1.1 Introduction: emergence of the variationist approach and sociohistorical linguistics

This chapter presents an overview of the variationist approach and illustrates how this theoretical and methodological framework arose through convergence of research investigating language variation and change within sociolinguistics and dialectology. This first section explains how the variationist approach was introduced and developed within (English) historical linguistics through the emergence of a research field variously called sociohistorical linguistics or historical sociolinguistics, terms often used interchangeably. Discussion of methodology follows in section 1.2, and some illustrative findings from this paradigm are presented in section 1.3. The final section (1.4) concludes with brief discussion of new tools for analysis, novel data sources, and theoretical frameworks, along with consideration of some remaining problems.

Weinreich et al.'s (Reference Weinreich, Labov, Herzog, Lehmann and Malkiel1968) ground-breaking discussion of empirical foundations for a theory of language change marks the birth of sociohistorical linguistics or historical sociolinguistics. Although the authors use neither of these terms, this publication heralded an important turning point in proposing a theory of language change embedded in both linguistic and social structure. Starting from the premise that synchronic variation of the type investigated by contemporary sociolinguists and dialectologists represents a stage in long-term change, they proposed using the present to explain the past and the past to explain the present. Most importantly, however, implementing the variationist approach requires taking account of the external history of a language in historical reconstruction and abandoning the strict Saussurean dichotomy between synchrony and diachrony.

A key tenet of the variationist approach is the idea of orderly or structured heterogeneity (Weinreich et al. Reference Weinreich, Labov, Herzog, Lehmann and Malkiel1968: 99–100). Variation is not random, but structured in regular ways along a number of internal linguistic and external social dimensions, including, for instance, linguistic context, speaker position/status, and text type (style, genre, etc.). A major finding resulting from decades of urban sociolinguistic work, much of it conducted in English-speaking communities, is that differences among social class dialects are quantitative and not qualitative. When viewed against the background of the community as a whole, variation in the speech of and between individuals previously seen as free or random is conditioned by social factors like social class, age, sex, and style in predictable ways. Inherent variability shows directional gradience through social groups, geographic space, and time. Understanding what Weinreich et al. (Reference Weinreich, Labov, Herzog, Lehmann and Malkiel1968) refer to as the ‘embedding problem’ (i.e. how variability and change are embedded in a social and linguistic matrix) in speech communities today and the social meanings attached to variants allows us to make some predictions about the pathways change follows as it moves through a community (see section 1.3).

A primary goal of sociohistorical linguistics is to provide an account of the forms and uses in which variation may manifest itself over time, and of how particular functions, uses and kinds of variation develop within particular language varieties, speech communities, social networks, individuals, texts, and genres (Romaine Reference Romaine1982). The incorporation of variability represents a significant departure from traditional historical linguistics, which focused primarily on the evolution of written standard varieties, often ignoring variation and non-literary language. Taking account of developments in a broader range of varieties, especially those from geographically or socially peripheral communities or individuals further removed from literate traditions presents opportunities for a richer account of language change because non-standard varieties used in rural and working class communities often preserve older stages of changes already completed and no longer directly observable in the standard. While standard handbooks typically treat the Great Vowel Shift as a fait accompli because they take standard English as their reference point, certain elements have yet to be implemented in some varieties of northern British English. Modern sociolinguistic studies reveal that retention of monophthongal /u:/ in words like house is a relic embedded in the most informal styles of contemporary working class speech in Scotland (Macaulay Reference Macaulay1977). Likewise, in the domain of syntax, the infiltration of WH-pronouns such as who, whose, and whom can be seen as completed in the modern standard written language and some varieties of educated spoken English, but it has not really affected modern spoken vernaculars, which prefer that or zero marking, e.g. the woman (^*whom, that, Ø) I met (Romaine Reference Romaine1982, Tagliamonte Reference Tagliamonte, van Kemenade and Los2006b).

1.2 Methods

Guided by the methodological requirements for conducting synchronic research within the variationist paradigm, the main task for diachronic studies is to devise procedures for reconstructing language in its social context. Historical linguists confront similar problems when assembling two kinds of data, linguistic and social, in order to establish correlations between them. These include selecting the individuals/communities/texts to be studied, identifying the linguistic variables and their variants, collecting and analysing the data, and interpreting the results.

1.3 Findings

This section illustrates some key findings from variationist historical linguistics, focusing especially on six of the major so-called ‘external’ factors constraining and facilitating the implementation of change and its transmission: social status, social network, gender, age, style, and region. Although spontaneous innovations occur all the time, introduced by acts of speaking/writing on the part of individuals, many will not enter the language system at all. Some changes can be quite rapid while others take centuries. The chance of survival of an innovation depends partly on where/when it is introduced into the social and linguistic system and by whom. Innovations follow predictable paths through social and linguistic structures.

1.4 Future directions

All change depends on variation. Much of the variation in language previously thought to be random is indeed systematic. This chapter has shown how understanding of the social and linguistic dimensions of variability provides a richer account of the history of English at the same time as it illuminates trajectories of language change. Although language change may involve multiple mechanisms and some changes may proceed with no apparent social weighting, the social evaluation of variants often plays a role in determining the fate of a change in progress as speakers make choices between rival forms based on their perceived associations with the groups using them (e.g. social class, age, gender, etc.) or the contexts (e.g. style, text types, etc.) in which they occur.

Nevertheless, the issues of status and methodology initially raised by Romaine (Reference Romaine1982) are still paramount. That is, our findings will only be as robust as our methods and the reliability of our data. The level of confidence in the results of any quantitative study depends heavily on the issues of representativeness and frequency. It is still an open question whether cases of seemingly free variation are instead the result of inadequate research methods, lack of sufficient data for analysis or some other factors. We cannot jump to the conclusion that non-occurrence of a feature in a corpus indicates its ungrammaticality. The explanation for its absence may lie in non-grammatical, contextual factors or even be due to chance (Butters Reference Butters2001). Variation is usually conditioned by multiple causes, which means that researchers need to identify multiple factors and assess the relative contribution of each. There is still much to learn about the intersection of grammatical and phonological variation and the nature of the constraints on variable systems over time. The emerging field of ‘cognitive sociolinguistics’ relying on a convergence of method and theoretical frameworks to study cognitive and social constraints on linguistic variation is beginning to make significant contributions to our understanding of changes like the replacement of -th by -s in the third-person singular present tense. Using a mixed-effects statistical model incorporating a large number of sociolinguistic, psycholinguistic, and phonological independent variables and their interaction with time, Gries and Hilpert (Reference Gries and Hilpert2010) found that a significant change occurred in the sixteenth century when writers corresponding with someone of the opposite sex were more likely to use the new suffix. Prior to that time, the old suffix prevailed regardless of who a writer wrote to.

Historians of English are fortunate because English has left a more fully documented written history than many other languages, and English is at the forefront of corpus development and tools for analysis. Theoretical linguists have relied heavily on English as the language of theorizing and exemplification, and given the predominance of English as a world language, it appears likely that data from English will continue to make substantial contributions to theories of variation and change. Much is to be gained by mining this rich source of data, which has become increasingly amenable to rigorous empirical investigation in the form of electronic texts easily searchable using a variety of analytical tools (see Chapter 2 by Hilpert and Gries and Chapter 8 by López-Couso in this volume). Tools have to be adapted to the material available at particular periods. Most conventional synchronic and diachronic multi- and single-genre corpora are still far too small for studying syntactic variables. In cases where an item or construction is rare or belongs to a genre or time period not represented in corpora, scholars wishing to push our knowledge of language change beyond the boundaries of the material currently available in corpora will need to rely on a set of best practices to optimize use of various kinds of electronic resources often developed for non-linguistic uses. Using the internet to collect data and the availability of web-based corpora and databases also pose new challenges. In 2004 Google began scanning millions of books as part of an ambitious project to make every page of every book ever published available and searchable on the internet. Now comprising more than two trillion words from fifteen million books published between 1473 and 2000 (c.11 per cent of all the books ever published) scanned from sources in over forty university libraries, Google Books is the largest megacorpus and a potentially rich resource for linguists (Michel et al. Reference Michel, Shen, Kui, Aviva, Veres, Gray, Pickett, Hoiberg, Clancy, Norvig, Orwant, Pinker, Nowak and Lieberman Aiden2011), especially for comparing change and variation in the two major varieties of English, British and American (Romaine Reference Romaine2013). Combining increasingly sophisticated statistical models and interfaces for manipulating large data sources, future prospects for the variationist approach are very bright indeed.

