2. Interpreting written Middle English 1

Our first concern in LAEME, as it was in LALME, is the recording of text languages and the display of their features in profiles and on regional maps.2 This has to be done in two stages. The first order taxonomy is orthographic: the primary written evidence of the manuscripts. But the creation of insightful interpretative maps (such as the LALME Dot Maps) involves consideration of one major second order property — phonetic substance. In LAEME (like LALME) we present first order Item Maps and implicitly second order Feature Maps; but for reasons that will become apparent below we also present maps explicitly indicating reconstructed phonetic substance. This reconstruction is particularly necessary for smoothing out the surface ‘nubbliness’ of the early Middle English continuum (Laing and Lass 2003, and see below §2.3.3).

Early Middle English writing is often extremely complex and difficult to interpret. It is above all not uniform. Like late Middle English, it varies from region to region and from scribe to scribe. Much more commonly than in late Middle English, early Middle English usage may also vary from text to text copied by the same scribe and (depending on textual histories) also from portion to portion of the same text (cf. §1.4 esp. n. 18).

The main reason for this complexity is the contingent rupture of the tradition of writing English that occurred after the Norman Conquest (cf. §1.2). A set of ‘text communities’ that had been accustomed to producing a large portion of its official and other documentation in its own vernacular was tipped into a new praxis. For a century English was generally not used as a written medium except for some copying of Old English legal documents into monastic registers and cartularies. In some centres Old English religious texts also continued to be copied. Although these ‘transitional’ texts show some orthographic developments that may well reflect changes in the contemporary language of the copyists, for the most part they are still recognisably Old English. We have no evidence, until the mid to late 12th century, of what could be called spontaneously produced up-to-date written English. Judging from the evidence of the written English that begins to appear post 1150, in the years after the Conquest there must have developed a kind of diglossia. Spoken English would have been used by the majority in the normal way, and continued to vary and change, differently in different regions, like any natural language; but written expression was nearly exclusively in Latin and French (the language of the conquerors), and scribal employment would have depended on mastery of these traditions. This is not a particularly unusual situation: there are many present-day cultures in which written and spoken discourse are in different languages (Arabic and Tamil are familiar examples). But the immediate post-Conquest English situation is rather different; in Arabic and Tamil the written language is an archaic and ‘classical’ (if not always directly ancestral) form of the current spoken language: Classical Arabic and written Tamil are still Arabic and Tamil.

By the early Middle English period, four generations after the Conquest, the high prestige languages in the written/spoken diglossia (though distantly cognate) were ‘foreign’: normally second and third languages, formally taught and learned. By the late 12th century it is virtually certain — except in the case of scribes from continuingly bilingual families (and we do not have the information that would allow us to identify them) — that none of the English scribes writing French were native speakers of French (though some may of course have been coordinate bilinguals).3 It is certain that none were native speakers of Latin, since after the genesis of the Romance vernaculars in the early centuries of this era it is most unlikely that there were any.

This means in effect that when the writing of the native vernacular was revived in the 12th century, the hiatus made it necessary for scribes to design new orthographies to represent the results of a century of massive and transformative change at the phonological level, which is the primary (though not the only) input to orthography.

Since English was not ‘classical’ or institutionalised it did not require fixed spelling: so different styles of orthographic design were able to flourish. And since the one-word/one spelling mode of most modern standards was not the norm, natural phonological and morphological variation was not prevented from surfacing in written forms. Early Middle English texts display a wide range of representational strategies. In addition to simple ‘phoneme’-to-‘grapheme’ mapping,4 we find logographic and morphographic writing, as well as litteral substitution. For these terms see §§2.2 and 2.3 below.

2.2.1 Levels of representation

A spelling system is a mapping of some chosen set (or sets) of linguistic units into a set of visual signs.5 The standard inventory of linguistic units is the word, the morphophonemic representation, the syllable6 and the ‘phoneme’. In relatively rare cases allophones of certain phonemes may be represented (e.g. the velar nasal in the elder Futhark, Gothic and Greek).

Modern and many ancient alphabetic scripts tend to represent language at what could be loosely called the phonemic level: a linear string of graphs is a rough icon for a string of structuralist phonemes of the sort one could arrive at by commutation — minimal pair tests or similar procedures. But trying to reverse-engineer historically bequeathed alphabets as redundancy-free and univocal systems of representation would generally disappoint us. This is true of early Middle English more than many other languages, for reasons we will discuss below. If spelling systems were designed that way, an ‘ideal’ alphabet would represent biuniquely: one grapheme per phoneme and vice versa. Such systems in fact are rare, and highly unlikely in languages with long histories and little spelling reform. Most orthographies carry considerable historical baggage: e.g. English ‘silent’ final <-e>, <gh> in eight, night, through, and <kn-, gn> in know, gnaw. They also tend to have much non-biunique representation (/ʃ/ in shoe, vicious, ocean, nation, passion, chic, schist ...). This may be exacerbated by intensive contact with other orthographies.7

Not all alphabetic systems are pure; many also use other types of representation. We find such mixed systems throughout Germanic. The commonest of the not strictly alphabetic, supraphonemic representational strategies are:

(i) Logography. Non-biunique phoneme/grapheme relations can enable a form of logographic writing: the spelling indicates not only a phonological string, but one particular member of a set of homophones: E right, wright, rite, Afrikaans ys ‘ice’, eis ‘claim’, both /ɛis/. These distinctions often reflect historical origin: right < OE riht, wright < OE wyrhta, rite < OF rite; but sometimes they do not, as in E deer, dear < OE dēor, dēore. We do not use ‘logographic’ in the way that ‘ideographic’ was used in older descriptions of Chinese: 1,2, @, & are not logographs, but rather icons or pictographs. In our definition, there are two main types of logography: (a) the discrimination of homophones in such a way as not to violate the graphotactic/representational rules of the system, e.g. dear/deer; (b) the consistent assignment of particular spellings to particular words where other spellings would, according to the structure of the system, be allowable, e.g. modern English bright with medial <gh>, where *brite would be equally well-formed.8

