A dialect atlas is, at least in part, a set of maps showing the distribution of linguistic features in space.2 Whether modern or historical, it is not made up, however, of static displays of dots or boundary markers on regional maps. Rather, it shows a continuum of overlapping distributions, where ‘dialects’ are assemblages of regional features that vary from map to map both spatially and temporally.3 This is because language change, like biological change, occurs differentially in different spatial settings. Change over time involves the vagaries of language use and thus introduces a necessary social dimension. Dialectology therefore operates on three planes: space, time and speech community.

We have no direct record of spoken language before the development of sound recording in the 19th century. For any earlier period our only source material is therefore written language. In other words, the ‘native speakers’ of past stages of a language are writers and copying scribes. For a medieval historical atlas therefore, ‘text languages’ (Fleischman 2000) must take the place of the live informants of a modern dialect survey. We cannot question our sources directly nor ask them for data that do not already appear in their surviving output. These are the three potential operational problems: (a) we have to work with the written rather than the spoken language; (b) our data are limited by the contingent survival of texts; (c) our knowledge of the personal details of the producer of any particular text is limited: medieval scribes are mostly anonymous. All we know is that they were literate and, at our early Middle English period, mostly adult male Catholics.4

By early Middle English we mean all written English from about 1150–1325, after which the language may be termed ‘late Middle English’. A Linguistic Atlas of Early Middle English (LAEME). is designed to cover the English that was written during the three or four generations before the period already covered in A Linguistic Atlas of Late Mediæval English (LALME). It aims, as far as the surviving material allows, to capture, display and analyse the written dialect continuum of this earlier stage in the language.5

How do we delimit the period ‘early Middle English’? Periodisation of languages is to a large extent arbitrary: 6

The ‘early’/’late’ division is largely ... a matter of convenience. The language of the 12th century and that of the 15th are different enough to justify separate identifying names. But language differences in time, like those in space, form continua: there are no sharp temporal boundaries. With respect to ‘archaism’ and ‘modernness’ Middle English before 1300 is certainly ‘early’ and that after 1350 ‘late’; the language written between is perhaps ‘transitional’, but this property exists only by virtue of the strong differences at both ends (Laing and Lass 2006: 418–19).

In the case of English, however, there is one very clear break in its written history, which enables selection of a terminus a quo. This break was one of the consequences of the Norman Conquest in 1066. Before the Conquest, the Anglo-Saxons, typically for Germanic rather than Romance cultures, used the vernacular (alongside Latin) to record their legal and administrative documents, as well as for religious, historical and literary works. Thereafter, the conquerors imposed their own native practice, Latin soon replacing English as the language of government 7 and of literature. The settlers also introduced their own vernacular literature. Alongside Latin, French became established as the other high status written language.

For LAEME, the most important effect of this disruption in the use of written English is the relative paucity of surviving material for our period compared with the more extensive sources available for LALME. It is likely in any case that more written texts will have been lost from a period so much earlier, but this problem is exacerbated by the fact that English was scarcely used as a written medium for the century immediately following the Conquest, and thereafter only gradually became reinstated. Until well on in our period, much of what does remain may be short or fragmentary texts written in manuscripts containing mainly Latin and/or French. 8

However, writing in English did not cease completely after the Conquest, since Old English texts continued to be copied and studied (and in the case of legal documents used) as historical objects themselves. Some of these copies, from before 1150, already sporadically display linguistic features we associate rather with Middle English than with Old English. 9 However, they are still recognisable as versions of an older stage of the language rather than as something that their copyists would have written spontaneously. There was apparently very little new composition in English in the century after the Conquest, apart from a few late additions to the Anglo-Saxon Chronicle. It is only towards the end of the 12th century that new writings in English begin to be attested in any quantity. These texts are linguistically clearly distinct from the copied Old English ones, indicating that the spoken language, as we might have expected, had continued to change throughout the post-Conquest period.

We take the work of the scribe of the second continuation of the Peterborough Chronicle, writing in 1154/5, as our terminus a quo for early Middle English. This text represents the earliest surviving example of truly Middle English language — that is, it reflects how the spoken language of the scribe’s region of origin had developed in the preceding century. After this there is a gap in attestation: there are very few other surviving texts in newly composed English that can be firmly dated earlier than the last quarter of the 12th century, most of these being ‘ca 1200’. Often manuscripts can be dated only from palaeographic evidence and this method is always approximate. Dates judged from script and decoration are usually given as lying somewhere within a quarter-century range. Were it not for the survival of the Peterborough Chronicle second continuation, we could reasonably say that our terminus a quo would be ca 1200. This is relevant to the selection of our terminus ad quem.