2.1 Introduction

English historical linguistics has a rich and long-standing tradition of corpus-based work (see the surveys in Rissanen Reference Rissanen, Lüdeling and Kytö2008, Kytö Reference Kytö, Bergs and Brinton2012). Resources such as the Helsinki Corpus, the Brown family of corpora and ARCHER have spawned active research programmes for the study of lexical and grammatical change, both long term (Curzan Reference Curzan, Lüdeling and Kytö2009) and short term (Mair Reference Mair, Lüdeling and Kytö2009). In addition, corpus resources inform the analysis of diachronic variation in genres (Hundt and Mair Reference Hundt and Mair1999), registers (Biber and Gray Reference Biber, Gray, Bhatia, Hernández and Pérez-Paredes2011b), and varieties (Tagliamonte Reference Tagliamonte, van Kemenade and Los2006b). The present chapter will discuss a currently developing line of research which uses the methods of quantitative corpus linguistics for the analysis of diachronic corpora. This research program draws on, and is informed by, the aforementioned areas, but at the same time, it uses particular kinds of data and handles that data in specific ways that merit discussion. Diachronic corpora are understood here as textual resources that represent comparable types of language use over sequential periods of time, thus comprising at least two periods, as in the Diachronic Corpus of Present-Day Spoken English (DCPSE, Wallis et al. 2006), but typically many more, as in the Corpus of Historical American English (COHA, Davies Reference Davies2010), a monitor corpus which at the time of writing samples twenty-one sequential decades of language use (see Chapter 8 by López-Couso in this volume). The English diachronic corpora that are currently available represent different varieties and text types and vary in their respective time depths, but it is a design feature of most diachronic corpora to hold the type of text constant, so that diachronic language change within a given text type may be studied with as few confounding factors as possible. Quantitative corpus linguistics (Biber and Jones Reference Biber and Conrad2009) is a research tradition in which research questions are formulated in such a way that frequency counts from corpora may provide answers. Quantitative corpus work thus often engages in hypothesis testing, so that a testable empirical question (e.g. ‘Have adolescent women been leading the development of the quotative be like in Tyneside English?’) may receive an answer in terms of either ‘yes’ or ‘no’. Of at least equal importance are so-called exploratory techniques, which are designed to transform a complex dataset into a summary (and often visual) representation (which may then be interpreted by the analyst and that may in turn lead to the formulation of hypotheses). To give an example, Szmrecsanyi (Reference Szmrecsanyi, Geeraerts, Kristiansen and Peirsman2010) studies the use of genitive constructions in different text types of British and American English in the 1960s and the 1990s, exploring whether there are changes that could be seen as Americanization or colloquialization (see Mair Reference Mair2006). The frequency counts that enter quantitative corpus studies often represent token frequencies, but a much wider variety of measures is routinely used, including measures of type frequency, dispersion, and collocation.

The main point of this chapter will be an overview of how the two, diachronic corpora and quantitative corpus linguistics, are put together in fruitful ways. Quantitative studies of how units of linguistic structure change across corpus periods can address questions of more general linguistic interest, including the following:

• When and how does a given change happen?

• Can a process of change be broken down into separate phases?

• Do formal and functional characteristics of a linguistic form change in lock-step or independently from one another?

• What are the factors that drive a change, what is their relative importance, and how do they change over time?

• How do cases of language variation in the past compare to variation in the present?

It is already apparent from these questions that quantitative studies of historical change have a great deal in common with quantitative studies of synchronic variation (Tagliamonte Reference Tagliamonte2006a), both on the theoretical and the methodological level. This commonality is of course no coincidence, as language variation is one key factor for explaining why languages change over time. The remainder of this chapter is organized in the following way. Section 2.2 motivates the approach that is taken here and explains how quantitative methods usefully complement qualitative approaches in the analysis of diachronic corpus data. Section 2.3 is concerned with approaches to the diachrony of variation in language, and it discusses desiderata of such approaches. Section 2.4 turns to exploratory techniques, which can guide the researcher towards discovering new, unanticipated aspects of language change or assist in the formulation of hypotheses. Section 2.5 offers a few pointers for future research and section 2.6 concludes.

2.2 Language change by the numbers

Historical linguistics, by its very nature, depends on the observation of authentic data. However, not all research questions in historical linguistics oblige the analyst to quantify that data. Many processes of linguistic change manifest themselves in qualitative differences, so that for instance lexical items disappear from usage, or word order patterns that once were common are no longer used. (Of course, such differences can be quantified as observed frequencies becoming zero.) For instance, the Old English (OE) word order shown in (1), an example from Ælfric's Homilies of the Anglo-Saxon Church (ÆC Hom I, 1.20.1), is no longer used in Present-day English (PDE).

1. (1)

   on twam	þingum	hæfde	God	þæs mannes	sawle	gegodod
   in two	things	had	God	the man's	soul	endowed
   ‘God had endowed man's soul with two things’

The crucial characteristic of the example is the fact that the finite verb hæfde ‘had’ appears after an initial constituent in what is called the ‘verb-second’ position (Fischer et al. Reference Fischer, Bermúdez-Otero, Denison, Hogg and McCully2000). A gloss such as With two things had God man's soul endowed, which retains this particular word order, might be acceptable as a deliberate anachronism, but it will not pass as an everyday PDE sentence. Hence in this case, a single historical example, in connection with the intuitions of a present-day speaker, is enough to establish that a change has taken place.

However, more rigorous quantification of diachronic data becomes necessary when research questions go beyond the mere detection of a change and into the internal dynamics of that change. This means that, often, approaches are required that meet the following criteria:

• They are multifactorial in that they take multiple formal, functional and language-external/social features into consideration as potential causes for linguistic choices.

• They involve interactions between the formal, functional, and language-external/social predictors so that one can determine whether a particular predictor has the same effect regardless of other predictors’ values. While most studies simply adopt the assumption that the effects of different predictors hold independently from one another, this need not be the case, and one can only identify such cases when tests for interactions are included.

• They involve interactions of, say, Time (or Corpus) on the one hand and formal, functional and language-external/social predictors on the other hand so that one can determine whether a predictor has the same effect in each time period or whether the role a particular feature plays for speakers’ choices changes over time. Without such interactions, it is nearly impossible to make principled comparisons between different time periods.

Some studies already involve these more sophisticated approaches, usually in the form of multifactorial regression analyses. Such regression analyses try to predict the outcome of a dependent variable (or response) on the basis of one or more independent variable(s) (or predictors). Crucially, both the response and the predictors can be of different kinds, i.e., they can be binary (ditransitive vs. prepositional dative), categorical and/or ordinal (human vs. animate vs. inanimate vs. abstract), or numeric (time or length of a word in phonemes); depending on the nature of the dependent variable, one would use binary logistic regression, multinomial or ordinal logistic regression, or linear regression. Also, a central advantage of these regression models is that they allow the researcher to study the effects of several predictors (and their interactions) at the same time (Baayen Reference Baayen2008: chs. 6–7; Gries Reference Gries2013: ch. 5) so that researchers can determine which predictors affect linguistic choices significantly, in which direction (does a particular predictor make a choice more or less likely), and how strongly. In spite of these many advantages of regression modelling, there are still many studies that do not involve the proper comparisons of observed frequencies of phenomenon P in different time periods (see Gries Reference Gries, Allan and Robinson2012 for discussion of an example).

Returning to our verb-second example from above, this means that if we want to find out how verb-second word order gave way to the patterns that are in use today, neither looking at individual examples nor mere tabulated frequencies of verb-second and other orders are sufficient. Rather, we need to identify the contexts in which verb-second disappeared first, and we would need to identify the formal, functional and language-external/social features that characterize these contexts. On an abstract level, the answers that one is usually looking for derive from all three above criteria: one wants to be able to indicate that ‘during time period X, context feature Y biased speakers towards the new, incoming word order pattern with a relative strength of Z’ (Hilpert Reference Hilpert2013: 50). By analysing the impact of a range of context features over a range of time periods, we thus arrive at a differentiated picture of how the change in question proceeded. Most importantly, we learn which contextual features play an important role and which ones do not, and we can find out whether the effects of these features change in strength over time. We might also find that two contextual features interact in such a way that, for instance, they only have an effect if they co-occur, but not if they occur in isolation. Observations of these kinds are difficult, if not impossible, to make on the basis of individual examples; quantitative corpus analysis thus works like a magnifying glass, allowing the researcher to detect phenomena that would not otherwise be open to inspection.

It is important to realize that this higher level of observational detail is no end in itself: having precise information about how a given change happened is a necessary prerequisite for discussions of why the change happened in the way it did. Are we looking at a change that can be connected to social developments (Americanization, colloquialization), do the data support the idea of culture-bound, genre-specific developments (complexification, simplification), or can the change receive a structurally motivated explanation (generalization, analogical levelling)? Claims that link observations of change to these potentially competing motivations of change must be based on analyses in which alternative explanations are considered with due diligence. It is here that quantitative techniques have a decisive advantage over qualitative assessments of change: a quantitative analysis can simultaneously weigh the relative impacts of several factors, thus separating the wheat from the chaff. The analysis may for instance demonstrate that a given factor only has a very small effect, or even no effect at all, so that explanations related to that factor can be ruled out – at least for the sample that is being analysed and the population for which it is representative. Demonstrating this on the basis of qualitative data, in a way that will convince a sceptical reader, is a very difficult task. While it goes without saying that any quantitative study is of course grounded in a fundament of qualitative insights, it should equally go without saying that the analysis of language change by the numbers is an indispensable tool for extending those insights. The next two sections flesh out this statement with a number of concrete examples.

