE haca) ‘hook’, etc. In Dutch (as well as in the other West Germanic languages) there were also geminate consonants that were lost word finally together with the final schwa. In the plural the geminate is still retained and, as expected, there is no The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 10

OSL: zon ~ zonnen ‘sun’ (< sonne ~ sonnen), cf. zon ~ zonen ‘son’ (< sone ~ sone) with the expected effects of OSL (note that the Modern Dutch plural suffixes may differ from their Middle Dutch counterparts, but this is an issue for morphology and does not alter the picture of phonological effects).2 In German the process was complicated by the so-called second consonant shift which created geminate consonants that blocked the application of OSL. This change operated on Old High German voiceless stops only changing them into either voiceless affricates or fricatives. Also, there were other general processes that created closed syllables: Old High German // was devoiced to // (and, it seems, also geminated to //) and other minor instances of geminaton before the suffixes -el and -er. A few collated examples follow (taken from Lahiri & Dresher 1999): (6) The second consonant shift and the absence of OSL OE wæter open wacian wicu

Old High German wazzer offan wahhēn, -ōn wehha/wohha

English water open wake week

The descendants of the above Old High German forms all have short vowels today: Wasser, offen, wachen and Woche (the English counterparts, with the exception of water whose length is not a ME phenomenon, all show the reflexes of OE vowels lengthened by MEOSL). In words with no second consonant shift, the application of OSL predictably occurs: Zahl (< zala, OE talu) ‘number’, Sohn (< sunu, OE sunu) ‘son’, Nahme (< namo, OE nama) ‘name’, etc. Given that the effects of the second consonant shift were such that all of the original voiceless stops were affricated or spirantised (and thus capable of creating closed syllables) and that another change devoiced (and geminated) the original // to //, the only remaining environment in which short vowels were in open syllables was before voiced consonants (cf. Lahiri & Dresher 1999: 688). There was another change but its outcome was also a 2 Lahiri & Dresher are silent about the actual pronunciation of the plural form zonnen ‘suns’. If there is no geminate consonant in zonnen and the vowel is a short //, then OSL is obviously a process which no longer applies and its effects are a diachronic residue of a once active synchronic process. A similar case is observed in German where the counterparts of Dutch ‘sun’ are Sonne // ~ Sonnen // with no geminate and no OSL. The original open syllables were subject to OSL: Sohn // ~ Söhne // ‘son’ (cf. Dutch zon ~ zonen). The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 11

voiced consonant: // > // (cf. OE pæ, with the original voiceless fricative vs. German Pfad). In Modern German there are no alternations similar to the Dutch case above (dag ~ dagen ‘day’): the once existing alternation between, for example, mono-syllabic singular and disyllabic plural forms (Tag ~ Tage ‘day’) was reanalysed in favour of the long vowel (Tag ~ Tage). The reanalysis was generally in favour of the long vowel (also before sonorants: cf. German Hohl vs. OE hol ‘hole’). Lahiri & Dresher (1999) contend that this may have been furthered by the fact that vowels are generally perceived as phonetically longer before voiced consonants. So, disregarding the original long vowels that generally remain unaffected by the consonant shift, in Modern German before voiced obstruents vowels can only be long (e.g. Tag ‘day’, Hasel ‘hazel’, etc.). Before voiceless ones, of course, they can be both long and short (the long vowels going back to original long ones): Schrot ‘chicken fodder’ vs. Schrott ‘scrap metal’. The situation before sonorants is similar: here, both long and short vowels can be found: (Jungge)selle ‘bachelor’ vs. Seele ‘soul’ (the short vowels are only found before originally geminate sonorants: cf. Sohn ‘son’ vs. Sonne ‘sun’). This constraint brought about by diachronic coincidence (i.e. the fact that the remaining single intervocalic consonants are all voiced) is so pervasive that King (1969: 53) claims that all short vowels are lengthened before voiced obstruents in early Modern German (with a few counterexamples like Ebbe ‘ebb’, Rogge ‘rye’, etc.). What escapes Lahiri & Dresher’s (1999) analysis, however, is that there is one environment left which clearly shows that the original motivation for lengthening was indeed the open syllabic environment and not the voiced nature of the following consonant: the Germanic or primary //3 has never undergone intervocalic voicing,4 or gemination and accordingly the vowel before it could undergo OSL: Ofen ‘oven’ (which in its result coincides with oben ‘above’ which also underwent OSL and had a short vowel historically). Such examples are difficult to come by. The only other example known to the author is Neffe ‘nephew’ which in Modern German has a short vowel but its recorded Middle 3 Note that this // can be called ‘primary’ because it was inherited from Common Germanic and was not the result the second consonant shift (cf. English open vs. German offen); it only later fell together with secondary //’s of OHG origin, as seen in offen (// < // < //). 4 As opposed to the inherited Germanic // (e.g. Hasel ‘hazel’) which underwent intervocalic voicing. It has never fallen together with ‘secondary’ /s/’s whose source is the Old High German /t/ through spirantisation and gemination. This explains why one gets // in Hasel but // in hassen (< hassēn, English hate). So, it seems, intervocalic voicing in German is just another diachronic relic which fails to affect ‘new’ //’s from // via //. The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 12