There is a cline, at one end of which are ‘genuine’ spellings of words (including logographs in the sense defined above, like dear/deer), and at the other are pure icons. In between is a domain that is of particular interest when dealing with medieval texts: abbreviation. Even abbreviated lexical items constitute a cline, because some retain more phonological clues than others. If deer/dear are ‘full’ logographs (since a pronunciation can be inferred from them according to the usual orthographic rules of the language), a further stage toward abstractness on the cline is something like barred thorn <> indicating that.9 How the vocalism of this ideal ‘that’ might be realised phonetically is undecidable. So the abbreviation means that lexically, but no particular form phonologically, except with respect to the initial consonant.10 The crossbar has no phonological import, but the thorn limits the range of representations to those beginning with [θ~ð]:11 <> is not merely a trigger. The same could be said for <S~> for saint. These might be called ‘partial’ or ‘impure’ logographs. There is also the occasional use in religious texts of initial letters only to stand for well known or much repeated word sequences (e.g. in Oxford, Merton College 248, for repeated quotations from the Pater Noster: ore ylk \ d. b. y. g. vs, ‘our each day’s bread thou give us’). At the other end of the cline are objects like ampersand <&> and the Tironian sign <> for and, or <xpc> for Christ, and <iħc> for Jesus. Even though these last two examples are made of letters, not abstract symbols,12 like <&> and <> they are not subject to phonological extrapolation. All these objects are mere triggers, i.e. their reference is non-phonologically conventional, outside the surrounding alphabetical schema. These are what we call ‘icons’.

(ii) Morphophonemic writing (‘morphography’). While alphabetic praxis generally represents at phoneme level, some languages represent morphophonemically or ‘abstractly’ as well. A systematic example is the writing of final obstruents in continental WGmc (except Yiddish). German, Dutch, Frisian and Afrikaans have final devoicing, but do not systematically indicate this in spelling, particularly for stops. Thus G Bund ‘league’, pl Bünde /bʊnt, byndə/, but bunt ‘colourful’, inflected bunte /bʊnt, bʊntə/.13 So, the final <d> in Bund (where /d/ according to surface phonotactics is ‘unpronounceable’) is a signal that /d/ will appear if a vowel-initial suffix follows. The writing <-d> then is indexical: it marks both morpheme identity and the existence of an alternation.14

Old English used similar abstract strategies: for instance, fricatives were voiceless initially and finally but voiced foot-medially between voiced sounds, so that a word like wulf wolf [wulf] would have the nom/acc plural wulfas [wulvas] with <f> written in both positions regardless of phonetic value (as opposed to ‘concrete’ ModE wolf/wolves where the voicing is indicated). This practice was maintained to some degree in early Middle English, though the preferential spelling mode appears to have been concrete or phonemic (as far as this term applies: see §2.3).15 Here are examples of both types of spelling from one early Middle English writer, the Worcester Tremulous Scribe (first half of the 13th century), illustrated by forms of the reflexive suffix -self (OE seolf/sylf). The fricative in absolute final position is always written <f>, e.g. sulf:

Concrete: final <f> ~ medial <u>16

P13XM he-sulf

P23<prX ham-suluen

P23X hi-sulue

Abstract: final and medial <f>

P13XM he-sulf

P11OiX [m]e-sulfum

P12GX þines-sulfes

(iii) Use of diacritics. Many Germanic scripts are rich in devices of this kind, e.g. letters marked with accents or diereses, and various superscript and subscript symbols. Some languages also use doubling of vowel or consonant letters for diacritic purposes.

In early Middle English, there are three main types of sporadic, and for the most part unsystematic, diacritic usage: (a) doubling of consonants to indicate that the preceding vowel is short; (b) doubling of vowels to indicate length; (c) the use of accents on vowels to indicate their quantity.17

(a) The doubling of a consonant to indicate a preceding short vowel is widespread18 though not usually at all regular. A frequent use is to disambiguate words that would otherwise be homographs, e.g. God and good (godd vs. god(e)). It is not a consistently applied practice, however. In most scribal systems short vowels before non-geminates are usually followed by single consonant graphs, especially where the spelling gives rise to no ambiguity. Some writing systems have double letters or digraphs like <ck> even after apparently long vowels, e.g. bock for book in both Laȝamon A, hand A and Laȝamon B. Orm is apparently the only author to have developed an explicit system of diacritic consonant doubling.

(b) Doubling of vowels to indicate length also occurs only sporadically: e,g, deed death < OE dēaþ in British Library, Royal 8.F.ii. Surprisingly, however, double vowel graphs may also appear for historical short vowels, even those followed by double consonants or consonant clusters (e.g. -seelf -self in Oxford, Bodleian Library, Laud Misc. 108, hand A). A further complication arises when a double vowel graph, apparently indicating a long vowel, is nevertheless followed by a double consonant graph, which would normally indicate a preceding short vowel (e.g. soðfastheedd soothfasthood in the short verse texts in British Library, Arundel 292).

(c) Accents on vowels to indicate quantity are also sporadic. We do not count as an accent the very frequent oblique stroke (like an acute accent) on <i>. It is a device for disambiguating sequences of minim strokes, helpful for all scripts, whether formal or informal. Although it is rarely employed with complete regularity, most scribes use it at least some of the time. The stroke (or dot) may appear on both historically short and long <i> (as well as on <y> and sometimes also on wynn and thorn). Where either the stroke or the dot appears in these contexts, we therefore take it to be integral to the shape of the letter, the equivalent of the dot on modern printed <i> and <j>, rather than having diacritic significance. Much less common is an acute accent on any other vowel but <i>/<y>, but when it occurs it apparently signifies vowel length. Sometimes the accent seems to have an additional function: to disambiguate a single <a> as a content word — e.g. the indefinite article or number one, or the word for ever — from <a> representing the preposition in. Even Orm does not use a consistent accent system. In his text, vowels may be marked by one, two or three acute accents. All three seem normally to indicate vowel length. Occasionally, he uses a breve above a vowel to indicate shortness, normally (but not consistently) to disambiguate pairs of homographs, e.g. lăte late and láte manner.