The terminus ad quem is less easily determinable — we do not have a last text in the same way as we have a first text. In addition, the existence of LALME creates certain responsibilities: we would like the two atlases to give as coherent a picture as possible of Middle English ‘as a whole’ without excessive duplication. A linguistic atlas displays how different forms of language change through time and how they vary across space. This leads to a further consideration: selection of too broad a time-span for a dialectal survey makes it more difficult to assess the relationship between diachronic change and regional variation. This is because over an extended period, the results of linguistic change are likely to present us with an intractably large typological range. On the other hand, choice of too narrow a time-span incurs the risk that the surviving texts will not provide sufficiently dense geographical coverage to constitute a continuum.

For LALME, a period of 100 years was considered optimal. For LAEME, in practice it is necessary to select a rather longer time-span10 to allow maximal regional coverage in a period from which fewer texts survived. In other words, wider temporal extension is a heuristic that increases our chances of encountering written material originating from different regions, which is especially important for early Middle English when the survival of texts is geographically very patchy. Moreover, dependence on data from an accidentally surviving sample means that our optimal time-span is not the same across the country. Even in LALME the 100-year period was different for the two areas of survey, North and South:

From before about 1350 there are very few sources for northern or North Midland English, so that, for those areas at least, the Atlas can go back no earlier; but for certain parts of southern England the evidence from texts of the earlier fourteenth century is so valuable that they could hardly be excluded, and the same applies even to a few late-thirteenth-century texts... Conversely, since the spread of written Standard English was earlier in the south than in the north, dialectal texts in the south become rare at a correspondingly earlier date. For the south, therefore, the ‘core’ evidence here utilised should be regarded as falling within the period ca. 1325­–1425, rather than that ca. 1350–1450 which applies for most of the Midlands and the north (LALME I, §1.1.2).

If LAEME and LALME together constitute an ‘atlas of Middle English’, in principle we would want to minimise the chronological gap between the LAEME sources and the LALME sources. In practice, this means that we too have adopted different termini for different parts of the country. Because of the small numbers of surviving texts of northern and North-Midland origins from pre-1350, we have included the few early texts that do survive, even if they are as late as the first half of the 14th century. In the southern half of England, survival of early texts is also uneven. There is no representation for the Central Midlands, very little for the true South, rather more for the East Midlands and relatively dense coverage for the South West Midlands. For the southern area of survey, the inclusion of some late-13th-century texts in LALME means that a terminus ad quem of 1300 for LAEME allows for sufficient overlap, with some late-13th-century texts appearing in both atlases.11

One text that is an apparent exception to our usual procedure is Ayenbite of Inwyt. Unusually for a medieval text, a colophon in the hand of its scribe gives his name (Dan Michel), origins (Canterbury) and the date of the text’s completion (1340). There is very little early Middle English surviving from the South East, but Dan Michel himself refers to his work as being in engliss of kent. We know that he was probably an old man in 134012 and can therefore take the language of his text to represent Kentish of the last quarter of the 13th century rather than of the mid 14th century. Accordingly this text appears in LAEME as well as in LALME.

In this section we outline the principles and methodology that underpinned the construction of LALME, and in the next section (§1.5) we identify the modifications that have been required for LAEME and introduce the resulting new methodology and data outputs.

1. Written language should be studied in its own right, not just as a representation of spoken language (cf. McIntosh 1956). Written language should be regarded as an autonomous linguistic system, if one with a special relation to phonology (see further chapter 2).

2. Regional dialects do not have discrete geographical boundaries. Interlectal variation normally constitutes a continuum (see above §1.1).

3. Anchor texts and the fit-technique.13 In surveys of modern dialects, localisation of informants is transparent: for any survey point one identifies a local witness and then plots his usage on the maps. Unfortunately, most medieval witnesses are more elusive: scribes, especially those of literary manuscripts, typically do not supply information about their local origins. As LALME felicitously puts it (1, §2.3.1): ‘It is rather as if the compilers of a modern dialect atlas had access to any number of speakers, all willing to be interviewed but very few of whom divulged where they came from’.