High German form neve suggests a long vowel (similarly to Ofen which appears as oven with the Dutch letter <v> for //, cf. German von //). TRISH also seems to have played no role in German: OHG fedara > Feder ‘feather’). Similarly to Dutch then, German OSL is not counteracted by TRISH. It is, nevertheless, blocked by gemination or, in other words, SHOCC (cf. Hasel vs. hassen). SHOCC (or, in a slightly different guise, the absence of lengthening) also appears in Dutch, as we have seen, in the case short vowels followed by geminate consonants (cf. zon ~ zonnen). It seems that in Dutch and German OSL was indeed lengthening of the stressed short vowel in open syllables uninfluenced by TRISH. 1.3.2 OSL as MEOSL Until recently, OSL in ME has been seen as a general process affecting the length of the stressed short vowel in an open syllable changing it into a long vowel: C0VCV  C0VCV (Wright & Wright 1928, Wardale 1958, Brunner 1970, Luick 1914). The traditional examples are OE talu > MoE tale, wicu > week, mete > meat, cradol > cradle, nacod > naked, etc. The ME data show the general weakening of OE unstressed vowels to a sound generally spelled <e> and in all likelihood pronounced // (see Minkova (1991) for a detailed study on the history of unstressed vowels in English from syntacticomorphological and phonological perspectives), i.e. OE <a>, <e>, <u>, <o>  ME <e> // (the OE vowels were probably pronounced as suggested by their written form). This schwa was ultimately lost in ME if word-final (tale, meat, week). However, a number of ideas have been put forward suggesting that there was no general MEOSL. Minkova’s (1982, 1985) basic counterclaim to the view of general MEOSL can be summarised as follows: the change (C0VCV  C0VCV) operates unfailingly only in those cases in which the original stressed short vowel is followed by // which is eventually lost (so, in words of the tale-type). Given this, it seems there are no MoE words of OE provenance of the C0VCV type which did not undergo MEOSL. In contrast to this, it appears there are nearly as many words of the OE C0VCVC type with long vowels as those with short ones (for an exact proportion expressed in percentages see Lahiri & Dresher (1999: 691); yet, logically there seems to be no rationale behind attributing any significance to any one group of the two almost equally distributed set of items): e.g. OE æcer > acre, cradol > cradle, hræfn > raven, stapol > staple, etc. vs. OE botm > bottom, ofen > oven, sadol

saddle, weder > weather, etc. Lahiri & Dresher (1999: 692) are forced to The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 13

admit that any attempt at drawing a conclusion based on statistical data alone is compromised on a number of accounts. A summary follows: (i) some words have not survived into MoE (ātor ‘posion’), (ii) some disyllabic items have short-vowelled reflexes (hænep > hemp),5 (iii) the selection of MoE items is arbitrary (OE cradol survives in non-standard dialects as craddle/creddle)6 and (iv) there may be OE alternations in declensional classes in relation to the ME outcome (MoE grave can descend from both OE græf and grafu; see the discussion below). So, it seems numbers per se can hardly be used to solve a linguistic issue. Up to this point we have seen that there was general MEOSL in OE words of the C0VCV type and the absence of such a general lengthening process in C0VCVC type words. We have, however, not investigated the generally assumed cause for the lack of such a general process. Before we attempt to do this, a slight digression is necessary to see what happened to OE words of the C0VC type. Words of this type in MoE can be found with both long and short vowels (a few items are listed from the list in Lahiri & Dresher 1999: 691): (i) OE bæþ > bath (the length of the vowel is not a ME phenomenon), broþ > broth, god > god, stæf > staff (again, the length is not a ME phenomenon), þæc > thatch, etc. vs. g(e)oc > yoke, dæl > dale, hwæl > whale, hol > hole, etc. MoE grave is thus indeterminate between OE græf and grafu: both OE words (of an originally different declensional class: a so-called a- and ō-stem, respectively) could potentially end up with a long vowel. It seems OE words of the C0VC type levelled in both directions and this is due to the morphology of the language at the time of the operation of the rule and analogical reorganisation on the basis of original OE C0VCV (i.e. talu) type of words. In (7) there is a typical (very) early ME paradigm of an OE monosyllabic noun (the representation of suffixes is rather conservative and follows the phonological rules of ME as applied to the respective OE suffixes shown in brackets): (7) A typical OE monosyllabic noun in ME illustrated (cf. Moore 1929) Singular Plural Nom. hol hōles (< -as) ‘hole’ Acc. hol hōles (< -as) Dat. hōle (< -e) hōle(n) (< -um) Gen. hōles (< -es) hōle (< -a) 5 One could also add original disyllabic items with a stressed long vowel (e.g. monaþ > month) that are equally affected. 6 Almost equally possibly, although not explicitly mentioned by the authors, one can assume OE sadol to have survived as sadle in dialects other than ‘standard’ BBC English. Wright & Wright (1928) supply a number of ‘recorded’ examples: stapple ‘staple’, even ‘even’. The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 14

We can see that an original OE monosyllabic noun was also subject to MEOSL in the majority of the oblique cases (even after the general ME extension of the plural suffix -es to the rest of the plural forms: hōles which is indeterminate between nom., acc., dat. and gen. plural; what matters is that the suffix is still vowel-initial). The conclusion one can reach is that there was analogical restructuring in (at least) some of the original OE monosyllabic nouns induced by the bisyllabic oblique forms (both singular and plural). Again, the direction of levelling is unpredictable (hole vs. god). The answer to the question of why the C0VCVC type of words end up in MoE with both short and long vowels (saddle vs. crādle) can be sought in the declensional characteristics of ME nouns. This is the point where traditionally TRISH steps in. The following typical paradigm illustrates this. (8) OE disyllabic nouns in ME and TRISH after suffix levelling Nom. Acc. Dat. Gen.