2.2.2 What readers have to know

A spelling system is a mnemonic for native speakers.19 Readers of any language at any period have to be able to cope with redundancy and historicity. The history of English of all periods shows that it is possible for spellings to be intelligible even when they belong to systems that include considerable merger or redundancy. There appears to be no general rule for how much of a language’s system actually has to be represented. Middle English generally did not distinguish between its two heights of long mid vowels, using <e(e)> and <o(o)> indifferently for both high mid and low mid. Though the common alternation in modern standard English between <ee> and <ea> (greet vs great) and between <oo> and <oa> (brood vs broad) goes back to the use in some Middle English dialects of <a> as a diacritic for the lower of each pair.20

But a literate native speaker can routinely understand spellings that may seem strikingly ‘defective’.21 All spelling systems have built-in redundancy, and interpretation of even bizarre spellings is possible as long as the reader knows the system and has a good idea in advance of what a word is likely to be, or what the range of choices is. In the present context, no reader who knows English would have any difficulty reconstituting the defective representations <spllng> or <rthgrphc>.

There are, however, complications. If an orthography lasts long enough it will tend to represent ‘ghost contrasts’ due to sound change not indicated by spelling change: e.g. for PDE, except for some Northern Scots, <kn-> vs <n->, and for many dialects <wh-> vs <w->. This kind of purely orthographic pseudo-contrast is generally removable only by deliberate spelling reform. A segment lost in isolation may remain as a diacritic or alternation index: e.g. in English non-rhotic dialects postvocalic <r> is a marker of length and sometimes quality for a preceding vowel; and final <-r> is only a tag for recovering the lost /r/ in external sandhi (law, lore /lɔ:/). It surfaces phonetically in lore when it is followed by a vowel, but not in law except in dialects with ‘intrusive r’. In whatever mode of representation, there is no a priori reason to expect consistency: e.g. English and German are close to ‘proper alphabetic’ systems (representing mostly at classical phonemic level), but with a fair number of logographs, morphophonemic writings and ‘unexpected’ spellings.

If a spelling system is a mnemonic for native speakers, as historians we have no right to expect systems that cohere with our modern European standard-language ideology of ‘good’ spelling practice, or with orthographic models derived from particular formal linguistic theories. Almost all the early Middle English scribes whose work survives were clerics or other institutionally trained writers. We can assume that the spellings they employed were interpretable to their readers. The systems they designed or adopted show differing degrees of internal variation and structural flexibility. Our task is to develop a hermeneutic that provides (as far as possible non-anachronistically) an interpretation for the work of all sane scribes.

2.3.1 The doctrine of littera

Up to this point we have been using standard terminology: ‘phoneme’, ‘grapheme’, etc., and representing these theoretical objects with the usual bracketing. We depart now from this framework, for reasons which will become evident. The most important of these is that such concepts do not always characterise what our scribes appear to be doing. They are frequently not ‘structuralists’, and it seems to us better to use a theoretical framework and notation that cohere more closely with what scribes would have experienced in their education — though we will take considerable liberties in exposition.

One work which probably all scribes would have been exposed to in the course of their training as the indispensable foundation of a medieval orthographic education, would have been the Ars maior of Aelius Donatus (fl. 5th century AD). In book I is a statement that can be taken as canonical:

Here is slightly exegetical translation:22

The littera is clearly an abstract object, which under the classical interpretation is a member of a universal phonetic alphabet. Each littera in this view has only one possible potestas. Thus Quintilian (Institutiones I), commenting on some deficiencies of the Latin writing system, notes that certain ‘necessariae litterarum’ present in Greek are lacking in Latin, and that for writing e.g. seruus and uulgus ‘Aeolicam digammon desideratur’ (‘the Aeolic digamma = [w] is wanted’). This aspect of the theory of littera (which is notable in the English tradition in John Hart’s 1569 discussions of the ‘abuse’ of letters, and appears as late as Wallis 1653) will not be treated here; Middle English scribes were not generally concerned with litterae in this sense; indeed our interpretive task would be much easier if they had been.

The stream of litterae in writing is represented by a sequence of figurae; indeed this is the way the littera becomes visible. Discourse about the detail of figurae is most often the preserve of palaeographical interpretation. For discourse about the interpretation of spellings once established by a palaeographical reading the term littera may rather used. For instance, it does not matter for the interpretation of a spelling whether a scribe uses a short <s>, a long <s>, a sigma-shaped <s> or a bean-shaped <s>. Certain types of script dictate the use of certain types of figurae in certain contexts: e.g. in Textura script, the use of 2-shaped <r> after a letter with a rightward facing bow (<o>, <b>, <p>). Except where the concepts littera and figura overlap, (e.g. when the litterae thorn and wynn are realised by identical figurae) at the level of system we therefore talk about a littera as the superordinate for all the different possible figurae that different scribes, or any one scribe, may adopt for it.

For notation we will follow the conventions developed by Michael Benskin (1997: 1 n. 1 and 2001: 194 n. 4) and used by us in Laing and Lass (2003). We put litterae in inverted commas, the figurae that occur in particular manuscript systems in angled brackets (not to be interpreted as ‘graphemes’), and potestates in square brackets: so ‘e’, <e>, [e].23 Glosses and the names of lexical items will be in small capitals. Standard citation forms, manuscript forms without specific litteral reference, etymological categories and reconstructions will be in ‘uninformative’ italics.

2.3.2 Substitution sets

On one parameter, Middle English spelling systems can be loosely classified into two types (Laing 1999, Laing and Lass 2003): ‘economical’ and ‘prodigal’. An economical system makes some approach towards the ideal of one littera one potestas; a prodigal system allows considerable multivocal relationship. These notions of course are relative: all systems are somewhere on a cline between the two, though the two ends of the cline can be easily recognised. And even the most prodigal systems may be economical in some particulars. For instance, we do not know of any system that uses ‘p’ in a strikingly multivocal way; but we know of many that use ‘eo’ for a very large number of potestates (Lass and Laing 2005). Similarly we do not know of any early Middle English systems with a plethora of representations for the potestas [i],24 but many with large substitution-sets for [x] (see below).