2.3 Quantitative analyses of diachronic variation in language

How does language variation in PDE compare to variation in earlier periods of English? In a study that addresses this question, Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) use the ARCHER corpus to investigate how variation in genitive and dative constructions has changed during Late Modern English. What these constructions have in common is that they are organized in paradigmatically related pairs, so-called alternations. The member constructions of an alternation are available as alternative ways of verbalizing the same, or at least fairly similar, conceptual content. Examples of the genitive alternation and the dative alternation are shown in (2).

1. (2a)

   the prince's horse

   the horse of the prince

1. (2b)

   I wrote him an email

   I wrote an email to him

Synchronic analyses of both the genitive alternation (Hinrichs and Szmrecsanyi Reference Hinrichs and Szmrecsanyi2007) and the dative alternation (Bresnan et al. Reference Bresnan, Featherston and Sternefeld2007) have identified several factors that probabilistically affect speakers’ choices between the respective alternative constructions. Those factors include semantic characteristics such as animacy, pragmatic characteristics such as the topicality/givenness of referents, and formal characteristics such as definiteness, pronominality, or length of (the referents of) possessors/possessees and recipients/patients. To illustrate the workings of just one factor with regard to the genitive alternation, the s-genitive construction is relatively less tolerant towards inanimate possessors than the of-genitive construction (?the water's temperature vs. the temperature of the water). In the dative alternation, the prepositional dative construction is relatively more tolerant towards syntactically heavy constituents in the recipient slot (?I wrote my sister, who lives in Spain, an email vs. I wrote an email to my sister, who lives in Spain). Experimental studies show that PDE speakers have internalized the complex ecologies of the determining factors in these alternations (Bresnan Reference Bresnan, Featherston and Sternefeld2007), but it stands to reason that, historically, there must have been developments leading up to the status quo. The exemplary study of Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) is a case where researchers aim to determine how these developments unfolded.

For each of the two alternations, Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) determine a variable context and retrieve all relevant examples from the ARCHER corpus. Each example is annotated in terms of a dependent variable, which marks the respective constructional choice, and in terms of several independent/explanatory variables, or predictors, such as animacy, topicality/givenness, definiteness, and crucially also the historical time period during which the example was produced. Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) then perform binary logistic regression analyses in order to obtain results that can be compared against earlier studies that analysed synchronic data, and that also indicate whether the impact of those factors has become weaker or stronger over time. Overall, the results that Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) obtain reaffirm findings based on synchronic data. The factors that are analysed show effects in the expected directions, which allows the conclusion that there has been substantial diachronic stability in the use of both genitive and dative constructions.

However, there have also been changes. For the genitive alternation, Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) find a diachronic change in the effect of the length of the possessed entity. Generally and following from a general short-before-long tendency in English, a longer possessed entity favours the s-genitive (John's sixteen-year-old stationwagon), but Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) note that the relation is non-linear when it comes to short possessed entities, especially in their early corpus data. Very short possessed entities are thus not necessarily strongly drawn towards the of-genitive. Over time, that non-linear relation becomes more linear: in the words of Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013), length is ‘more well-behaved’ in later corpus periods. A second change involves the semantic factor of animacy. Whereas s-genitives in eighteenth- and nineteenth-century English rarely occurred with collective, locative, or temporal possessors (e.g., the Academy's decision, the island's inhabitants, today's technology), the frequencies of these options sharply increase in the twentieth century, pointing to a process of semantic generalization. As for the diachronic development of the dative alternation, animacy is also shown to play a role. Inanimate recipients, as in The herbs gave the soup a nice flavour, have become more frequent in the ditransitive construction in the twentieth century.

These findings demonstrate that the probabilistic usage patterns of constructions undergo fine-grained changes that could not be detected through the comparison of individual examples, but that do lend themselves to meaningful interpretations in terms of general processes of language change.

In another study that targets change in variation, Buchstaller (Reference Buchstaller2011) investigates quotation markers in Tyneside speech on the basis of a tripartite corpus that consists of sociolinguistic interviews collected in the 1960s, the 1990s, and the late 2000s. Whereas the variation of genitive or dative constructions involves only two alternative expressions, so that the dependent variable has only two levels, matters are a little more complex in the case of quotation markers. Here, speakers can draw on a set of several forms, and the recent addition of innovative variants such as go or be like suggests that the system of quotative markers is currently undergoing a substantial reorganization. Examples of some of the variants in Buchstaller's data (Reference Buchstaller2011: 59) are given in (3).

1. (3a) I never say ‘howay man’

2. (3b) I shouted back ‘well if you stop kicking the door…’

3. (3c) I just went up to him and Ø ‘excuse me mister…’

4. (3d) She was like ‘eeh! It's a rodent!’

5. (3e) She goes ‘I might not wear them’

6. (3f) I'm all, ‘Dude, you're not helping your cause!’

Buchstaller (Reference Buchstaller2011) sets out to investigate diachronic changes in the extralinguistic and intralinguistic factors that influence speakers’ choices in that system of variants. Again, research on synchronic variation (Buchstaller and D'Arcy Reference Buchstaller2009) has identified several determining factors, such as the content of the quote, i.e. whether a thought, utterance, or noise is quoted, the grammatical tense that frames the quotation marker, the grammatical person of the quoted speaker, the distinction between narrative and other texts, and social variables such as age, social class and gender. Buchstaller (Reference Buchstaller2011) exhaustively retrieves examples of quotation from her corpus. The results identify say as the most frequent variant throughout, which however decreases over time in relative frequency with the emergence of go and be like. But how are these frequency developments reflected in a changing ecology of determining factors? In order to approach this issue, Buchstaller (Reference Buchstaller2011) first examines each factor on its own.

As for the extralinguistic factors, the quotative system in the 1960s is differentiated by gender, but not by age or social class. This subsequently changes: with go becoming more frequent in the 1990s, and be like even surpassing it in frequency in the 2000s, age and class, in addition to gender, become relevant determinants. Young women are the speakers that adopt be like to the greatest extent. As for the intralinguistic factors of quotation content, grammatical person, and grammatical tense, these exert an influence throughout the three corpus periods, but patterns of change emerge here, too. For instance, whereas say is the preferred marker of first-person quotations in the 1960s, it has ceded that role to be like in the 2000s, during which say and go show an inclination towards third-person quotations.

With a complex dataset that reflects several factors influencing speakers’ choices between several forms, the analyst has to rely on multivariate statistics to arrive at reliable generalizations. Buchstaller (Reference Buchstaller2011) performs a multinomial regression analysis (Gries Reference Gries2013: ch. 5) of the complete dataset, which reveals that the effects of age, social class, grammatical tense, and narrative are measurably different across the three subcorpora. In other words, the emergence of new variants in the quotative system of Tyneside English goes along with a reorganization of the selection processes that speakers of different age groups and different social classes make. Unlike the system of genitive and dative constructions, which undergoes just minor rearrangements, the results show that the system of English quotation markers is currently in a state of upheaval that might either stabilize or see further change through the repeated intrusion of new variants. Buchstaller's quantitative analysis (Reference Buchstaller2011) pinpoints the exact loci of change and indicates what factors change at what time. It thus gives an affirming answer to the question whether young women have been spearheading the emergence of quotative be like, but at the same time, the results offer a picture that is much more differentiated than that.

It was mentioned in the introduction of this chapter that quantitative corpus-based methods are commonly applied in order to test hypotheses. Whereas the two case studies that were described above address fairly specific research questions, their primary aim was not to decide between two rivalling hypotheses. A study that checks the validity of a pre-existing hypothesis is presented by Geeraerts et al. (Reference Geeraerts, Gevaert, Speelman, Allan and Robinson2012), who investigate the emergence of anger as a term that ousted its near-synonyms ire and wrath during Middle English. Diller (Reference Diller, Fernández, Fuster Márquez and Calvo1994) suggests a socially motivated explanation for this development, hypothesizing that anger emerged as an expression for annoyance in lower-ranked persons, as opposed to the ire and wrath of socially powerful beings such as kings or deities. From this hypothesis, Geeraerts et al. (Reference Geeraerts, Gevaert, Speelman, Allan and Robinson2012) derive the predictions that anger should be used to describe situations in which the social status of the experiencer is low, the offense affects only the experiencer, rather than having more profound consequences, and the experiencer's reaction to the offense is non-violent. Geeraerts et al. (Reference Geeraerts, Gevaert, Speelman, Allan and Robinson2012) retrieve all tokens of ire, wrath, and anger from a collection of Middle English text and annotate those in terms of the semantic factors outlined above, as well as distinguishing between tokens from religious and non-religious text and between translated and natively produced texts. The analysis further includes historical time as a variable, distinguishing examples from approximately 1300, 1400, and 1500. Analyses of each individual semantic variable across those three time periods reveal processes of change for the social status of the experiencer and the affectedness of the experiencer, but not for the violence or non-violence of the reaction to the offense. Geeraerts et al. (Reference Geeraerts, Gevaert, Speelman, Allan and Robinson2012) then use a binary logistic regression (Gries Reference Gries2013: ch. 5) to assess the combined effects of the described factors, the dependent variable is modelled as a contrast between anger and the combined tokens of ire and wrath. The results are largely in line with Diller's hypothesis (Reference Diller, Fernández, Fuster Márquez and Calvo1994). The use of anger at 1400 is favoured by contexts of personal offences with non-violent reactions, a marginally significant effect is observed for low social ranks of the experiencer. The effect of non-violent reactions is stronger in non-religious texts than in religious texts. The data further show that over time, as anger becomes the default term for the emotion it denotes, these effects weaken. Examples from around 1500 thus have a relatively higher likelihood than earlier ones to denote public offences of high-ranking experiencers that react violently to those offenses.