Singular sādel/crādel sādel/crādel sadele/cradele sadeles/cradeles

Plural sadeles/cradeles sadeles/cradeles sadeles/cradeles sadeles/cradeles

‘saddle/cradle’

The traditional explanation is that here (or more precisely, in ‘half’ of the cases) there was also analogical levelling: in saddle, for example, the short vowel from the trisyllabic forms was generalised to the base form (here, in a pre-theoretical case understood as the nominative case), whereas in cradle this did not happen. There is no principle which predicts which of the two forms (one with a short and one with a long vowel) will be generalised. It can be seen that TRISH exercised a counteracting effect on MEOSL: in this respect, ME is different from the rest of the West Germanic languages we have examined so far. It is not generally influenced by OSL. To continue our discussion of MEOSL, the next batch of OE words to consider is disyllabic nouns with a stressed long vowel. Taking into account the workings of ME phonological rules (MEOSL and, as we have seen, TRISH) this class of nouns is also expected to undergo the effects of TRISH, but not MEOSL (as the vowel is already long, of course), see (9):

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 15

(9) OE disyllabic nouns with a long vowel in ME Nom. Acc. Dat. Gen.

Singular hǣring hǣring hæringe hæringes

Plural hæringes hæringes hæringes hæringes

‘herring’

As expected, some of the oblique forms show the effects of TRISH. The display, again, predicts that there may be analogical levelling in either direction. This trend seems to be substantiated by the data: there are nearly as many nouns with long vowels as there are with short ones. Some data follow to support this (from Lahiri & Dresher: 692): OE bēacen > beacon, ǣfenn > even(ing), hūsl > housel ‘Eucharist’, tācn > token, etc. vs. OE bōsm> bosom, dēofol > devil, mōdor > mother, wǣpen > weapon, etc. In conclusion, it seems that the only class of OE nouns that show no diachronic alternation with respect to the stressed MoE vowel is the class of C0VCV nouns (the tale-type) where TRISH simply did not have the right grounds to work on (there are no trisyllabic forms). Lahiri & Dresher (1999) arrive at the same conclusion but claim that there are a few exceptions (unfortunately, no examples are supplied). The classes in which alternation is expected and recorded are the OE (i) C0VC type nouns (hole vs. god) where the long vowel may have been extended from the oblique cases, (ii) C0VCVC type nouns (cradle vs. saddle) where either MEOSL or TRISH takes the upper hand and (iii) C0VCVC type nouns where the results of TRISH have either made their way into the base form or not.

2 Is MEOSL really MEOSL? 2.1 TRISH again As we have seen in 1.1, TRISH as a diachronically valid phonological process observed in morphologically related alternating pairs like vain ~ vanity can be questioned on a number counts. The fact that pairs of words like vain ~ vanity are considered related today is (according to Lahiri & Fikkert 1999) because a number of Romance suffixes, such as –ity, have accidentally become productive in the history of the language and, as a result of this, originally

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 16

independent words (with their respective long and short vowels) like vain and vanity have been drawn closer together by the now productive suffix. One of the problems with TRISH in ME is that it offers little explanation because in this stage of the language there would have been very few words with no trisyllabic forms (cf. Ritt (1994) for a similar conclusion). In other words, nouns, adjectives and verbs (all of the major lexical categories) would have appeared in a trisyllabic form at least now and then: e.g. ME makede ‘he made’, makedest ‘you (sing.) made’, makende ‘making’ vs. māked ‘made, ppl.’; smal ‘small’, smāle ‘small, pl.’ vs. smaleste ‘the smallest (one)’, etc. If TRISH is accepted as a valid explanation for OE sadol having a short vowel in MoE (saddle) as opposed to OE cradol having a long vowel (cradle), the whole machinery for deriving MoE values appears rather suspect because a principle is invoked rather arbitrarily whenever the need arises on the basis of MoE forms. Perhaps the regularity behind the MoE values vis-à-vis their ME values is more elusive than appears at first sight (to be discussed in Section 3). Ritt (1994) assumes that the only reason why Luick (1914) added TRISH (alongside SHOCC, MEOSL and HOL) to his list of quantity adjustments in ME is because this was the only way to salvage saddle from not undergoing MEOSL. Ritt (1994) gives a probabilistic ‘natural’ formula (as explained above) to account for MEOSL and derives the individual changes from the base form: in his view, one of Luick’s shortcomings was his unwillingness to accept that sometimes the very same phonological environment (e.g. sadol vs. cradol) can lead to different results (saddle vs. cradle). Apparently, the only phonologically plausible solution was TRISH and analogy in the establishment of the base form (saddle was reanalysed as containing a short vowel, originally the result of TRISH, in its lexical representation). The exact degree to which trisyllabic words were preserved into ME is very difficult to ascertain. In this article no attempt can be made at full coverage of this issue. A satisfactory account of this issue is probably next to impossible given the extremely varied dialectal picture of ME and the later spread of the dialectal forms into the ‘standard’ language after a change had become inoperative/obsolete (see 3.5). The issue of trisyllabic forms is inextricably intertwined with the ME vowel spelt <e> and probably pronounced as a schwa. A comprehensive account is given in Minkova (1991) who shows that the loss of schwa was not only complicated by morphological factors (some of the ME schwas showed various inflectional categories such as person, number, definiteness, etc.), but also by the euphonic principle of stress-clash avoidance (as, for example, in the retention of // in late ME in the so-called definite declension: e.g. the man was black vs. the blacke man). The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 17