We can illustrate the flavour of the two system types by looking at the spellings of OE -ht (as in night, thought and not < OE noht)25 in four hands in the same manuscript: Cambridge, Trinity College (323) (hereafter Trinity) (SW Midlands, second half of the 13th century). Here are the patterns for the four main hands that contribute text in English, with frequencies of different spellings:

Hand A: ‘st’ 89, ‘t’ 55, ‘tt’ 4, ‘cst’ 3, ‘ct’ 2, ‘th’ 2, ‘chit’ 1, ‘cht’ 1, ‘cðth’ 1, ‘sþ’ 1, ‘th’ 1, ‘thth’ 1, ‘tth’ 1

Hand B: ‘st’ 90, ‘t’ 21, ‘tt’ 3 ‘d’ 1

Hand C: ‘t’ 19, ‘tt’ 3

Hand D: ‘t’ 11, ‘st’ 10, ‘cht’ 8, ‘ch’ 4, ‘ct’ 3, ‘d’ 2, ‘th’ 2, ‘tht’ 2, ‘ȝt’ 2, ‘dt’ 1, ‘tf’ 1 ‘tt’ 1

It is clear that Hands B and C are (at least relatively) systematically economical in their approach to words containing this historical sequence; Hands A and D are profligate.

A set of litterae in variation for the same potestas or etymological category we call a Litteral Substitution Set (LSS: cf. Laing 1999). Thus Hand C has for OE -ht the LSS {‘t’, ‘tt’}, Hand B has the LSS {‘st’, ‘t’, ‘tt’ ‘d’}, Hand D has the LSS {‘t’, ‘st’, ‘cht’, ‘ch’, ‘ct’, ‘d’, ‘th’, ‘tht’, ‘ȝt’, ‘dt’, ‘tf’, ‘tt’} and Hand A has the LSS {‘st’, ‘t’, ‘tt’, ‘cst’, ‘ct’, ‘th’, ‘chit’, ‘cht’, ‘cðth’, ‘sþ’, ‘th’, ‘thth’, ‘tth’}. The inverse of an LSS we call a Potestatic Substitution Set26 (PSS: Laing and Lass 2003: 262–263). So in Trinity Hand D the littera ‘ȝ’ maps to the PSS {[h], [x], [j], [w], [ɣ]}. For instance [h]: ȝu how < OE hū; [x]27 driȝten lord < OE dryhten; [j]: ȝe ye < OE gē; [w]: roȝen to row < OE rōwan; [ɣ]: daȝes days < OE dagas.28 A system that is prodigal in one direction is likely to be so in the other — prodigality is a fundamental design style.29

Within the framework of the ‘doctrine of littera’ this substitutive praxis is not ‘classical’ but revisionist. According to Donatus’ definition, each littera has a potestas, as inseparable from it as its name and shape (accidunt cuique littera tria ...). Just as the potestas is a local property (‘accident’) of the littera, so each littera would seem to be appropriately connected with just one potestas; at least nothing in the text appears to grant a licence for multiple representation. Certainly according to Quintilian — and we would imagine, three centuries later, to Donatus — a foundational principle of the doctrine was the univocal binding of figura and potestas in a single universal unit. So allowing one potestas to be represented by multiple litterae would be a violation. This does not vitiate our use of the doctrine of littera as a hermeneutic device; but it suggests that the medieval notion is different in major ways from the late antique one.

2.3.3 Why potestatic interpretation is necessary

Since all our Middle English witnesses are written texts, why should we attempt to assign potestates to litterae at all? It would seem on the face of it that the obvious strategy would be to map litteral representations, and construct an orthographic dialect geography and history of early Middle English. This would appear to be a ‘safe’ strategy, as it would require minimal use of inference, and give an accurate picture of the distribution of forms.

But would it? The very possibility of litteral substitution raises some serious problems, as does the proposed restriction to spelling. Since we are interested in a language, and spellings are spellings of words and morphemes, it would be arbitrary and unnecessarily constraining (and information-destroying) not to consider also the ‘reality’ or phonetic substance that underlies the spellings. Language is phonic before it is orthographical; and without some substance there would be nothing for the LSSs to represent.30 And once the notion of littera is accepted, potestas (except in the special case discussed below in §2.3.5) is indissolubly bound to it. In fact as we will see even the simplest taxonomic judgements (e.g. the grouping of spelling types) depend ultimately on phonetic judgements.

The statement in LALME (vol. 1, 6) that the maps constitute ‘a dialect atlas of written Middle English’, and that texts are ‘treated as examples of a system of written language in its own right’ is often misinterpreted. The emphasis on the independent value of written evidence was particularly apposite two decades ago, given the post-Bloomfieldian view that was current then (and to a large extent still is) that writing is of no independent linguistic interest, but merely ‘parasitic on’ speech. But this must not be misunderstood and taken to imply that phonological interpretation is per se unnecessary. The LALME editors take no such line. They were fully aware of the potential phonological implications of their data. LALME is rich in phonological commentary, while the series of Dot Maps (vol. 1) crucially depends on acknowledging the relationship between sound and symbol.

The history of a language cannot be restricted to its orthography. We take spelling with the utmost seriousness, in no sense ‘merely’ as indicative of phonology; but we take phonology equally seriously. LAEME is not an atlas of early Middle English orthographic forms, but an atlas (like LALME) of both first-order data and the second-order but equally important information deducible, or otherwise arguable, from it. The history it portrays is that of orthography, phonology and their interactions. The inclusion of both is logically necessary: we could not assume that any two orthographic objects represented ‘the same’ item unless we assumed a system-level rather than utterance-level entity lying behind them, and tying them together as sames (see also §2.4.2).

But beyond these general linguistic considerations are specifically dialectological ones. In the LAEME materials we find frequent surface irregularity and lack of directional or graded variation of spellings across space. There is an apparent mismatch between this and the uniformitarian imperative (except in the face of clear disconfirmatory evidence) that any cluster of adjacent cognate dialects forms a continuum.31 One way of extracting the expected underlying continuum from the apparent scatter of divergent forms is by reducing the ‘free’ distribution of many spellings to a smaller number of types, and then tracking these across space.