Some manuscripts, however, can be localised on extra-linguistic grounds. ‘Local documents’ such as personal correspondence, local records or legal instruments,14 constitute the one large body of texts whose places of origin or execution are for the most part explicitly indicated. Such texts, which are referred to as anchor texts, are the necessary starting point for a historical linguistic atlas. They enable the setting up of a matrix of defined localities, within which texts not carrying such information can be localised. This is possible because of a central empirical claim incorporated into the LALME tradition: ‘it is fundamental that dialect changes in, for the most part, an orderly way over space, and that the domains of the various dialectal characteristics (again, for the most part) overlap differentially (Benskin 1981: xxx). For LALME, the language of local documents provided valuable data itself, but its importance was far greater as the means by which the unlocalised literary manuscripts, which formed the greater part of the LALME source material, could also be utilised. The methodology is known as the fit-technique. ‘Fitting’ is done by comparing, map by map, spellings particular to an unlocalised text with those already placed in the localised matrix. For each map, areas where those or similar spellings are not found are then eliminated, until (in the ideal case) only a single, well-defined location is left where the whole assemblage of spellings could plausibly occur. For exemplification and further discussion see Benskin (1991a).

4. The questionnaire.15 The traditional analytical tool for the dialectologist is the questionnaire. This involves the preselection of items, which the investigator has reason to believe are likely to be dialectal discriminants. ‘The term item [is] used to denote the heading for a collection of different forms that are regarded as equivalent in function and/or meaning, and may therefore, potentially at least, differentiate dialects’ (LALME 1, §2.1.1). Constructing a suitable questionnaire out of these functional equivalents is a matter in the first instance of informed experimentation. Ideally it must be large and varied enough in the range of linguistic phenomena it elicits to provide a conspectus of the language under investigation. Choice of a questionnaire item requires not only that its forms show variation but also that it occurs in most of the source texts. The combination of formal variation and frequency of attestation results in what we might call the discriminatory yield. Maximal discriminatory yield may be elicited by different items in different regions. It is for this reason that the two areas of survey in LALME used somewhat different,16 though (for obvious comparative reasons) partly overlapping, questionnaires. The combined questionnaire consists of 280 items, some of them subdivided, of which about a third form the common core. A completed questionnaire is processed to form a Linguistic Profile (LP) of the scribal language. If the localisation of the scribal language is known, the forms recorded for each Linguistic Profile may then be entered at the relevant place on the dialect maps. In LALME there are two types of map: Item Maps (vol. 2) that display actual manuscript spellings, and Dot Maps (vol 1) ‘that highlight the distributions of specified variants or classes of variant’ (LALME vol. 1: 297): e.g. ‘kir-’ and ‘kyr-’ types of the item church (LALME vol. 1: Dot Map 388).

For LAEME we elected not to use a questionnaire, so the problem of constructing differential regional questionnaires does not arise. For explanation of the alternative methodology adopted see §1.5.4 below.

5. Scribal practice. Very few surviving Middle English texts are holographs. Most by far are scribal copies at some remove from their archetypes. Before the work of the late Middle English dialect survey that resulted in LALME, it had been widely believed that the language of all such copied texts must represent dialect mixture. It was assumed that a copied text must be made up of a combination of elements from the authorial language and whatever linguistic forms could be ascribed to the series of copyists between the original and the surviving version in question. Early in the investigation, however, Angus McIntosh (1973 [1989: 92])17 observed that most copied Middle English texts were in fact in language that was dialectally homogeneous: some scribes ‘translate’, or convert the language of their exemplar wholesale into their own variety. If a copyist is a consistent translator his language overrides that of his exemplar. However far removed the scribal language might be from that of the original, it nevertheless represents the usage of an individual and therefore has independent value as belonging to a native informant. In what might be called the ‘LALME tradition’ — as opposed to more conventional forms of textual study — the emphasis is shifted away from a putative ‘authorial’ language to the actual language of a copy and its scribe. The output of a translating scribe is thus always useful for mapping.