Some of the reasons for the possibility of ascertaining the full impact of trisyllabic forms in ME (and thus the effect of TRISH) is diluted by the following facts: (i) morphologically simplex OE trisyllabic forms are very difficult to come by (the usually cited monomorphemic forms are ǣrende ‘errand’ and sūþerne ‘southern’ and they show a short vowel today), (ii) compounds are rather numerous in OE, but as they are morphologically complex and consist of more than one morphological domain, phonological rules are expected to treat them differently (of these compounds the majority is lost in MoE: e.g. bereman ‘potter’, andsaca ‘apostate’); some OE compounds survive into MoE but their form is complicated by other phonological changes and thus the impact of TRISH is difficult to trace (e.g. the vocalisation of OE //, the loss of intervocalic // or some other less usual changes): hlǣfdige > lady, hēafodu > head; the remaining compounds, if reanalysed as monomorphemic, show a lax vowel today: hāligdæg > holiday, bēoforlic > Beverly), (iii) the existence of secondary stress of OE origin in ME (cf. Campbell 1959, Kim 2002) may also have had an impact on the rapidity of the loss of schwa. Moore (1929) assumes that secondary stress is maintained in early ME polysyllabic words and schwas are not lost in one fell swoop: mýnecē[ne ‘nun’ > mínechě[ne (loss of length) > mínchene (loss of medial schwa) > mínchen (loss of final schwa) > mínchen (loss of secondary stress). Wright & Wright (1923) assume that schwas were lost earlier in disyllabic words with a short stem syllable than in those with a long stem syllable. Yet, no textual justification is supplied. The various accounts of schwas loss are reviewed in Minkova (1991: Chapter 2). The effect of trisyllabic forms in inflected nouns has already been exemplified: crādel ~ crădeles with MoE cradle showing the continuation of ME crādel and hǣring ~ hringes with a short vowel in MoE herring from the inflected form. Lahiri & Dresher (1999: 694), criticising Minkova (1982) for her choice of data, say that ME trisyllabic verb forms result in a lax vowel in MoE: e.g. OE gædrian > ME gadrien > MoE gather. However, even a superficial survey of some of the surviving OE verbs show that trisyllabic forms with no internal cluster do not seem to have impeded MEOSL: OE macian > make, werian > wear, hopian > hope, wacian > wake (but possibly also < wacan), sparian > spare, etc. It seems then that OE trisyllabic verb forms behave identically to disyllabic verbs (teran > tear, scacan > shake) with MEOSL and verbs having an original long vowel (sēarian > sear, hlēapan > leap, bītan > bite, lāþian > loathe) which could not undergo MEOSL. If one assumes that TRISH was an active phonological phenomenon of ME one would expect at least some examples of trisyllabic verbs with no internal consonant cluster (as opposed to verbs like gædrian) to have survived The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 18

with a short vowel. In addition, no account seems to have claimed the importance of inflected verb forms in analogical formations, e.g. the levelling of the short vowel of makede ‘he made’ into the disyllabic form māken > MoE mac. The only examples of levelled vowels as a result of TRISH into disyllabic forms with long vowels have always come from the nominal paradigm (e.g. saddle vs. cradle). Although the full implications of this and related phenomena cannot be investigated it seems that TRISH is at best unsubstantiated as a ME phonological phenomenon. The further discussion of this and related issues will reveal a number of other problems too. 2.2 An overlooked regularity? Up to this point we have seen a number of late OE and ME phonological regularities, most notably that of MEOSL and an alleged counterbalancing effect called TRISH. We have also seen a number of reasons why TRISH cannot be the solution to all of the problems. 2.2.1 Halo vs. hallow? One of the overlooked regularities of the lengthening processes in ME is comprised by words of OE origin which have reflexes in MoE of the C0VCVV template (with VV representing the diphthong //): e.g. mellow, yellow, etc. Limitations of space preclude a full justification of the term ‘template’, but I am going to use it in a non-traditional sense (i.e. not in the ‘traditional’ sense of the morphological limitations observable in the various Semitic languages, for example, where certain morphological formations, such as past tense or inchoative, must fall within the scope of a template of a given CV length): in the present use, the term template will simply be used to characterise the surface phonological form of a word with no intentions of saying anything about its morphological structure (see also section 3). The OE source of the MoE diphthong // is twofold: (i) it either represents the reflex of OE u (as in the so-called u- and ō-declensions) if followed by another vowel (in all likelihood, a schwa is the only candidate for this function) in which case the pronunciation of this // was probably // or (ii) the vocalised reflex of OE //. The pronunciation of OE // was environmentally conditioned: generally, the velar fricative [] in non-final positions (other than word-initial) after back vowels or consonants or its