Consider for instance a sample of the attested spellings for such in the LAEME corpus: swylk, swilke, swilk, suylk, suilke, sulk, swiche, suich, swyche, sweche, swoche, suche, soche, soch, sich, seche. Are these all ‘individuals’, or do they reduce to a smaller set of ‘types’? Entirely by eye, invoking no phonetic intuitions they could be reduced to a certain extent: -ch- vs -lk- and sw- vs su- vs so- vs si- vs se. But the maximally parsimonious sorting will also invoke assumptions about what the spellings represent phonetically. E.g. it seems reasonable to take the sw-, su- forms as beginning with [sw] and all those with initial s- + V as not having a following [w]. It is reasonable to assume that those ending in -k have a final velar (which typically correlates with initial sw-), and those in -ch have a final palatal. In effect we can subdivide the attested material into more abstract types: swVch vs swVlk as against sVch. Further, we can group nuclear i, y together, and e, o, u as constituting different groups. We can take nuclear u, o as being more closely related by virtue of their being back vowels, though given the presence of an LSS {‘u’, ‘o’} in some text languages they may both represent [u]. By any criteria, e would be something quite different. It is these phonetically based representations that serve to identify the continuum, which tends to be occluded if the set of attested spellings is too rich. The richness may be made more tractable by subdivision according to phonetic as well as orthographic criteria.

Potestas then is an ‘accident’ of littera only in the technical Latin grammatical sense ‘something that belongs to’ it (cf. Lewis and Short 1879 s.v. ac-cidō B.3); it is an essential and inseparable part of the concept and its interpretation.

Since spelling ultimately has some relation, however indirect, with sound substance, considerations of litteral substitution often raise potestatic questions. Some of the choices in one of the Trinity hands illustrated in §2.3.2 above will make this clear, and suggest the kinds of knowledge that have to be brought to bear on matters of litteral interpretation. First we should explain ‘st’ as a representation for historical -ht (expected synchronic [xt]). This is a choice found, beside other representations, in three of the hands, and it might seem phonologically odd. The compound littera32 ‘st’ is also used in this text for historical -st (e.g. in beste best); its use for -ht however does not reflect a change [xt] > [st] in English, but is an inverse spelling based on a French change [st] > [xt ~ çt ~ ht] (Pope 1934: §§1178(ii), 1216).33

There is also nothing problematic about ‘ct’, ‘cht’, or ‘ȝt’: they are to be read in the first instance as some obstruent before [t], and by various historical arguments (see §2.4.1) as more precisely something in the palatal or velar region. The litteral sequences ‘ct’ and ‘cht’ have a history going back to early Old English, and form part of a series that later resolved primarily to ‘ht’, which indeed remained in many early Middle English text languages. But other substitutions are not that straightforward, and may have potestatic implications as well. Difficulties arise here with ‘t’, ‘tt’, ‘th’ and ‘tht’. The following are the reflex-sets of four Old English items in Trinity Hand D (tagging is omitted as only the root phonology is at issue; capitalisation is as in the manuscript):

ǣht property: Acte 1, hachte 1, achte 2, haite 1

dryhten lord: drichen 1, dristin 2, driȝten1, Drittin 1

cniht knight: cniches 1, cnith 1, cnit 1, Cnites 1

liht light: litht 1

We group ‘t’ and ‘tt’ together, since they appear to show absence of the reflex of OE h: at least we assume they do because this fricative was lost eventually, and these spellings could easily be interpreted as early variants showing loss. That is, while ‘ct’ and ‘cht’ may be interpreted as [xt],34 the general usage of this writer (and others), and the history, suggest that ‘t’ or ‘tt’ are better taken as lacking the preceding fricative.35 The other problematic cases are ‘th’ and ‘tht’. On the face of it, they might appear to represent [θ(t)]; is there evidence for this? There is no doubt that -ht can become [θ]; this is attested in modern Scots at least.36 So the development itself is not outlandish, though its current regional restriction makes it somewhat problematic to attribute it to an early Middle English SW Midlands text. But an examination of Hand D’s system of writing makes the interpretation of ‘th’ as [θ] more doubtful. There are six examples of ‘th’ in this text, all representing historic -t- (sitthest sittest < sittest), or -ht (cnith knight < cniht, mitht might < miht, litht light < liht, noth not = nought < nāwiht > nāht, sothede sot-hood, folly < OF sot). There are no initial ‘th’ for OE þ (hand D’s system has ‘þ’ throughout); and postvocalically in stressed syllables there are 20 examples of ‘þ’. So from a systematic point of view it does not look as if ‘th’ is a likely writing for [θ] in this text. This emphasises the important point that spellings cannot be looked at in isolation: one must consider the system employed in a particular text language before making interpretive judgements.37

2.3.5 Zero

It should be clear now that our interpretation of the litteral praxis of our scribes is not intended to be purely synchronic. What we take as the content or import of litterae is often etymological. This is because it is often not possible to produce the kind of synchronic analyses of our text languages that would allow for an implementation of the Saussurean dichotomy — even if we considered it desirable (see the discussions in Lass 1997: chapter 1, Williamson 2004). The most convenient descriptions of the import of particular litterae are often historical: we do not know precisely what ‘cht’ in Trinity D means synchronically, but ‘OE -ht’ is a useful and accurate label for it, and one that is relevant to the entire subject matter of LAEME.

One case where this essentially historical analysis becomes particularly relevant is in dealing with ‘zero elements’; it is not really possible to characterise these intelligibly except on historical grounds. Zero is not an element in classical theory of littera; we introduce it here because both zero litterae and zero potestates are well attested in our materials, and their use was familiar and apparently unproblematic for our scribes.