In summary, the case studies presented in this section illustrate three issues. First, the variationist approach to analysing the use of alternative expressions with similar functions is fruitfully transferred to the usually regression-based analysis of variation over historical time. With a diachronic corpus that represents sequential periods of English, time can be included into the analysis as one (interacting) predictor among others, and it can be determined how variation in the present compares to variation in the past. Second, this type of analysis offers nuanced accounts of what has happened, so that it can be specified what factors had an effect at what time. The contrast between the studies by Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) and Buchstaller (Reference Buchstaller2011) shows that the dynamics of diachronic variation may range from relative stability to substantial reorganization, which requires a fine level of observational granularity: an analysis has to do more than just ask whether or not a particular factor has an effect – it has to ask when this effect obtained and how it varied in strength over time. Third, the observations that these studies offer importantly include the absence of effects, which is evidence that can in principle serve to rule out hypotheses that predict those effects. An aspect that has not received much attention in the discussion above is that the findings from quantitative studies usefully feed back into the development of linguistic theories, either enriching already existing theoretical claims or generating altogether new hypotheses. The idea of using quantitative corpus-based methods to generate new ideas is taken up more extensively in the following section.

2.4 Quantitative analyses of diachronic change: exploratory approaches

An attractive potential of quantitative corpus-based methods that has yet to be fully realized in diachronic studies lies in exploratory, bottom-up approaches (Gries Reference Gries, Allan and Robinson2012). The label ‘bottom-up’ stands for a set of techniques in which the data are processed statistically in order to discover structures that had not necessarily been anticipated by the analyst. Compared to the approaches that were presented in the previous section, these methods often reverse the order of qualitative and quantitative analysis. Whereas for instance a logistic regression analysis requires a fundament of qualitative analysis which is subsequently scrutinized statistically, bottom-up approaches may start with the statistical processing of raw data, which then yields results that function as a stepping stone for a qualitative analysis. Starting with automated computational procedures has the benefit of a ‘fresh start’ that may serve to eliminate preconceptions and to reveal previously overlooked aspects of a given phenomenon.

One example for such an approach is Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012), who apply a bottom-up computational approach to the study of lexical semantic change. Whereas word meaning is usually thought of as an area of study in which the intuitions of a human analyst are completely indispensable, research in natural language processing has developed a range of methods that operationalize the meaning of a given word in terms of the elements and structures that occur in the linguistic context of that word. J. R. Firth's dictum that ‘You shall know a word by the company it keeps’ (Reference Firth1957: 11) has thus found its way into methods such as latent semantic analysis (Landauer et al. Reference Landauer, Foltz and Laham1998), which produce results that stand up to comparisons with human processing of word meaning. Latent semantic analysis uses corpus data to characterize word types in terms of frequency lists of their collocates. For instance, the noun toast frequently occurs close to nouns such as tea, cheese, slice, and coffee. A statistically processed frequency list of all collocates of toast is called its semantic vector. Semantic analysis enters the picture when semantic vectors of several words are compared. Two words are in a semantic relation if their semantic vectors are highly similar. For instance, near-synonyms such as cup and mug will have similar semantic vectors, but also converses such as doctor and patient and even antonyms such as hot and cold. If a large group of semantic vectors is analysed with a dimension-reducing technique such as multidimensional scaling (Wheeler Reference Wheeler, Köhler, Altmann and Piotrowski2005) or correspondence analysis (Greenacre Reference Greenacre2007), semantic relations between those words can be visualized in two-/three-dimensional graphs in which words with close semantic ties are positioned in close proximity whereas semantically unrelated words are placed further apart.

Whereas most applications of latent semantic analysis analyse word types, thus averaging collocate frequencies over many occurrences of the same word, Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012) use an approach that operates at the level of word tokens, thus capturing meaning differences between individual occurrences of the same word. In order to overcome data sparsity, that method uses not only the direct collocates of the target word, but also second-order collocates, that is, the collocates of collocates. Given a concordance line such as he prescribed tea and toast and a small bit of steak, the second-order collocates would include the word doctor (a collocate of prescribed) and coffee (a collocate of tea). The latter will be relatively more important, since it is also a collocate of toast itself. Applied in this way, latent semantic analysis can transform a simple key word concordance of a word such as toast into a two-dimensional scatterplot that arranges data points representing concordance lines with similar sets of context words in close spatial proximity while placing data points that have markedly different collocates further apart. Semantic patterns such as homonymy are thus reflected in different clusters of data points, yielding one cloud for tokens that signify roasted bread and a separate one for tokens signifying that people raise a glass and drink to someone's health. In their study, Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012: 171) use this procedure to investigate semantic change in the words dog and deer. The general course of the semantic developments of these elements is well-known: Old English docga semantically broadened so that the word dog today refers to not just a breed of dog, but an entire species. Conversely, Old English deor used to mean ‘animal’, today's deer has thus undergone semantic narrowing. Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012) exhaustively retrieve examples of dog and deer from the Helsinki Corpus, construct semantic vectors for each concordance line, and visualize the results using multidimensional scaling. For the word dog (and its earlier spelling variants), the resulting visualization, a scatterplot of points in a two-dimensional coordinate system, reflects the process of semantic broadening. Data points from earlier corpus data occupy a smaller, more densely populated area of the scatterplot; that area grows across the subsequent corpus periods. These results align with what is generally known about the semantic development of dog and thus vouch for the general feasibility of the method. Beyond that, they allow a glimpse into the temporal dynamics of that development that would be hard to infer from the analysis of individual examples.

For the word deer, the results are less straightforward. Instead of a systematic shrinkage of the clouds of data points over time, Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012: 177) observe successive shifts that show relatively little overlap between the different corpus periods. They interpret this as suggestive evidence that deer has undergone changes that go beyond the well-documented process of narrowing. The quantitative investigation thus prompts a more in-depth, qualitative investigation of the shifts that have taken place. What the computational procedure offers is a fresh look at data that lays bare phenomena for investigation that would have been overlooked, or perhaps considered unimportant, otherwise. Visualization techniques of the kind Sagi et al. use in that connection (see also Szmrecsanyi Reference Szmrecsanyi, Geeraerts, Kristiansen and Peirsman2010, Hilpert Reference Hilpert2011) can be of considerable help for that purpose.

The rearrangement of data to facilitate qualitative analysis also lies at the heart of an exploratory analytical method that investigates shifts in the collocational behaviour of grammatical constructions (Hilpert Reference Hilpert2006, Reference Hilpert2008). This approach draws on the method of distinctive collexeme analysis (Gries and Stefanowitsch Reference Gries and Stefanowitsch2004), which is used to contrast the collocational profiles of two or more constructions that have an open slot that accommodates different lexical types. Gries and Stefanowitsch (Reference Gries and Stefanowitsch2004: 106) exemplify the procedure with the constructions of the dative alternation. The ditransitive construction and the prepositional dative construction share a substantial number of verb types, but those shared types are not equally likely to be used in either construction. By comparing the text frequencies of both constructions against the frequencies of the verbs in either construction, verbs that significantly deviate from their expected frequencies can be identified. For instance, the ditransitive construction is significantly attracted to the verbs give, tell, and show whereas the prepositional dative construction is typically used with bring, play, and take. These preferences are in line with the idea that the two constructions differ semantically in the distance between agent and recipient (with the ditransitive construction encoding closer proximity) and with the proposal that the ditransitive construction primarily expresses that ‘X causes Y to receive Z’ whereas the prepositional dative expresses that ‘X causes Z to move to Y’ (Goldberg Reference Goldberg1995: 75–6). The purpose of a collostructional analysis is the exploratory semantic study of grammatical constructions via their most strongly attracted collocates. Applied to diachrony, the method can be used to contrast collocate sets of the same construction across a number of historical corpus periods. What the method provides are lists of significantly attracted collocates for each of the corpus periods that are analysed. Differences across those lists can be interpreted as a reflex of semantic change. If a construction broadens semantically, it will occur with a larger, semantically more diverse set of collocates. If a construction retreats into a particular semantic niche, it will increasingly occur with collocates that are semantically related to that niche.