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 19

voiceless pair [] word-finally after back vowels).7 These two developments are probably the only native sources for MoE post-tonic non-primary stressed full vowels. A short remark is necessary in connection with the OE u- and ō-declension: the regular template for this class of nouns was C0VCV in OE (duru ‘door’, talu ‘tale’, wudu ‘wood’, sunu ‘son’, etc.) continued into ME as tale, sune, etc. after the usual levelling of OE unstressed vowels in //. However, some C0VCV forms have survived into ME in their oblique forms which had a vowel following the template which was thus modified to C0VCwV (e.g. the OE nominative sc(e)adu regularly gives ME shāde; as opposed to this the OE oblique form sc(e)ad(u)we gives ME shad(e)we continued as MoE shadow). Note that OE and ME spelling is not always a reliable source from which one can judge the actual pronunciation: the cited OE sc(e)ad(u)we shows that some scribes recorded this word as sc(e)adwe and some as sc(e)aduwe possibly influenced by the non-oblique shadu. The vocalised reflex of OE // in intervocalic environment between back vowels is generally taken to be // which later came to form a diphthongal or a long nucleus with the preceding vowel (the MoE reflexes corroborate this): e.g. OE dragan [] > ME drawen // > draw //, OE boga [] > bowe // > bow // ‘weapon for shooting arrows’, OE būgan [] > bowen // > bow // ‘bend’. Although the exact details of this change would require a lengthy discussion, the general state of affairs can now be stated: OE vocalised [] and [] seem to have coalesced with OE // followed by a vowel (in which case it was [] possibly combining in one nucleus with the vowel following it) resulting in a ME post-tonic non-primary stressed unreduced vowel. Based on the MoE value // the ME vowel // suggest itself as a possible source. A survey of the ODEE entries has revealed that the surviving examples of the kind of changes exemplified above in MoE all unfailingly have a short vowel in an open syllable followed by <ow> //. Some examples follow: OE ar(e)we > ME arwe > arrow, ?OE (Anglian) belgan > belwen > bellow ‘roar as a bull’, OE (plural of bel(i)g ‘belly’) belga > belwes/belows > bellows, OE burg/burh (with an indication of [x] <h>, the 7 In OE there are two sources for word-final []: one is the devoiced [] which appears as one of the environmentally conditioned allophones of // (and derives from IE /h/) and the other is OE // (deriving from IE //). From the point of view of the vocalisation process there are no differences between the two []’s: compare MoE farrow < OE fearh < Germanic *farxaz < IE porkos to MoE borough < OE burh/burg < Germanic *burg-s (if there existed an IE source, it could only have been //). The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 20

devoiced variant of [], in spelling) > borough, OE borgian > borrow, OE fearh > farrow, OE fealgian > fallow (v), OE fealg/fealh > fallow (n), OE folgian > follow, OE mærg/mærh > marrow, OE sc(e)adwe > shadow (cf. shade < sc(e)adu), swelgan > swallow (v), OE sarh/sarg > sorrow, ME talg/talug/talow > tallow, etc. These forms show the vocalisation of OE // and the reinterpretation of OE // followed by a vowel as // which probably coalesced with the following vowel to yield ME // (cf. shadow vs. shade). One could also suppose that, as exemplified by sc(e)adwe, for example, final // merged into //. The minutiae of these changes are irrelevant now. As we can see, the above examples all behave identically to original OE // in the appropriate context: OE melwa > mellow, OE *mynwe > minnow, OE swealwe > swallow (n), OE w(e)alwian > wallow. It seems that there are no MoE words of OE origin of the above type that have a long vowel before //. The reason for this must be sought in some ME constraint which must have become inoperative after a time and, consequently, in MoE one can find words having a long vowel before // but these must be later borrowings: this is testified by halo which, according to the ODEE, appears in the 16th century, that is after the supposed time of the operation of the change. The word must have entered the language with both vowels long. Lahiri & Dresher (1991: 690) discussing those words in MoE that do not have long vowelled reflexes also take up the question of the C0VC<ow> type of words and claim that “shadow derives from ME shadwe, where w either created a consonant cluster or a disyllable which would have trisyllabic inflected forms.” Both of the assertions are questionable: the first one assumes that // was cluster and thus the MoE short vowel is the result of closed syllable shortening or SHOCC (shād.we) or is due to the absence of lengthening in such an environment. This does not seem to be the best solution because OE and ME did have onset clusters of the // type (e.g. dwellan ‘dwell’, dweorh ‘dwarf’), so one could potentially expect // to behave as such in shadwe (i.e. shā.dwe). The other claim is that the length of the MoE vowel is due to TRISH in an even longer form. This is difficult to substantiate given that shadwe is already an oblique form (there would have been no possibility of expanding it even further inflectionally even in case it was reanalysed as morphologically simple: a ME trisyllabic form like shadw- s ‘shadows’ is impossible). As we can see, an explanation resting on TRISH and SHOCC is not well motivated not only on internal grounds but also on the rather unlikely possibility of not finding at least some surviving forms with a long vowel generalised by analogy on forms that did not The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 21