A simple case of zero potestas is ‘excrescent h’, the writing of ‘h’ in places where history tells us that it should not have had potestatic import.38 Two examples from Trinity Hand D again shown as reflexes of OE forms:

ǣht property: Acte 1, hachte 1, achte 2, haite 1

ic I: ich 7, ic 5, Hi 1 Hic 1, I, 1, hich 1, y 1

For the category OE word-initial vowel (#V-) the LSS is then {0, ‘h’}, but the PSS may very probably be {0}. The reason we can infer this is the existence of its inverse, e.g.

hæbbe have: abbe 2, haue 1

hit it: hit 8, it 7, hid 1

In these cases the ‘h’ is etymological, so that forms without it represent what we could call a zero littera. The LSS for OE h- then is {0, ‘h’}, and the PSS is {0, [h]}. Of course in one sense the use of a littera with zero potestas as above is simply what is known as ‘inverse spelling’; the reason we want to rename it for the moment is to make it coherent with the rest of what appears to be a reasonable model for scribal praxis.

2.3.6 Substance and structure: why we choose littera as our basic unit

The theory of littera enables us to bypass a question that for our purposes is vexatious rather than informative: whether we intend our representations to be ‘phonemic’, ‘phonetic’, ‘abstract’ or anything else within contemporary metalanguage. This issue is also avoided by the handbook tradition with its ‘uninformative’ italics.

Of course, we do not know in precise phonetic detail what any historical spellings represent; but in terms of the parameters of LAEME, this question is largely irrelevant. A representation of a Middle English form in [ ] is an indication of phonetic substance at some level, but not of status within a system in any structuralist (including generativist) sense. As indicated above, systemic status rather is relevant within the confines of specific text languages. We disavow any attempt to set up an overall pattern or system for ‘Middle English in general’. In the first place we do not think it can be done, and in the second our aims are comparative and variationist, not reductive and generalist. As dialectologists we are interested in description and comparison and as linguistic historians we have to work within the constraints of the data that survive. That does not mean that we eschew interpretation, extrapolation or theoretical argumentation. Simply, there is no need to make a commitment to any particular theory of system structure in an endeavour such as LAEME. LAEME provides the tools to test and to create such theories. The option for producing a structural analysis of any kind is always open to the user. Indeed, if such analyses turn out to be possible or useful, the corpus provides precisely the kind of material one could use to make them.39

In LAEME, we are concerned with word- or affix-histories and distributions.40 Whether a given symbol is to be taken as representing something ‘emic’ or ‘etic’ is a consideration that belongs to a different kind of discourse. The previous discussion of substitution has already indicated why formal structural analysis based on distinctiveness is not appropriate for application to most of the text languages in the LAEME corpus. Many of these languages, as the previous discussion of substitution suggests, have writing systems that display two distinctly non-structuralist properties: (a) their creators are not particularly concerned with biunique grapheme/phoneme mapping; and (b) the orthographies (and frequently the languages they represent) are highly variable. The evidence for structuralist ‘eme’ systems is too weak to make it sensible for us to use this kind of modelling. Distinctiveness in the usual sense does not seem to be a high priority, which makes techniques like commutation nearly useless.

A case in point is Trinity Hand A. For taxonomic convenience we set up a typological inventory (based on the handbook accounts of ‘the Middle English sound system at large’) from which the potestates represented by any early Middle English orthography would be likely to have been drawn:41

Here we differentiate between ‘inventory’ as the set of available sound types and ‘system’ as the deployment of a subset of these types in the phonology of a language.

The total inventory of litterae that occur in Hand A’s output in what we assume are accented syllables is:

‘a’, ‘e’, ‘i’, ‘o’, ‘u’

‘ai’, ‘au’, ‘ay’, ‘ea’, ‘ei’, ‘eo’, ‘eu’, ‘ey’, ‘ie’, ‘oe’, ‘oi’, ‘oo’, ‘ou’, ‘ui’

These nineteen litterae happen to be the same numerically as the number of vowel types one would expect in a maximal system; but the arrangement seems odd. We might expect 12 monophthongs and 7 diphthongs. However, if (as is usual in Middle English vowel orthographies) length is not represented and ‘open’ and ‘close’ long mid vowels are not distinguished, the 12 monophthongs would be represented by just the first five litterae listed in our inventory.42 We then have an expectation of 7 possible diphthongs with 15 digraphs actually occurring in Hand A’s output. The problem then is matching all 19 vowel-type representations to the likely potestatic inventory. Some cases are relatively simple. E.g. ‘i’ appears in many forms where we would expect it, like (h)ic I, min(e) my/mine. On the basis of standard protocols (see §2.4.1 below) we can assign the potestates [i], [i:] respectively.43 And in general throughout this text wherever we find reflexes of OE i or ī they will be represented by ‘i’.

But now consider the representation of stressed nuclei in the following items from this text language:

LSSs containing both simplex litterae and digraphs are frequent in this text: e.g. the 3rd person plural genitive pronoun their is represented with nuclear {‘e’, ‘ei’, ‘eo’, ‘uo’}. If we take one member of this set — ‘ei’, historical evidence suggests that there should be at least two or perhaps three potestates associated with this littera: [ai] as in awei away, [e:] as in feit feet, and whatever ‘ei’ might be presumed to stand for in unstressed syllables, as in afteir after (see note 53 below). It is therefore clear that digraphs need not represent diphthongs (though they may); and that there is no way that sense can be made of a system like this from the inside except (a) historically, and (b) in terms of litteral substitution. The scribe’s eschewing of biuniqueness makes a standard synchronic phonemic analysis impossible: there are too many overlaps and failures to distinguish what we know independently to be etymologically distinct and unmerged categories for such an analysis. Hand A’s orthographic system has a number of LSSs for the nuclear vowels of particular lexemes, and only that; assignment of potestatic representations is at least partially an etymological act.

Such orthographies are not based on ‘structural’ analyses of a lect (quite unlike the First Grammarian’s earlier analysis of Icelandic, which is essentially based on minimal pairs). Rather they are complex systems whose inventors apparently lack certain interests that they ‘ought to have’. Hand A writes a language that, with some knowledge of both prior and subsequent linguistic history, we can interpret perfectly well. But it is not clear on orthographic evidence, which is our primary data, what the internal structure of its sound system is.47 We cannot know from the evidence of the LSSs themselves whether any two or more etymological categories have merged or have remained separate. It should be clear at this point why we are interested in the histories of forms and of individual systems, not in an ‘overall system’.