The first process characterizes the development of the English be going to construction between the eighteenth and the twentieth century (Hilpert Reference Hilpert2008: 120). Whereas early uses of be going to attract main verbs that involve animate, intentionally acting agents as their subjects, the data from later corpus periods show how the construction broadens semantically so that the attracted elements include highly general verbs such as be or verbs such as happen, which denote spontaneous events, rather than deliberate actions.

The English future construction with the modal auxiliary shall exhibits a very different developmental trajectory. Between the sixteenth and the twentieth century, shall continually decreases in text frequency and simultaneously undergoes a change towards increased usage as a text-structuring device in expressions such as I shall return to this issue in the conclusion or I shall discuss quantum theory in Chapter 5 (Hilpert Reference Hilpert2006: 252). What this suggests is that the change in question is not necessarily semantic, but rather stylistic in nature. Like the method that Sagi et al. (Reference Sagi, Kaufmann, Clark, Allan and Robinson2012) employ to visualize phenomena of change, distinctive collexeme analysis serves to draw the analyst's attention towards those aspects of a linguistic unit that have changed over time. The quantitative method merely picks out the elements for which there is a significant difference between expected and observed frequency. A necessary second step is a qualitative analysis involving a close examination of the concrete example sentences with those significantly attracted elements, and ultimately ideally an interpretation that relates the empirical findings to a more general account of how and why the construction changed.

Other bottom-up quantitative techniques to be discussed in this overview are tools for a specific problem of diachronic corpus linguistics, namely the division of data points from different historical dates into sequential periods. All of the case studies that have been discussed up to now relied on some contrast between earlier and later data, often with intermediate stages in between. Typically, diachronic corpus data are divided into temporal stages in a way that either captures well-established historical stages of a language or, if a more fine-grained temporal resolution is desired, in a way that uses intervals of thirty to forty years to capture changes between subsequent generations of language users. Gries and Hilpert (Reference Gries and Hilpert2008, Reference Hilpert, Nevalainen and Traugott2012b) make the case that this procedure is not without its problems. By creating equidistant time periods in a top-down kind of way, the analyst may combine corpus parts that actually behave very differently, thus creating misleading statistics/trends. Thus, one approach of Gries and Hilpert's is a data-driven approach to data periodization. The basic logic of such an approach is that (1) parts of the data that exhibit similar characteristics should form part of the same corpus period and (2) breaks between different periods should be inserted at points in time where there are measurable shifts in the characteristics of the data. Thus, periods need not be equidistant, allowing for the possibility that there are longer times of stasis that are interrupted by fits and starts of development (see Figure 2.2 for one example).

Their approach is implemented as a hierarchical clustering algorithm (Gries Reference Gries2013: ch. 5). Hierarchical clustering is, like the multivariate procedures discussed earlier, a procedure that takes as its input complex datasets in which each observation (e.g., a concordance line of a particular expression in its context) is characterized in terms of a range of different variables. A common purpose of clustering approaches is to then categorize a set of n observations into m<n different, hierarchically ordered groups. For this purpose, the procedure compares observations, assesses their mutual similarities, and iteratively merges those two points of a set that exhibit the greatest mutual similarity. Of course, diachronic data require a small twist of that procedure. In a diachronic dataset, it is fully possible for two data points to be fairly similar despite the fact that they represent very distant historical periods. Gries and Hilpert's clustering algorithm can therefore only merge two data points if those data points are temporally directly adjacent, which motivates the algorithm's label variability-based neighbour clustering (VNC).

To illustrate the procedure and its advantages, Gries and Hilpert apply the algorithm to the data that informs Hilpert's analysis of shall that was discussed above. The data in question consist of collexeme strengths of the lexical verbs that occur with shall in corpus data from the sixteenth to the twentieth century, which combines parts of the Penn Parsed Corpora and the CLMET into a database of nearly twelve million words. The corpora from which the data is taken consist of six seventy-year periods which have been combined into pairs to yield three equidistant periods of 140 years in Hilpert's analysis. A question that VNC can answer is whether this particular periodization makes sense, or whether it would be more sensible to create different period boundaries. The raw input data for the VNC algorithm consists of six lists that contain the verbs occurring with shall, which add up to 1,201 types, and their respective collexeme strengths. Using a correlation statistic such as Pearson's r (Gries Reference Gries2013: sections 3.2.3 and 4.4), degrees of mutual similarity between temporally adjacent pairs can be computed. On its first iteration, the VNC algorithm finds the two lists that exhibit the greatest mutual similarity and merges the two, so that a set of only five lists remains to be analysed. The algorithm continues and finds the next pair, reiterating until all lists are merged. The result of that procedure can be visualized as a tree structure that captures mutual similarities across the six initial lists. Figure 2.1 shows the result.

Figure 2.1 VNC-based periodization of shall+V

(Figure 3 of Gries and Hilpert Reference Gries and Hilpert2008: 70)

The tree structure shows one thing very clearly, namely that lists three and four are not similar at all. What it suggests in terms of a reasonable periodization of the data would either be a binary split into an earlier and a later period, or a fourfold distinction of (1) periods 1 and 2, (2) period 3, (3) period 4, and (4) periods 5 and 6. A subsequent analysis that adopts such a periodization would have the benefit of seeing more pronounced differences between the corpus periods, so that statements about change can be made with greater accuracy and reliability.

A second practical benefit of VNC is that it can be used for the detection of outliers in historical data, that is, data points that deviate considerably from the overall trend in the data and/or from other temporally close data points, and that may therefore reflect ‘anomalies’ in the data (which in turn may result from sampling problems, author idiosyncrasies, etc.). Gries and Hilpert (Reference Gries and Hilpert2010) analyse the relative frequency of the third-person singular present tense suffixes -(e)th and -(e)s in the Parsed Corpus of Early English Correspondence, and when the relative frequency decrease of the -(e)th suffix is plotted year by year, several outlier measurements are immediately apparent to the human analyst, as can be seen in Figure 2.2. Removing outliers from datasets is a practice that is common in the empirical sciences, but ideally there should be transparent conditions on outlier removal. VNC provides such conditions. Gries and Hilpert (Reference Gries and Hilpert2010: 301) use the VNC algorithm to exclude as outliers those data points that form very small clusters (i.e. an individual year) that are surrounded by much larger clusters (i.e. more than fifty years). The logic behind this approach is simple: if a measurement is a ‘bad neighbour’, so that it differs considerably from contemporary sources which in themselves are relatively homogeneous, this is evidence that we are dealing with an outlier.

Figure 2.2 The growth of third-person singular present tense -(e)s

(from Gries and Hilpert Reference Gries and Hilpert2010)

A second type of exploratory analysis is Hilpert and Gries's (Reference Hilpert, Gries and Th2009) iterative sequential interval estimation (ISIE). This approach is a visual tool to compare the diachronic development of observed frequencies or ratios against what would be expected on the basis of prior temporal changes and their variability. From each time period to the next, the algorithm computes and plots how more or less frequent a word/structure should be (with a kind of confidence interval) such that more recent temporal developments affect predicted developments more, and then an analyst can scan the resulting plot to determine the homogeneity of the diachronic trends and where unexpected developments occur.

In sum, exploratory tools have a lot to offer to historical corpus linguistics: they can help to discern distributional structure in data invisible to the naked eye to discover trends, temporal stages, and (un)expected diachronic trends, which can then either inform more qualitative or additional quantitative analysis (see below for an example).

2.5 Desiderata for future developments

While corpus linguistics is by definition an empirical discipline and the frequencies of (co-)occurrence and dispersion statistics it provides are a natural fit with statistical methodology, the adoption of more advanced statistical tools is a slow process. Apart from very general issues that have more to do with the reporting of quantitative analyses (see Wilkinson and the Task Force on Statistical Inference Reference Wilkinson1999) that require all practitioners’ attention, in this section, we will outline a few methodological approaches whose broader adoption we think would help elevate diachronic corpus linguistics to ‘the next level’.

First, there are a variety of ways in which particularly variationist kinds of study can be improved. While the field is slowly discovering that generalized linear regression models are a tool much superior to traditional Varbrul analysis – recall our discussion in section 2.2 – important developments still await wider adoption. For example, simple regressions can, in fact should, be followed up with model criticism and evaluation to determine when levels of predictors should be conflated – does one really need to distinguish human, animate, and inanimate possessors, or is it enough to distinguish human/animate vs. inanimate (see Bretz et al. Reference Bretz, Hothorn and Westfall2010)? Also, generalized linear mixed-effects models have become used more widely in linguistics as a whole because of their abilities to incorporate speaker/writer-specific effects, lexically specific effects, and to better handle unbalanced data of the kind that corpus linguists face. Thus, this method can also be very beneficial in diachronic studies. Two studies we have already mentioned showcase the power of this method: Gries and Hilpert (Reference Gries and Hilpert2010) follow up their VNC-based data periodization with such a model to study which factors drive the emergence of the -(e)s third-person singular, and the approach allows them to obtain a classification accuracy exceeding 94 per cent. Wolk et al. (Reference Wolk, Bresnan, Rosenbach and Szmrecsanyi2013) also use this method in their studies of genitives and datives, with similarly high success. While high classification accuracies are not the ultimate goal of these studies, they reflect that with the right tools, all statistics will be more precise and, thus, more relevant to the task at hand, understanding what facilitates/inhibits change. As the development of mixed-effects models (and generalized estimating equations) matures, this method will provide ever more useful results; see also Rietveld et al. (Reference Rietveld, van Hout and Ernestus2004) for more discussion of pitfalls in quantitative corpus research.