Not all scribes were translators. McIntosh identified two other possible scribal copying strategies, the outputs of one of which are also potentially valuable as witnesses for linguistic geography. Aside from translating, a scribe may copy a text exactly, or he may only partially translate. The first strategy is known as literatim copying, and as long as the language of the exemplar is itself homogeneous, it will result in a homogeneous copied text language.18 The second strategy, and any literatim copy of its results, will always result in a so-called Mischsprache, A true Mischsprache is defined as linguistic output containing two or more elements that are mutually incompatible: that is, from non-contiguous areas within the established dialect continuum (see further Benskin and Laing 1981:76–77). It is obvious therefore that a Mischsprache is not as such mappable. Further analysis may make it possible to isolate the disparate linguistic elements and thus to map subsets of the scribal usage. In LALME such processed subsets were not used on the maps — there was sufficient homogeneous source material to render this unnecessary. For LAEME, we are not so fortunate, and for some cases we have had to devise strategies to make use of data from dialectally mixed texts (see §1.5.5 below).

Our survey of linguistic variation in the period 1150–1325, while being firmly in the tradition of LALME, has developed an independent identity. When the LAEME project was initiated 20 years ago, we were aware that the new investigation would have its own problems and difficulties, but we thought that the principles for constructing a medieval dialect atlas had already been established and just needed to be adapted in relation to the new data. In the event the very different nature of the available materials for early Middle English drove us to modify the established principles and to create a new conceptual and procedural model. The modifications resulted in the adoption and development of a radically new methodology and also prompted an expansion of what we thought was the original remit of an atlas.

The principles and praxis adopted in the creation of LALME require discussion in relation to the rather different demands of LAEME.

Any historical linguistic survey has to address the fact that different kinds of text will prompt different linguistic choices. Pulling forms out of their linguistic context and taxonomising them by means of a questionnaire creates a comparability that irons out some of these variables. That indeed, is partly what a questionnaire is designed to do. But for historical surveys such as LALME and LAEME, the dependence on the contingent survival of texts as linguistic witnesses means that some of these differences are in practice obvious. The questionnaires completed for LALME were processed to form taxonomic listings for each scribe, known as Linguistic Profiles (LPs).19 A comparison of the LPs in volume 3 of LALME quickly reveals that some are considerably denser in their responses than others. ‘Thin’ LPs, with multiple null responses to particular questionnaire items, are mostly those derived from very short texts (such as some of the otherwise crucially important documentary ‘anchors’). But comparative thinness or density may also be the result of lexical and morphosyntactic choices. For instance, a past tense narrative may not have examples of items that elicit present tense verb morphology,20 while instruction manuals may not have examples of those that elicit past tenses.21 The result is that for some items the linguistic maps in LALME are variously patchy in attestation. Where the matrix of survey points is dense this does not have such serious consequences for our view of the continuum as a whole. But where the matrix is itself thin, lack of attestation of forms for any particular item makes for a more gapped continuum.

There are many other variables that may affect the outcomes of linguistic analysis, such as whether a text is verse or prose or whether it is a translation from Latin or from French or from an older form of English itself. The problem of comparability cannot easily be solved for a historical dialect survey of any period, but it becomes more acute the thinner the corpus of witnesses (Laing 2000a: 103). For the early Middle English period, even when all the available texts are utilised, we are still left with less than a tenth of the number of texts that were chosen for analysis from the vastly larger surviving corpus of manuscripts available to the compilers of LALME.

For early Middle English we have information from certain areas from early in our period and from others only from late in the period. For some regions (see §1.3) there are no witnesses at all. We therefore have considerable gaps in both the space and time continua, so that LAEME is patchy in its basic coverage of survey points. Complete, coherent and cohesive dialectal patterns across the whole area of survey are not to be had. In some areas, however, such as the South West Midlands, there are enough local texts for something approaching a dialect continuum to be observable. Moreover, early Middle English dialects should not be studied in isolation from later Middle English dialects. The space continuum in LAEME is incomplete, but the time continuum is inextricably entwined with that of LALME. In order to allow empirically responsible comparison with the later period, and a conspectus of diachronic and diatopic variation for Middle English as a whole, we have produced text dictionaries as the equivalent of the LALME LPs, and a series of item and feature maps, which to some extent parallel the item maps and dot maps in LALME. The LAEME text dictionaries and maps are derived from a corpus of lexico-grammatically tagged texts,22 which form the core of the LAEME database. The corpus methodology and outputs are described in part II. Other display types that have been made possible by our revised methodology are discussed and illustrated in part III.