Before we produce a more detailed model of litteral praxis, it is important to scrutinise the theoretical framework that enables us to assign potestatic values to written materials from the past.

2.4.1 Protocols for potestatic interpretation

How do we assign a credible phonetic value to an orthographic string from our period? We have so far taken the ‘canonical’ handbook representations shown in the inventories above as unproblematical. But the principles involved are not without difficulties, and these deserve to be made explicit. We will take as an example a token spelled niht and clearly meaning night in an early Middle English text. What are our sources of information as to its likely phonetic representation?

(i) Comparative evidence from ancient Indo-European. Forms like L noct-, Gr núkt-, etc. imply (given our background knowledge of Indo-European comparative linguistics) a distant reconstruction *nokt-. This already suggests that the shape of the root in our form ought to be CVCC. We can also use descriptive phonetic evidence from those ancient languages that have such a tradition, e.g. Greek and Latin (cf. Allen 1965: 14–16). There is comparative evidence from modern Gmc as well: G Nacht, Du nacht cohere with the suggested ancient Indo-European forms having undergone spirantisation of their penultimate consonant. Sw natt [nat:], ModE night [nait] suggest further major changes: here deletion of the segment represented as [k] in Latin and [x] in German, and compensatory lengthening of the final consonant in Swedish, and lengthening and diphthongisation of the root vowel in English.

(ii) The history of forms within Old English. We find among others naeht, neaht and niht; further exploration of Old English sound changes suggests that the original Germanic vowel was *a, with later changes involving fronting to *æ, ‘breaking’ to ea, and raising eventually to i, giving a high vowel in English whereas all other Germanic dialects have a low vowel.

(iii) Later developments in English. Spelling, rhyme practice and phonetic description from the 16th–17th centuries suggest coexistence of types with medial [x], [ç] or [h] (generally preceded by short vowels), and forms with a long vowel but no consonant before the final [t] (see Lass 1999a: 116–18 for details). We have here a common type of trajectory, in which a consonant weakens from stop (Indo-European) to oral fricative, to [h], and then to zero.

(iv) Modern dialect evidence. Forms of ‘night’ with a nonanterior fricative exist in Scots, e.g. [nixt]; unless we want to propose something as odd as late [x]-epenthesis, which the historical evidence in any case does not support, we must interpret this as a survival.

Historical reasoning is typically not ‘linear’ but ‘reticulate’. We make claims on the basis of convergence or ‘consilience’ of many different arguments from different temporal strata and theoretical positions. In the case above we use evidence from ancient forms, subsequent forms, and the period under investigation. This argument has been a particularly simple one. Characterisation of vowels, for instance, may be more complex, involving detailed comparative reconstruction, judgements on the range of likely potestates of Latin and Greek litterae, the behaviour of Latin and Greek loans in Germanic and studies of 16th–17th-century orthoepic testimony. We will not go into any further examples: our judgements are based on general consideration of the available evidence and the long tradition of scholarship that has led to consensus.

2.4.2. Level of resolution

Assuming that we need some kind of phonetic representation in order to do our work, the Hard Question remains: how fine should our phonetic resolution be? In our representation of historical categories and potestates we have used conventional IPA symbols broadly, as signs for ‘ranges’ of phonetic quality. This is necessary in historical reconstruction, where we do not have live informants. There are two polar approaches to this problem in the literature. One is the ‘hard structuralist’ assumption that all that counts are systems of oppositions, and that phonetic value means little. The other is a strongly realist view that would have the outputs of reconstruction be something close to ‘pronunciations’. The second may be desirable but is unattainable. The first is untenable, not only for the reasons already demonstrated above, but also for the reason argued below.

What do we actually mean by the phonetic symbols we use? We could be phonetically agnostic and say simply that they stand for points of opposition in a contrast space. This way we assign them nothing but ‘mnemonic’ phonetic value.48 So for mnemonic convenience we might say that Old English had two vowel sets:

and call them ‘front’ vs. ‘back’, but not mean anything more (despite the mnemonic transparency) than if we wrote:

We could call these two sets ‘men’ vs ‘women’, and still capture the insight that there are two groups of entities, and each set has some major feature in common which distinguishes it from the other. And yet as a matter of historical fact there is a well-known major ‘sound change’ (or ‘sex change’) in which Women vowels become Men vowels if followed by Fred in the next syllable. (This is an austere ‘classical’ structuralist version of i-umlaut.) If we contrast this statement with the phonetically specific alternative ‘Back vowels become Front if followed by [i] in the next syllable’, this, along with the claim that every distinct phonological string represents a real word, may be sufficient to show that some phonetic specification is necessary. Without it, history becomes arbitrary and no longer ‘naturalistic’; e.g. assimilation is no more transparent than any arbitrary (even cross-linguistically unattested) change.

Somewhere between [Fred] and [i] (= ‘precisely the value of Cardinal Vowel 1’) is the domain in which we operate in historical reconstruction. In practice we limit both the number of phonetic parameters and the number of points on each parameter to a ‘safe’ minimum. This allows for naturalistic specification at a fineness of detail adequate for our purposes and historically insightful, which is in keeping with what we know of both historical input and present development — as well as with what we know of language history in general. Our experience over decades of work on the history of English suggests that for our period and its predecessors, we can retrieve distinctions of only a limited degree of fineness, but with a certain confidence.

The level of phonetic representation we choose might best be called ‘poorly resolved broad transcription’ (cf. Laing and Lass 2003). This is, we think, the right way to represent most historical sound substance. We operate on the assumption that our reconstructions are well enough supported so that if a responsible phonetician equipped with a time machine were able to hear the items represented, the symbol in question would be a reasonable transcriptional response. This is partly standard wishful thinking, and partly our positive assessment of the results of work in comparative and historical linguistics over the past two centuries. For instance, we would be very much surprised if what we choose to represent as [me:] (the oblique form of the 1 pers sg pronoun), would have anything other than a labial nasal in first position and a mid front vowel (probably high mid) in second position.49 So each segmental representation is fuzzy; it is equivalent to a penumbral range around the denotational centre of the equivalent IPA symbol. Therefore the use of [ ] in the maps, etymologies and statements of changes is essentially ‘typological’ rather than ‘phonetic’ in any more highly resolved sense. It is a representation of ‘sound substance’ at an unspecified but undoubtedly coarse level.