As another example, other more sophisticated tools that can improve diachronic corpus studies involve related methods that can handle curvature/nonlinearities in the data, a frequent characteristic of diachronic data. Hilpert and Gries (Reference Hilpert, Gries and Th2009) discuss regression with breakpoints as one rather simple tool, but down the road more advanced methods – polynomial regressions, splines, generalized additive models (see Zuur et al. Reference Zuur, Ileno and Smith2010 for discussion) – will also be indispensable to the quantitative study of historical corpus data.

Similar advances in the domain of exploratory tools await adoption in diachronic corpus linguistics. There is now a variety of methods to follow up on cluster-analytic results. Dendrograms such as Figure 2.1 can be studied with regard to (1) how many clusters should be distinguished (using so-called average silhouette widths), (2) how ‘clean’ the resulting clusters are (with F-values), (3) which features are most distinctive for the clusters (using t-scores), and (4) how well they map onto other cluster-analytic results regarding the same phenomenon; see Divjak and Gries (Reference Divjak and Gries2008) for exemplification and discussion.

Finally, the statistical area of robust statistics is potentially also very useful to the study of historical corpus data. Robust statistics are statistics that are less based on the assumptions underlying many traditional statistical tools (normality, homogeneity of variances, no outliers, etc.), or that are less vulnerable in the presence of such violations, which are the rule in the kind of noisy observational data that diachronic corpus linguists study. Fields such as second-language acquisition have begun to discover this area (Larson-Hall and Herrington Reference Larson-Hall and Herrington2009), and in diachronic corpus linguistics, some work is under way. For instance, Lijffijt et al. (Reference Lijffijt, Papapetrou, Puolamäki and Mannila2011) develop an approach to the study of text frequency change that dispenses with the bag-of-words assumption. Much like language itself, diachronic corpus linguistics will continue to evolve.

2.6 Concluding remarks

This chapter has argued that quantitative methods hold considerable potential for diachronic corpus analysis. There are two main selling points. First, in order to make sense of the complex variation that is at play in processes of language change it is a simple matter of necessity to have analytical tools that can cope with that complexity and that offer the analyst a nuanced view of what happened. If we want to understand why a certain change happened, thorough understanding of how it happened is the first step towards that goal. Second, quantitative analytical methods can make phenomena visible that would otherwise not be open for inspection. These methods can offer a fresh, unbiased look at phenomena that seem familiar, but which still remain to be fully understood. Importantly, it is early days for diachronic corpus linguistics. All of the methods discussed in this chapter are still in a phase of development, awaiting further testing and replication, and we can look forward to studies that will further enlarge the toolkit of diachronic corpus linguistics in the near future. At the same time, it has to be pointed out that even a high level of analytical sophistication cannot remedy the problem of data sparseness that is one of the natural limits of endeavour in historical linguistics. Evidently, any analytical method can only produce satisfying results on the basis of rich empirical data and analysts who are aware of the inherent restrictions of their methods. When the historical record is poor, the best shot that we have at nonetheless understanding it to some degree is to take present-day variation as a model, and to see whether the historical data varies in comparable ways. To make this happen, historical corpus linguists and sociolinguistically oriented corpus linguists need to join forces both at the level of methodology and at the level of linguistic theory. Pioneering work in that direction has been carried out (Nevalainen and Raumolin-Brunberg Reference Nevalainen and Raumolin-Brunberg2003) and the recent studies that were discussed in this chapter show that the problems and pitfalls that haunt diachronic corpus linguistics are being addressed from a variety of angles.

3.1 Introduction

Within the space of about twenty years, English historical linguistics has been deeply changed and enriched by the introduction and expansion of pragmatic approaches. While, in the previous decades, English historical linguistics was concentrated on reconstructing and mapping forms, the 1980s and 1990s were the time when it started to focus on ‘meaning-making processes in past contexts’, i.e. in how meaning is negotiated, and how more is conveyed than what is said, in texts from the past (Taavitsainen and Jucker Reference Taavitsainen, Taavitsainen and Pahta2010: 6). Historical pragmatics can be defined as ‘a field of study that wants to understand the patterns of intentional human interaction (as determined by the conditions of society) of earlier periods, the historical developments of these patterns, and the general principles underlying such developments’ (Jucker Reference Jucker2008: 895). The main aims of historical pragmatics can be subsumed under two overarching research questions:

1. 1 What do texts tell us about the way in which pragmatic functions (i.e. context-bound, negotiated meanings) were encoded in language in the past?

2. 2 What can we reconstruct about the way in which pragmatic values of language forms develop over time in a language? (Jacobs and Jucker Reference Jacobs, Jucker and Jucker1995: 5–6)

Historical pragmatics considers language as ‘diachronic, social and dynamic and not just as a synchronic and static system’ (Taavitsainen and Jucker Reference Taavitsainen, Taavitsainen and Pahta2010: 4). It is therefore a field in which the context and the aims of communicative acts are considered important factors in the analysis of language forms; historical pragmatics focuses particularly on the specific combination of contextual features and sets of communicative intentions in each text, rather than on broad and relatively stable categories. As in other branches of pragmatics, the interface with semantics is strong, with a focus on diachronic change in semantic values, especially connotational ones. To give just one example, if we consider the English term of address sir, e.g. in dramatic dialogue or in letter openings, we can study its distribution in texts considering who addresses whom, in terms of registers (was it more or less formal in the past than it is today? Was it used more to social superiors, to equals, or to inferiors?), or we can study its change in semantic values over time (especially if we consider its relative sirrah, which started as a ‘neutral’ term of address for young men and later took on a pejorative meaning); within historical pragmatics, we would look at instances of dialogue in which the term occurs, and consider the relationship between the characters involved, the specific context of situation, and the apparent communicative intention represented in the relevant co-text, to try to determine what the term represents in that particular exchange, e.g. if it aims at conveying deference, or insult, etc. (Mazzon Reference Mazzon, Taavitsainen and Jucker2003b, Reference Mazzon2009: 26).

The present chapter first reviews the thematic and chronological development of English historical pragmatics and the input it received from neighbouring fields; then, it mentions some of the data that are studied in historical pragmatics and some of the most promising results; finally, it discusses the challenges and limitations of this approach as against the aims outlined.

3.2 The development of historical-pragmatic and sociopragmatic approaches

This development had several precursors, i.e. studies that were not unified by any theoretical framework, but that are still the starting point of much current research. The decisive step towards the present expansion of the field was the application of theories and frameworks from synchronic European pragmatics (itself recognized as a separate field only about a decade earlier) to documents from the past. Authoritative and comprehensive works on (particularly English) historical pragmatics started to be published in the early 1990s (a precursor was Stein 1985b; see, e.g., Busse Reference Busse1991, Ehlich Reference Ehlich, Watts, Ide and Ehlich1992, the fundamental Jucker Reference Jucker1995, and the Journal of Historical Pragmatics, which was started in 2000) and are now being published at a growing rate (see Jucker and Taavitsainen Reference Jucker, Fried, Östman and Verschueren2010, relevant chapters or sections in Bergs and Brinton Reference Bergs and Brinton2012 and in Nevalainen and Traugott Reference Nevalainen, Bergs and Brinton2012); the corresponding multiplication of conferences, workshops, and projects in this area further testifies to the liveliness of this field.Footnote ¹ The subfields that most directly contributed to what, already some years ago, Jucker et al. (Reference Jucker, Jucker, Fritz and Lebsanft1999b: 14) defined as the ‘pragmaticization of English historical linguistics’, were intercultural (or cross-cultural) pragmatics (Blum-Kulka et al. Reference Blum-Kulka, House and Kasper1989, Wierzbicka Reference Wierzbicka1991) and variational pragmatics (Schneider Reference Schneider, Fried, Östman and Verschueren2010, Pichler Reference Pichler2013); within these branches, scholars realized the potential (and subsequently the limitations) of studying the pragmatics of groups separated by a distance.Footnote ² The possibility of developing comparative studies of different communities concerning single pragmatic issues stimulated the idea that similar efforts could be directed at studying groups separated by time rather than by space or culture. This was followed by input from relational and interactional pragmatics, Gricean pragmatics, and politeness theory (see Kopytko Reference Kopytko1993, Traugott Reference Traugott, Horn and Ward2006: 540–8, Jucker and Taavitsainen Reference Jucker, Nevalainen and Traugott2012), and, more recently, from interpersonal pragmatics (Locher and Graham Reference Locher and Graham2010) and impoliteness theory (Culpeper Reference Culpeper2011).