In constructing LAEME we faced a further problem that is closely linked to the general problem of the gapped time/space continuum. For LALME, the numbers of surviving local documentary texts in English from 1350 onwards, though not completely even in their distribution, provided a reasonable matrix of localised anchors (LALME 1, §2.3.2). As indicated in §1.2 above, in the early Middle English period the vernacular was not used for the kinds of official document that provide explicit local origins.23 The only official documents appearing in English in the late 12th and the 13th centuries are copies from pre-Conquest originals found in cartularies and monastic registers. Only very few of these were linguistically updated by the copying scribes. Usually, presumably at least partly for reasons of ‘authenticity’, the language of these copies is preserved by the monastic registrars as unmodernised Old English (this is particularly common in the case of boundary clauses — Kitson 1995: 48–49). In spite of the date of copying, these texts therefore do not belong to the Middle English corpus. In some cases, the later copyists had only a very shaky grasp of Old English, so that their attempts accurately to transcribe their exemplars produce garbled versions that are equally useless as samples of early Middle English (Laing 1991: 38–40). When these are discounted, there remains only a very partial matrix of genuinely original or updated documentary texts (Laing 2000a: 104–106).

This documentary matrix is so exiguous that alone it would be virtually useless as an aid for fitting unlocalised material. Although we were aware from the outset of the almost total lack of original documentary material pre-1350, we had hoped that more of the Old English texts copied during our period would turn out to be of use. The recognition that almost all were not usable provided a further setback to our expectation of using LALME methodology to produce something like a parallel early Middle English atlas. Indications of local origins for early Middle English texts are not, to be sure, confined to documents. Literary texts may also be associated on non-linguistic grounds with particular places. There is however a hierarchy in such associative clues. Extra-linguistic local associations in documentary texts are for the most part more likely to be reliable than those for literary manuscripts. Local records or legal instruments were mostly drawn up by local scribes, so can usually be trusted to attest forms of language of their stated place of origin or of somewhere nearby. The few exceptions are likely to be recognised because the body of local documents for any place would normally constitute a tradition of scribal practice against which non-local deviation is obvious. Non-linguistic associations in literary manuscripts represent a much broader spectrum of localising evidence.

The single most convincing ascription of local origins in a literary manuscript in our corpus is Dan Michel’s colophon in London, British Library, Arundel 57 containing Ayenbite of Inwyt (see above §1.3). This constitutes local association at least as convincing as any evidence for a documentary source, and accordingly the Ayenbite functions as a firm anchor. But other indications to be found in literary manucripts are variously less convincing.

The so called Worcester Tremulous Scribe was interested in Old English texts, and he provided Latin (and in three manuscripts also Middle English) glosses in at least twenty manuscripts known to have been in Worcester during the Middle Ages (Franzen 1991: 29–83). His work has been ascribed to the first half of the 13th century (Ker 1937) thus falling squarely in our period. The fact that he was working on manuscripts owned by Worcester Priory at the time need not entail that he was brought up and trained in Worcester; but lacking evidence to the contrary it is a good working assumption. Apart from his major work of glossing, he also wrote longer texts in English. These are preserved in Worcester Cathedral, Dean and Chapter Library F. 174: a copy of Ælfric’s Grammar and Glossary, which is at least partly ‘translated’ from Old English to Middle English, and the so-called Worcester fragments which are probably copied from a late 12th-century exemplar (Moffat 1987). We have not included his Middle English glosses in the LAEME corpus, but the Tremulous Scribe’s extended English work provides a literary anchor text for Worcester, especially as, allowing for date, its dialect appears consonant with Worcester material from the later period.

Further down the hierarchy lies a manuscript like Oxford, Bodleian Library, Digby 86. This is a trilingual manuscript (Latin, French and English) written in the last quarter of the 13th century (Tschann and Parkes 1996: xxxvii–xxxviii) by one main scribe with a second scribe contributing only a very small amount of French text. The manuscript contains obits (one written by the main scribe himself) of people associated with places in SW Worcs and N Gloucs (Laing 2000b: 524 and references there cited).

Less probative still, are manuscript associations provided by evidence for ownership, such as library pressmarks or ex libris inscriptions. Medieval library catalogues may or may not be able to establish the ownership of a particular manuscript at the time when it was actually produced. Later ownership does not guarantee that the scribe or scribes who contributed to its contents were local to its present home. Even contemporary ownership need not imply that the scribes themselves had not come from elsewhere. A cautionary example is provided by the British Library, Cotton Cleopatra MS of Ancrene Riwle, for which the earliest evidence of ownership is an inscription of ca 1300 indicating that it belonged to Canonsleigh Abbey in Devon (Dobson 1972: xxv). But we believe from linguistic evidence that of the three Middle English scribes contributing to it, two came from North Worcestershire or North Herefordshire, and one from West Norfolk (cf. Laing 2000a: 114).