undergo TRISH (as we saw was quite common and unpredictable in OE words of the C0VCVC template: saddle vs. cradle). That such a result (i.e. having shadow, for example, with //) is not encountered is strongly suggestive of the fact that ME did not have general OSL. In addition, it seems that not only were short vowels precluded from undergoing lengthening but long vowels were shortened in such environments: OE mǣdwe (oblique form of mǣd ‘mead’) comes down in MoE as meadow (further examples: OE hālga > hallow (n), hālgian > hallow (v)). 2.2.2 Further doubts about MEOSL Another point of interest in connection with MEOSL lies in the history of OE words containing -ig [] > ME [i] (e.g. OE bodig > body). Lahiri & Dresher (1999: 694) assume that (i) there was length variation in ME (bōdy and body, although no support is given for this assumption) and (ii) both original OE long vowels as well as short vowels are found short in MoE. Whatever the original situation, in MoE there are no long vowels before OE words containing -ig, i.e. the stressed vowel of the OE C0VC<ig> template (> ME C0VCV after the coalescence of the OE [] with [] in ME) is always short in monomorphemic words. A short survey of the ODEE supports this: OE belig

belly, OE bodig > body, byr(i)gan > bury, bisig/bysig > busy, myrig > merry, etc. Also, it seems that vowels were shortened before -ig: ǣnig > any, sārig > sorry, ME rēdy > ready (OE gerǣde, possibly also gerǣdig), etc. The shortening of the vowel also happened in forms that lost their morphological complexity: hāligdæg > holiday. One explanation for the prevalence of short vowels is linked to a possibility of the second vowel bearing secondary stress (e.g. Lass 1992: 73). Should this be so, then even OE hǣring can be part of this picture, similarly to OE ǣnig. Note that holy (< hālig) still has a long vowel in MoE but this seems due to its morphological complexity: there are two domains, (w)hole and -y, i.e. the word is still not monomorphemic, hence the vowel retains its length (note incidentally that OE hāl has acquired an unetymological <w> in spelling if in isolation (whole) but not if suffixed, as in holy). Based on this, it seems that MoE holiday contains a short vowel not because it happened to have been in a trisyllabic form in OE/ME but because hālig was reanalysed as a monomorphemic word (similarly to sārig > sorry, cf. sār > sore). The history of OE -ig > ME y shows that it followed a different path to the rest of the non-primary stressed vowels. First, it did not partake in the general late OE levelling of unstressed vowels (cf. OE talu > tale): this can The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 22

mean that the coalescence of OE [] with the preceding [] happened at a later time than the general loss of contrastiveness in unstressed vowels. Secondly, this vowel, as shown by the effects of the Great Vowel Shift, was a short //. The Great Vowel Shift only affected the long monophthongs and had the vowel been long, it would have been diphthongised to // in MoE (cf. ME fire // > MoE //). This is how it is possible for MoE // to come from either ME // (< OE -ig) or ME // of Old French origin (e.g. ME cite(e) > city) via the Great Vowel Shift. In MoE phonology // is taken to represent the neutralised contrast between unstressed // and // word-finally, as for example in city (sti ~ st ~ sti). In conclusion, it seems that MEOSL can hardly be a general process of OSL. What traditionally seems certain is that (i) ME words only underwent OSL unfailingly if the template was C0VC (ME talu > tale), (ii) ME words of the C0VCVC template (i.e. disyllables with an original short stressed vowel) show both long and short vowels in MoE (cradle vs. saddle) due to the interference of analogy based on TRISH and (iii) OE disyllables with long stressed vowels can also have both long and short reflexes in MoE (steeple vs. herring). The environments (ii) and (iii) revolve around the same principle, that of TRISH, working either from non-oblique into oblique forms or the other way around. Obscure compounds (i.e. compounds that are treated as monomorphemic) are also suggested to have undergone TRISH. All in all, MEOSL was counteracted by TRISH. What we have found has lent further doubts to a general process of OSL. First, we have questioned the importance of TRISH and then given further examples of overlooked ME regularities: (i) there are no long vowels before // in words of OE origin and (ii) in morphologically simplex forms long vowels are also missing from before // of OE -ig origin (what is more, original long vowels were shortened both before // and //). Also, we have suggested OE hǣring > herring is identical in its development to (the once morphologically complex) sārig > sorry (all have a short vowel today). What is more, even OE hāligdæg (after becoming morphologically simple) has acquired a short vowel, similarly to OE sārig: all seem to behave identically to the originally simple hǣring > herring. All in all, the general workings of OSL in ME have been questioned on a number accounts. The next section investigates a possible analysis of these phenomena.