Another element that contributed to historical pragmatics was the development of historical sociolinguistics, a branch that has also grown rapidly (see Chapter 1 by Romaine in this volume). Particularly relevant to historical pragmatics are studies based on network analysis (Milroy Reference Milroy1992a), or on the partly overlapping concept of ‘community of practice’, where specific linguistic markers contribute to defining group membership (see, e.g., McColl Millar Reference Millar2012: 9).Footnote ³ Nevalainen (Reference Nevalainen, Bergs and Brinton2012) reports on historical applications of other approaches, such as sociology of language and social dialectology (see also Chapter 17 by Nevala in this volume), or correlational sociolinguistics and interactional sociolinguistics (in turn, influenced by functional approaches; Palander-Collin et al. Reference Palander-Collin, Nevala, Nurmi, Nurmi, Nevala and Palander-Collin2009: 1–2).Footnote ⁴

In particular, interactional sociolinguistics has contributed to historical pragmatics by adding the micro-elements of personal background, social networking, and communicative function to the analysis of correspondence (Fitzmaurice Reference Fitzmaurice2002c, Nevalainen and Raumolin-Brunberg Reference Nevalainen and Raumolin-Brunberg2003, Tieken-Boon van Ostade Reference Tieken-Boon van Ostade, Juan and Hernández-Campoy2005), and has offered new insight into dialogue studies, e.g. by tracing power dynamics in interaction, also in terms of ‘conversational dominance’ (Itakura Reference Itakura2001, Locher Reference Locher2004). This ‘cross-fertilization’ has been termed sociopragmatics, and, according to one recent definition, ‘focuses on the way in which speakers exploit [socially conditioned language] norms to generate particular meanings, take up particular social positionings’ (Culpeper Reference Culpeper, Andreas and Taavitsainen2010: 73; see also the studies in Pahta et al. Reference Pahta, Nevala, Nurmi and Palander-Collin2010). The study of genres and text-types, connected to the notion of language register, is within the subfield of sociopragmatics, and so is the study of politeness conventions and facework within discourse communities and, as mentioned, ‘social networks’, keeping in mind the impact of class- and gender-related variation.

Historical discourse analysis is closely related to, and often overlaps with, historical pragmatics (Brinton Reference Brinton, Schiffrin, Tannen and Hamilton2001, Taavitsainen Reference Taavitsainen, Fanego, Méndez-Naya and Seoane2002, Hiltunen and Skaffari Reference Hiltunen and Skaffari2003, Skaffari et al. Reference Skaffari, Peikola, Carroll, Hiltunen and Wårvik2005). It principally deals with the fact that language items diachronically take on (and may eventually lose) discourse-organizing functions. At the same time, the existence of separate discourse communities, practising different discourse styles, can be traced back to the past – in fact, the features identifying text-types and discourse levels can be even more prominent for times in which there was a strict separation between ‘clerics’ and (largely illiterate) lay people. Courtroom discourse (Archer Reference Archer2005), scientific and didactic discourse (Fries Reference Fries, Borgmeier, Grabes and Jucker1998, Taavitsainen and Pahta Reference Taavitsainen, Taavitsainen and Pahta2004), religious discourse (Kohnen Reference Kohnen, Pahta, Taavitsainen, Nevalainen, Tyrkkö, Jucker, Schreier and Hundt2009, Rütten Reference Ratia2011), and news discourse (Raymond Reference Raymond2006, Valle Reference Valle and Brownlees2006), have received the highest degree of attention so far, with the contribution of historical stylistics, which also partly concentrates on text-type-specific features (Studer Reference Studer2008; see relevant chapters in Jucker and Taavitsainen Reference Jucker, Fried, Östman and Verschueren2010 and Claridge Reference Claridge, Bergs and Brinton2012; see also Chapter 16 by Taavitsainen in this volume). Historical discourse analysis investigates the codification of conventions pertaining to specific text types as indicative of the existence, degree of cohesion, and diachronic change of ‘communities of practice’ and discourse communities. It studies, for instance, degree of formality, information structure, and level of personal involvement and stance marking in various text types, addresser–addressee relationships, text-specific language structures, and vocabulary items and their developments (Diller and Görlach Reference Diller, Diller and Görlach2001, Claridge Reference Claridge, Bergs and Brinton2012). It also studies the ‘discursivization’ paths of specific items, e.g. connectors, markers, etc. (Jucker and Taavitsainen Reference Jucker, Nevalainen and Traugott2012: 302–3), which often involves meaning shift and a range of processes that lead a language item to acquire discourse-internal meanings and/or functions; this step can also be followed by a further shift towards subjectification, i.e. ‘towards meanings increasingly situated in the speaker's subjective believe-state or attitude’ (Brinton Reference Brinton, van Kemenade and Bettelou2006: 310; for original theorizations, see Traugott Reference Traugott, Lehmann and Malkiel1982), and further towards intersubjectification, i.e. ‘subjectified meanings may in turn be recruited to convey the speaker's attention to the addressee’ (López-Couso Reference Lange2010: 131).

3.3 Aims, objects of study, methods, and results

The first research question mentioned above (i.e. how were pragmatic functions encoded in the past?) has produced mostly synchronic studies of one period, text type, or communicative environment, e.g. investigating how terms of address conveyed social and pragmatic values in EModE, or how textual organization was encoded in correspondence between eighteenth-century scientists. Many such studies are within pragmaphilology (a term introduced by Jacobs and Jucker Reference Jacobs, Jucker and Jucker1995), i.e. they concentrate on single texts or groups of texts, taking into account not only internal elements but also the extralinguistic context of its production and use (Traugott Reference Traugott, Horn and Ward2006: 538, Jucker Reference Jucker, Fried, Östman and Verschueren2010); examples of such studies are Mazzon (Reference Mazzon and Hart2003a), which studies the area of modal meanings within a restricted time span and in a relatively homogeneous set of text types (i.e. Chancery documents), Pakkala-Weckström (Reference Pakkala-Weckström, Jucker and Taavitsainen2010), and Busse and Busse (Reference Busse, Busse, Jucker and Taavitsainen2010), where the emphasis is on the pragmatic meanings encoded by individual authors, respectively Chaucer and Shakespeare, through their use of terms of address, speech acts, discourse markers, and other relevant items.

The second type of question (i.e. how do the pragmatic values of a language item develop over time?), which is addressed by the approach often separately defined as diachronic pragmatics (Jacobs and Jucker Reference Jacobs, Jucker and Jucker1995), has led scholars to investigate, for instance, how a main clause with a propositional content such as you know or I mean, or an adverb like well, lose their literal meanings and become discourse markers with textual or interactive functions over time (form-to-function mapping), or, conversely, how a certain communicative function (e.g. greeting or apologizing) has changed its realizations over time (function-to-form mapping; see Jucker Reference Jucker, Fried, Östman and Verschueren2010: 110). Most such studies compare occurrences of the relevant items in various contexts, and draw on a variety of external and internal factors (e.g. relation between the interlocutors/correspondents/characters, or specific, independently ascertained communicative aim of the dialogue) in order to assess the pragmatic values conveyed.

3.4 Challenges and limitations

The challenge faced by historical pragmatics is related first of all to the nature of the evidence: pragmatic values are more likely to emerge in speech, while our evidence is at the utmost ‘speech-like’ written text; moreover, there is the problem of the piecemeal preservation of only a few text-types in patchy distribution, mostly produced by a narrow subset of text-producers. A further problem is the lack of ‘common ground’ between the researcher and the text-producers, a limitation first identified by intercultural pragmatics (for a diachronic perspective see Jucker and Taavitsainen 2012: 300; see also Chapter 15 by Fitzmaurice in this volume); this reduces drastically the amount of information on the communicative situations investigated. In other words, the challenge is that of trying to glean from texts something that goes beyond the literal without having direct access to the world knowledge and to the ‘communicative competence’ of the speakers who produced those texts, or of the recipients of such communicative acts (Taavitsainen Reference Taavitsainen, Bergs and Brinton2012b: 1464). This challenge has been taken up widely, and historical pragmatics, sociopragmatics, and discourse analysis are today among the liveliest branches of English historical linguistics; at the same time, the related limitations, hurdles, and possible pitfalls briefly mentioned here, are only escapable if we consider written texts as valid objects of study in their own right (Jucker et al. Reference Jucker, Jucker, Fritz and Lebsanft1999b: 16), i.e. without yielding to the temptation of considering the language indicators present in texts in the same way as we consider recorded spoken language tokens. A useful review of the problems in historical pragmatics is offered by Taavitsainen and Jucker (Reference Taavitsainen, Taavitsainen and Pahta2010: 15–20) – at the forefront is a semasiological problem, related to the difficulty of identifying the precise function of a certain form, and of determining the intended meaning of an item in a text or group of texts belonging to a certain time, space, and text type: how can we hope to reconstruct the exact shades of meaning (or layers of meaning) conveyed by a certain item in a context whose specifications are largely unknown to us?