Indications of provenance provided by manuscript contents, scripts or illuminations must be similarly treated with caution. However, the language of these texts may be taken, at least provisionally, to represent the language of the place of origin of the manuscript, subject to confirmation by other local material.

We can therefore cautiously supplement our documentary matrix of anchors with literary anchors. Even so, the business of fitting text languages that have no extra-linguistic indications of provenance is still compromised by the unevenness and overall thinness of the early Middle English anchor matrix. In this situation we are fortunate to have the much more densely attested dialect continuum for later Middle English displayed in LALME. A dozen or so of our texts were already localised in LALME. More detailed analysis has led to modifications of some of these placings. In two cases, where the contributions of more than one scribe were amalgamated, these have been separated. In general, however, the localisations of these early texts hold good, whether as anchor texts or fitted texts. Where necessary, then, the LALME continuum can be used as scaffolding for the putative early Middle English continuum. In areas where there is little or no early Middle English attestation, we can interpolate unlocalised text languages of the earlier period into the LALME continuum in order to establish plausible localisations. But it must be understood that ‘fitting’ for LAEME is not anywhere near such a robust concept as it is for LALME. The LALME configuration has been used to help with some fittings, but in the sparsest areas any localisation is bound to be very approximate indeed and will always be subject to subsequent revision if more data or information becomes available. For much of LAEME, the display of linguistic data in map form at all is a convenient but highly generalised abstraction.

The apparently exact placings attempted for LAEME are a function of perceived patterning in relation to other texts of similar language within a kind of abstract linguistic space that also takes into account the time axis. They are also driven by the necessity, for mapping purposes, of putting in some specific place text languages that appear to be homogeneous and local. Where text languages exist in larger numbers, and the configuration is denser (e.g. in the SW Midlands) the concept of linguistic space becomes even more important. In the early Middle English period, religious houses and a number of early-established town schools (Orme 1973: 295–325) would have provided opportunities for learning the art of writing and copying. For the SW Midlands, surviving early Middle English texts, in somewhat differing forms of language, outnumber the most likely places of origin of written local dialect systems. The complex of texts that include those in London, Lambeth Palace Library 487, London, British Library, Cotton Nero A xiv, London, British Library, Cotton Caligula A ix, part I (Laȝamon A), and part II (Owl and the Nightingale) are all very similar to each other, and also to the language of the Worcester Tremulous Hand and other material with Worcester associations. It is possible that varying Worcester language is what this complex may represent. But it would be very difficult to display the material cartographically all on one spot; so for the purposes of mapping, texts have sometimes been spread out, according to the usual criteria for fitting, across areas in which there were few or no contemporary centres of teaching and learning.

The advantage of a questionnaire is that the same list of items is used for every informant. This enables the dialectologist to address the key desiderata of dialectology: description and comparison. The questionnaire’s two main weaknesses are that (a) it requires considerable prior knowledge of the language (or else a great deal of trial and error) before an effective questionnaire can be finalised; and (b) because its contents are by definition limited by the investigator’s selection of items, its function as a linguistic net only ever results in a limited catch (Williamson 1992/3, p. 139).

There is a further disadvantage, which is that questionnaire analysis has to be done by hand. This makes its use particularly problematic for the English of our period. As part of the point of the survey of the earlier period was comparison with the later material, we would want most of the LALME items to be included also on the LAEME questionnaire. For very good methodological reasons (chaque mot a son histoire) the LALME questionnaire had not included many items eliciting phonological categories. For those that were there, it was operationally difficult for the analyst to collect consistently for them. For items such as a/o (reflexes of OE ā) or wh- (reflexes of OE hw-) one had to record for that item the segments not just from the relevant words that featured already on the questionnaire as items in their own right (for which one also had to record the full spelling in the relevant place), but one also had to note them for all the words not on the questionnaire already. For early Middle English, while fully acknowledging lexical specificity, we wanted to include many more phonological items that we thought would be of particular interest at the period, such as the reflexes of OE . It would have been impractical to include as separate items all the possible lexical sources for this category. We were also interested in similarly ‘catch-all’ orthographical items such as the distributions of the letters thorn and edh, whether they were used initially, medially or finally.