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 23

3 ME as a templatic language? 3.1 Introduction The following sections will be speculative in nature and none of the claims can be investigated fully. The aim of this discussion will not centre around the justification of some of the phonological theories proposed to day or be very specific about certain issues given that for the present purposes some of the minutiae are irrelevant. With the introduction of Strict CV phonology (Lowenstamm 1999), the Coda Mirror (Ségéral & Scheer 1999) and VC phonology (Szigetvári 1999) it has became clear that every language’s phonological skeleton is built up of strictly alternating CV or VC pairs depending on theoretical flavour. As a consequence, every cluster, be it vocalic or consonantal, is virtual: a long vowel (e.g. //) encloses an unpronounced consonant so much as a consonantal cluster does an unpronounced vowel (e.g. //). For a substantiation of this and similar issues see Scheer (2004, 2006). The pronunciation of empty positions, especially those of vowels, needs special care and there are a number of ways of silencing an empty vocalic slot: e.g. by syncope, by being locked in a burial domain, by being parametrically licensed to remain silent, etc. (for a recent discussion of this and related issues, as well as their gradual development throughout the various theories see Balogné 2005). If every language is composed of strictly alternating CV or VC sequences then every language has the same template (i.e. a skeleton comprised of CV/VC units), i.e. languages are all templatic in nature, with various forces conspiring to make this less transparent: e.g. vocalic material spreading from one vocalic slot to another over the head of an empty consonant resulting in a long vowel (e.g. //) or the establishment of a relationship between two consonants leading to a muted vowel and thus to a consonant cluster (e.g. //). This latter type of relationship results in what is traditionally known as a coda-onset cluster. The representation of the socalled onset clusters (e.g. //) is more disputed (cf. Szigetvári (1999) and the references therein).

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 24

3.2 OE as a templatic language? In the Germanic languages syllable weight is straightforward: open syllables with a short vowel are light and closed syllables (i.e. syllables closed by a coda consonant) or syllables containing a long vowel are heavy. Dresher & Lahiri (1991) argue that the Gemanic metrical foot must dominate at least two moras ([μ μ (μ)] μ), where the head is shown in square brackets. Connected to this is the idea of syllable-weight resolution: if the second mora to satisfy the head’s requirement cannot come form the same syllable (i.e. when the syllable is light), the head’s weight is ‘resolved’ or bound together by the mora coming from the second syllable (regardless of its weight) to form a single metrical foot. The equivalence of a single heavy syllable and a light syllable followed by any type of syllable (heavy/light) has been noticed to play a role in a number of Germanic phonological regularities including the assignment of main and secondary stress, high vowel deletion in OE and Siever’s Law in Germanic and Gothic. These issues cannot be investigated critically here (see Dresher & Lahiri (1991) on the issue of the Germanic foot and Kim (2002) for OE and ME stress) but what seems certain is that OE had a ‘window’ or template outside of which certain phonological restrictions were no longer operative. Such a restriction concerned the distribution of the pre-OE high vowels // and // that were lost under specific conditions (e.g. Campbell 1959: 144ff), more specifically after stressed heavy syllables or a stressed light syllable followed by another syllable (this is know as a ‘resolved foot’ in Dresher & Lahiri’s account): OE word < *wordu ‘words’ weorod < *weorodu ‘troops’ vs. fatu ‘vats’, scipu ‘ships’, etc.; OE fēt < fēti ‘feet’ vs. hnyte < hnyti ‘nuts’ with <e> showing a reduced // in recorded OE). As the examples show, the high vowels were retained after a light syllable. They, however, also remained after a stressed heavy syllable followed by a light syllable: hēafodu ‘head’. If a word contained two high vowels one after the other, it was always the one closer to the heavy or resolved ‘window’ that was lost: *strangiþu > OE strengþu ‘strength’, *rīkiu > rīcu ‘riches’, *hēafodu > hēafdu ‘head’, etc. From the point of view of Dresher & Lahiri’s account, the high vowels were saved from erasure if they fell within the template of the Germanic foot, i.e. if they supplied a mora to the weight of the head (i.e. if they were preceded by a light syllable with which they formed a resolved foot). If they were not needed in terms of weight considerations, they were lost. This cannot be the whole picture because if there existed a general constraint on the erasure of ‘unneeded’ high vowels falling outside an allowed ‘window’, The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 25

then in OE we would have rīc from < rīkiu (and not the attested rīcu). The conclusion is, therefore, that the loss of the high vowels was not a phonologically conditioned regularity expressed solely in terms of syllable weight and its effect on the following i or u (if it had been, one would not expect to find forms like hēafodu in which u is preceded by a sequence of syllables that could potentially add up to something weighing more that just a light syllable), rather it must have been bound to the satisfaction of the OE template of having (at least) two moras in the head of the foot. Once the template was satisfied, there was no further need for i/u to undergo erasure (the situation with the rest of the OE vowels will not be discussed). The process is lazy and applies only as long as there is a minimal need for it (this can be contrasted with a truly (phonetico-)phonological process like wordfinal de-voicing in German where the length of words is irrelevant: Rad ‘wheel’ ends in [] just as much as Klassenkamarad ‘class mate’ does). So, this is why it can be said that there was no general loss of the high vowels in OE based on weight alone. There must have been at least some interplay between a phonological and a templatic constraint (which means that there is a ‘window’ beyond which phonology was blind at the time). The templatic constraint can also be observed in MoE in connection with the so-called minimal-word constraint (cf. Szigetvári (1999) for a discussion). Without going into detail and assuming some rather rough-hewn formulations, it seems MoE only allows minimal words of the C 0VC or C0VV size at the right edge of a word: // and // are well formed, but *// is not. Yet, a word cannot be salvaged from being sub-minimal even with the addition of further phonological material comprising a stressed syllable, for example, to the right of such a sub-minimal word (probably stress domains in Szigetvári’s (1999) interpretation or feet in more traditional terms): // is well formed (note that this account does not attempt to make a difference between secondary and tertiary stress at this point because the issue is irrelevant), but // is not, where // is taken to represent some subminimal phonological material. Actually, not even // or // would escape the filtering effect of the minimal-word constraint. It seems then that in MoE words are scanned for the constraint beginning with the right edge of the word. If the minimal-word template is satisfied, the word is parsed as well-formed, if not, irrespective of how well-satisfied the template would ‘potentially’ be with the addition of further feet (or stress domains in VC phonology where a stress domain is understood to start with a stressed vowel extending up to, but not including, the next stressed vowel), the word remains subminimal. The hypothetical words // and // are well-formed, of course. Languages do not count, it seems, and as a consequence of this The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 26