What implications does the theory of littera have for modelling the writing process in early Middle English scribes? It is clear from the discussion above that any adequate model would look rather different from one based on standard phoneme/grapheme mappings. We also suspect that its basic construal would be different. Writing in early Middle English is a procedure that requires more intervention from the agent than the (apparently) unproblematic writing behaviour of modern standard-language literates. We do not (except rarely) make orthographic choices when we write; the majority of our scribes, judging from their substitutional habits, appear to do precisely that. We cannot of course reconstruct their individual motivations for particular choices, but we can model their behaviour based on the phenomenology of their texts.50 This notion of choice holds even when (as may be the majority case) the scribe is copying an exemplar not his own. As the well motivated distinction between literatim and translating scribes makes clear, every act of writing represents a choice point. The choices, however, are more constrained for the literatim scribe though they may well not have been for the writer of his exemplar.

There are two aspects to writing: it is an action, but it is also a system or set of protocols. In other words, before he actually writes, the scribe asks:51 How shall I represent this linguistic object? What are my options? Which strategy type (economical or profligate) shall I use? How constrained am I by pre-existing design? What is the difference in the range of choices I have if I am copying as opposed to composing? We can assume that in most cases an author would not produce a fair copy in the first instance. His rough copy might be the exemplar for himself or for some other scribe; we rarely know which. Orm, however, gives us considerable insight into authorial scribal choices and how they may change through time. We would presume that Orm was less constrained (except by his own orthographic theory) than copyists of other people’s work, because everything was coming out of his head. He was not working with an exemplar in the strict sense. If he had a rough copy at any stage in his work, his editing processes on it are lost, and the transitions between it and his fair copy are now invisible. But his habit of continuing to work on the copy of the Ormulum that survives, and the layered corrections it evidences, illustrate how even self imposed constraints can evolve.52

We suggest that a reasonable model for general scribal practice is defined by a ‘scribal lexicon’; this consists of a set of PSS and LSS listings, and a set of word and affix templates. The general listings would look like this:53

1. For each potestas there is a canonical LSS, e.g. [o:] ⇔ {‘o’, ‘oi’, ‘ohi’}

2. For each littera there is a canonical PSS, e.g. ‘oi’ ⇔ {[o], [o:], [ɔ:], [u:], [u]}

The inventory of these constitutes the superstructure of the lexicon, the material out of which word and affix shapes may be constructed.

Subsets of the canonical LSSs and PSSs are then associated with particular forms. This is an important point: for most of our sources there is at least some degree of lexical specificity, such that for a given potestas in a given form only a subset (most often but not always a proper subset) of a given LSS is available. We would then conceive typical lexical representations (here from Trinity Hand A) as being of this type:

good n.

#

[g] ⇔ {‘g’}

[o:] ⇔ {‘o’, ‘oi’, ‘ohi’}

[d] ⇔ {‘d’}

#

We include # ‘word terminus’ because certain substitutions are limited to positions adjacent to # (see below); in complex words we include + ‘formative boundary’, which may control processes such as syncope in verbal endings. An example illustrating the role of # would be:

#

[f] ⇔ {‘f’}

[e:] ⇔ {‘ei’, ‘ehi’}

[t] ⇔ {‘t’, ‘d’}

#

The littera ‘d’ (an inverse spelling based on optional final stop devoicing) can only appear in this word to the immediate left of #. Another example of the role of # in this text is voiced variants of historical voiceless initial fricatives: these can only appear to the immediate right of #. As an example of lexical specificity (at least as applies to the surviving sample of this text language) we might note that the voiced variants do not appear in fire (< OE fȳr, 6 tokens), whereas they do (spelled ‘w’, ‘v’) in fair (< OE fæger, 4 out of 18 tokens).

Another important phenomenon has to be accounted for: this is zero (see the discussion under §2.3.5 above). Aside from mere absence, zero may be manifested as either excrescent ‘h’ or ‘otiose’ final ‘e’. Zero for etymological ‘h’ can be represented in the model so far, by simply having zero as a member of the LSS for initial [h]. Many forms have a final ‘e’ that is not etymologically justified, but is presumably ‘decorative’: e.g. in Trinity Hand A (which has fewer than many other texts) in nominative tuelue twelve, breste breast < OE twelf, brēost. Below are models for words with excrescent ‘h’ and otiose final ‘e’, where empty [ ] indicates a potestas-free but littera-accepting slot:

#

[ ]54 ⇔ {‘h’}

[ɔ:] ⇔ {‘oi’}

[t] ⇔ {‘t’}55

#

#

[b] ⇔ [‘b’}

[r] ⇔ {‘r’}

[e(:)]⇔ {‘e’}56

[s] ⇔ {‘s’}

[t] ⇔ [‘t’]

[ ] ⇔ {‘e’}

#

In both these examples the forms represented are the only ones in the text; to illustrate just how complex the substitutions in both zero and filled positions can be in a single word, consider the spellings of after (adverb, conjunction and preposition) also from Trinity Hand A:

hefter, Afteir, afteir, aftir, aster, efter, -hefteir, -hefter, -hester

Here we for the first time we encounter multiple potestates in a single slot:

#

{0} ⇔ {0, ‘h’}

{[a], [e]} ⇔ {‘a’, ‘e’}57

{[f]} ⇔ {‘f’, ‘s’}

{[t]} ⇔ {‘t’}

{[e]} ⇔ {‘e’, ‘i’, ‘ei’}58

{[r]} ⇔ {‘r’}

#

So the model for litteral praxis consists of the canonical littera/potestas mappings plus a set of individual word templates. It is not intended to be a fully developed formal model in the usual sense; rather it is a ‘diagram’ — a procedural layout driven by the actual data presented to us by our text witnesses.

This chapter has provided the background for interpreting the orthography of Middle English forms and demonstrates the necessity of phonetic reconstruction within the LAEME framework. In the next chapter we describe the corpus of forms itself.