During the early Middle English period, the language is in a process of rapid change from a state with still a considerable degree of inflection to one in which endings are being levelled. Prepositions are increasingly taking over the functions of some verbal prefixes and of inflexions. It is in principle likely that such changes were happening at different times in different places. We could not be content simply to record the spellings of the stems of nouns and adjectives: there was a case for recording not only the singular and plural endings of each as categories in their own right, but also whether they had subject and object function in the sentence. The pronouns still show complex inflectional systems and even the definite article may still be marked for case number and gender. The relationships between linguistic form and function in these instances are potentially of dialectal significance. It will be obvious by now how impractically large and complicated the early Middle English questionnaire had become:

The operational difficulty in noting each form and function of a word separately is compounded if one also tries to record separately its inflectional ending for comparison with others used in the same function. ... If all this potentially significant information is to be noted, a single word or morpheme in the text may have to be recorded in a number of different places in the questionnaire.

It was apparent early in the investigation that the method of ‘analysis by hand’ described above was totally impractical. It was time-consuming, over-complex and the inherent likelihood of error and inconsistency was unacceptable (Laing 1993: 126).

In these circumstances we decided to use a different analytical tool for the investigation of early Middle English linguistic variation.

Instead of using a questionnaire, we elected to transcribe all the early Middle English texts (or extensive samples of very long texts) and to put them on disk in a form enabling electronic processing: 24

The advantage of this method is that all the linguistic data can be subjected to analysis without the investigator being committed to a pre-selected set of dialectal discriminants. The results of the analysis may then inform the selection of items for linguistic profiles and dialect mapping (Laing 1994:127).

The resulting corpus of lexico-grammatically tagged texts is a database that can be sorted and analysed (using software written for the purpose by Keith Williamson). From each tagged text is derived a text dictionary, which is the taxonomised inventory of each text language. Any pair or n-tuple of these may be compared electronically. Such a taxonomised inventory is what we call an assemblage, which is a proper subset of a given scribe’s total usage. Thus the database, in association with the processing software, becomes itself the instrument of selection. Because of its open-endedness, it develops a heuristic function impossible from self-limiting procedures such as the questionnaire. Further, the corpus allows for extended use (see chapters 3 and 4): rather than being solely a dialectological resource, it also becomes a powerful research tool for studies in historical phonology, morphology, syntax, semantics and pragmatics.

For late Middle English, McIntosh observed that many scribes were translators, and many also perpetrated Mischsprachen, but that literatim copying was comparatively rare. The situation seems to have been different in early Middle English, where there were far fewer translating scribes and proportionally more literatim copyists. As a result, we find more manuscripts containing linguistically composite texts. Considering the relative paucity of data at this period, we do not have the luxury of rejecting a large proportion of our source material as being in mixed language. It is therefore necessary to subject the texts to more detailed analysis so as to establish their stratigraphy. It has proved possible in a number of cases to isolate within a text in mixed language one or more linguistic layers which can be taken to represent genuine regional usages.

Benskin and Laing (1981) details for late Middle English the copying strategies of scribes, and the kinds of Mischsprachen and pseudo-Mischsprachen that result from phenomena such as relict usage and constrained selection. Relicts, by definition, are confined to the output of translating scribes. A relict is a form from the usage of the exemplar that is alien to the dialect of the scribe and which he has allowed to slip through his translating net. This may happen either by accident as the translating scribe is working in to his task, or perhaps sometimes deliberately because it helps the text retain an exotic flavour. Constrained selection invokes the concept of a scribe’s passive repertoire of spellings. A translating scribe will normally translate all exemplar forms that are alien to his own usage — his active repertoire. If, however, a spelling not his own is familiar to him because it is to be found in neighbouring areas, it forms part of his passive repertoire, and he may well not feel the need to translate it. In this case his own usage is constrained by that of his exemplar, and his copied text, while not containing any forms alien to his area of origin, may show skewed frequencies for the variant spellings of particular items.

Following the terminology established in Benskin and Laing (1981), Laing (2004) describes the types of scribal complexity, both spatial and temporal, to be found in early Middle English texts, and the major analytical techniques for disentangling them. References to specific published studies on early Middle English texts undertaken for the creation of LAEME may be found under individual manuscript entries in the Index of Sources, while notes on particular textual interpretations are often to be found in the tagged texts themselves.25

This chapter has introduced and defined the concept ‘early Middle English text’. Chapter 2 addresses the main theoretical issues concerning orthographical and phonological interpretation of the LAEME materials.