MoE also looks only for the satisfaction of the minimal right-edge template. In other words, there may be as many CV/VC pairs in a word as one could wish but without the satisfaction of the right-edge template the word would still be ungrammatical. OE, from this perspective, does no better than MoE: if the template is satisfied no further need arises for the elimination of other high vowels falling outside the appropriate template. 3.3 ME compensatory lengthening It has already been argued that MEOSL is far from being a general process of OSL in ME. The importance of TRISH has also been questioned. Minkova (1982) offers a solution to MEOSL in terms of the preservation of syllable weight after the word-final schwa was lost. In other words, the mora previously associated with the word-final schwa was re-associated to the stressed vowel and thus made it long. Hayes (1989) also discusses the ME problem and supplies the following representation: (10) ME compensatory lengthening X

X

X

X

X

X

t a l



t a

l

t a

l

In terms of moraic phonology it seems unclear why the mora formerly associated to the lost schwa was not linked to the word-final consonant. The result, on the face of it, would have qualified for a minimal word: tal would have been as well-formed as OE hol ‘hole’, hwæl ‘whale’, schip ‘ship’, etc. Kim (1993) notes several problems with Minkova’s analysis. First, she disagrees with Minkova’s statement that the loss of schwa was simultaneous with compensatory lengthening. Second, she notes, there are dialects of English that preserve a more extensive set of lengthened ME forms. What is more, Minkova’s account of compensatory lengthening relies on the loss of schwa. However, this seems to preclude a compensatory analysis of words like hazel, cradle, etc. Kim (1993) suggests that MEOSL was brought about by the general weakening of vowels to schwa and thus lengthening compensated for the overlight final syllable. This accounts for the absence of lengthening in OE -ig words. Yet, again, MoE <ow> words of ME origin are not mentioned, nor is the rather general process of shortening of the original OE long vowels before the <y> (mǣnig ‘many’) and <ow> (hālga ‘hallow’) type of words. A more in-depth and critical analysis of this and several other The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 27

issues connected to the general process of compensatory lengthening is not possible here (a summary of the various types of compensatory lengthening is provided by Kavitskaya 2002). However, Lahiri & Dresher (1999: 712) note that one of the problems with Kim’s (1993) analysis is that it is unable to account for the so many apparent exceptions to this general compensation in disyllables of the C0VCVC template, i.e. for the distinction between saddle vs. cradle, even vs. heaven, later vs. latter, etc. This cannot have been due to a difference between the quantity or quality of the second syllables, they conclude, and go on to propose an analysis based on analogy and a subsequent levelling of either the short (via TRISH) or the long (via OSL) vowel into the base form. 3.4 A ME template? As we have seen the only source that unfailingly gave long vowels in ME was the OE words of the C0VCV template (OE talu > ME tāl). We have also concluded that there was no lengthening of OE stressed short vowels if the next vowel was // < OE -ig or // (the result of the vocalisation of OE //, for example), i.e. after any of the ME full (non-primary stressed) vowels. Although the details of an analysis that makes reference to a template are still wanting, let us assume that ME acquired a constraint that required ME lexical words to link to a template of the …V C V… (since the details are unknown, the representation is deliberately vague as to whether this template is based on CV or VC units). Let us also assume that this matching of the template onto available phonological material was done from the right edge of the word onto vocalic slots with melodic material. In VC phonology it is assumed that vocalic slots in the skeleton may lexically link to melodic material (this gives the difference between // and //, for example). If nothing links to a V slot and this V is not ‘extinguished’ (e.g. it is not syncopated or is not locked in a burial domain between the members of a coda-onset cluster, for example), it is pronounced as some ‘default’ empty vowel (it is empty because no melody links to it lexically, but it is still pronounced). The pronunciation of this empty vowel may vary from language to language: in English it is assumed to be // and we can also take it to hold for ME. If we accept these assumptions for the sake of argument, let see what happened when this template was matched onto a ME word like tal:

The Even Yearbook 7 (2006), © Attila Starčević

Middle English Quantity Changes – Further Squibs 28

(11) The matching of a …V C V… template onto a ME word of the tale-type C

V

C

V

t

a

l



…V C V…

tāl

beli

folo

medo