This problem is intertwined with another major methodological problem, especially of sociopragmatics, i.e. overcoming the dichotomy between the macro-level of social structure and its linguistic expression on the one side, and the micro-level represented by the piecemeal and indirect evidence of individual texts on the other.Footnote ⁶ There is an overall problem of generalizability and representativeness within a research field formed by scattered and hybrid evidence (see Section 2 of this volume), given the impossibility of accessing comparable amounts of evidence from different groups, and to the difficulty of creating solid social models and of categorizing ‘informants’ in any uncontroversial way, with the partial exception of gender (see, e.g., Nurmi Reference Nurmi, Granger and Petch-Tyson2003a, Lutzky Reference Lutzky, Mazzon and Fodde2012; see also Chapter 1 by Romaine in this volume).

A further challenge is trying to overcome the inevitability of superimposing the modern perspective, i.e. reading history backwards in view of ensuing developments, a danger of all historical research (Auer and Voeste Reference Auer, Voeste, Hernández-Campoy and Conde-Silvestre2012). This is made more difficult by the haphazard, socially conditioned, or editorially manipulated survival of documents, which makes more acute the dangers of falling into ‘historiographic elitarianism’ and of feeding various possible myths of a unified history of the language (Milroy Reference Milroy, Watts and Trudgill2002, Watts Reference Watts2011). Another considerable range of problems, as mentioned, comes from the danger of ‘reification’ or overgeneralization of results, especially related to the gap between spoken and written language. Deeper reflections on the orality–literacy relationship (e.g. see the precursor studies in the collection Stein Reference Stein1992; and later Arnovick Reference Arnovick2006, Taavitsainen and Fitzmaurice Reference Fitzmaurice2007: 18–21, Schaefer Reference Schaefer, Bergs and Brinton2012) are contributing to the development towards a more realistic and heuristically correct stance in this respect.

By way of example of the limitations of the field, let us reconsider the advances in historical speech-act analysis, which are based on the interpretation of the intended speech act as it emerges from the text. Of course, there are possible pitfalls in such interpretations, as the classification of acts often relies on the presence of explicit performative verbs (e.g. promise), as well as of additional elements such as modality items and other boosters or downtoners (e.g. absolutely), from which one can infer the relative force of the speech act. Recent research shows sensitivity to this potential circularity, while extensive corpus studies on wider samples allow for a more precise mapping of the possible illocutionary force conveyed by specific items in certain text types. As in the case of PDE speech-act analysis, where exhaustive and uncontroversial classifications do not seem possible, the key to such descriptions could lie in the adoption of a prototypical approach that allows us to establish categories that are not watertight, but relatively fuzzy (Taavitsainen and Jucker Reference Taavitsainen2007: 108, Culpeper and Archer Reference Culpeper, Archer, Jucker and Taavitsainen2008: 47); this is valid especially when it comes to the reconstruction of indirect speech acts, which can often only be revealed by the analysis of metapragmatics and of the whole ‘pragmatic space’, i.e. adjacent speech acts (Culpeper and Archer Reference Culpeper, Archer, Jucker and Taavitsainen2008: 57ff.).

3.5 Conclusion

As I hope to have shown in these pages, historical pragmatics has contributed a great deal to historical linguistics; in the last few years, acting synergistically with historical sociolinguistics and corpus linguistics, it has expanded and produced interesting results. There is no doubt that a lot is still to be done; the metapragmatic level mentioned above represents one of the subfields that looks more promising for future research, and so do dialogue studies, as well as all those approaches that combine sociolinguistic, discourse-analytic, conversation-analytic, and pragmatic methodologies to increase our understanding of texts as dynamic products. In particular, a promising direction seems to lie in the integrated analysis of several indicators whose combination conveys a specific pragmatic effect. In short, historical pragmatics represents a possible way in which we can try to fill the linguistic and sociocultural gap that separates us from texts produced in the past, allowing us to consider them not as fixed objects, but as the expression of culturally determined communicative processes.

4.1 Introduction

This chapter provides a review of some of the ways in which the framework of construction grammar has been used to explain patterns of change in English, with a focus on morphological, morphosyntactic, and semantic change. While there are a number of variants of construction grammar, the research reported on here takes a largely usage-based, cognitive approach to the architecture of language (see Langacker Reference Langacker1987, Bybee Reference Bybee2010); some comments relating to the use of a more formal model of construction grammar in accounting for language change appear in section 4.4. The chapter is organized as follows. Section 4.2 outlines some of the principles of constructional approaches to language (see Goldberg Reference Goldberg, Hoffmann and Trousdale2013), and establishes those features which are particularly relevant to language change (on which, see Fried Reference Fried, Hoffmann and Trousdale2013, Hilpert Reference Hilpert2013, Traugott and Trousdale Reference Traugott and Trousdale2013). The objective is to sketch briefly how early constructional approaches challenged some of the dominant linguistic theories in the late twentieth century, then go on to show their application to historical linguistics. Some reference is made to methodological issues in a usage-based approach to language change, but for more detail see Chapter 2 by Hilpert and Gries in this volume.

Section 4.3 considers applications of construction grammar to areas of English historical linguistics, focusing on the development of and changes to morphological and syntactic schemas, and the development of grammatical and lexical micro-constructions. The focus is on how construction grammar helps to elucidate some problematic issues in theories of language change, but also consider ways in which current approaches to constructional change might be refined or developed.

The discussion of morphological schemas (section 4.3.1) is couched within the framework of constructional morphology (see particularly Booij Reference Booij2010). The focus will be on quantitative approaches to productivity changes in the history of English (the V-ment construction, Hilpert Reference Hilpert2013), and the relationship between lexicalization and the development of word-formation schemas (the X-dom construction; Haselow Reference Haselow2011, Traugott and Trousdale Reference Traugott and Trousdale2013). The subsection on syntactic schemas (4.3.2) looks at changes to existing constructions (illustrated by semantic specialization in the English ditransitive construction; see Colleman and De Clerck Reference Colleman and Clerck2011) as well as the creation of new constructions (illustrated by the development of cleft constructions in English (Traugott Reference Traugott, Cooper and Kempson2008, Patten Reference Patten2012)). The section on the loss of schemas (4.3.3) focuses on reduction in constructional space, such as when a productive word-formation schema or argument structure construction falls into disuse. Examples are taken from recent work on constructional change.

The fourth section provides a brief overview of the advantages and disadvantages of a constructional approach to change, with some comments regarding future directions of research. A brief concluding section summarizes the main issues discussed.

4.2 Construction grammar: an introductory sketch, and its application to historical linguistics

Goldberg (Reference Goldberg, Hoffmann and Trousdale2013) identifies the following as features common to varieties of construction grammar which distinguish them from mainstream generative approaches to the architecture of human language:

• From lexical items to phrasal constructions, language is a system of linked, conventionalized, form–meaning pairings.

• There are no operations (such as merge or move) which transform one structure into another.

• Language is a conceptual network (see also Hudson Reference Hudson2007 for a related model) in which inheritance and extension links serve to associate one constructional node with another.

• Language is a variable phenomenon; similarities across languages can be accounted for either by properties of the constructions themselves, or by ‘domain-general cognitive processes’ (Goldberg Reference Goldberg, Hoffmann and Trousdale2013: 16).

Goldberg then notes one other feature that is shared by many but not all constructional approaches, namely that knowledge of language is a product of language use (the ‘usage-based model’). This is an important feature for work in historical linguistics, and contrasts sharply with some other approaches to language change. One area of intersection between historical linguistics more generally, and construction grammar, is historical–comparative reconstruction. I do not discuss this topic in the present chapter (see Barðdal Reference Barðdal, Hoffmann and Trousdale2013 for a summary of current thinking).

The features listed above serve to establish what is generally shared across the various ‘constructional’ approaches to language, such as Radical Construction Grammar (e.g. Croft Reference Croft2001) and Cognitive Construction Grammar (e.g. Goldberg Reference Goldberg2006). As is generally the case for linguistic theories, construction grammar was originally designed as a tool to model the linguistic knowledge of speakers synchronically, rather than as a theory of language change. It developed as a reaction to mainstream generative linguistics in the United States in the 1980s and 1990s, and shared some – but not all – features with other cognitive theories of language being developed at the time, including Notional Grammar (e.g. Anderson Reference Anderson1977), Word Grammar (e.g. Hudson Reference Hudson1984), and Cognitive Grammar (e.g. Langacker Reference Langacker1987), all of which, like construction grammar, have since been used to account for historical changes. Fillmore (Reference Fillmore, Hoffmann and Trousdale2013: 111) notes that Berkeley Construction Grammar developed from work ‘centered on discovering the idiomatic and “irregular” parts of language, demonstrating their frequency in text and their centrality in the linguistic knowledge of speakers’. Precisely such topics are of concern to historical linguists, as the following suggests:

• Work on grammaticalization (e.g. Lehmann Reference Lehmann1985, Hopper and Traugott Reference Hopper and Traugott2003) and lexicalization (e.g. Brinton and Traugott Reference Brinton and Traugott2005) has tried to explore how the ‘idiomatic’ parts of language are associated with both procedural meaning (e.g. the aspectual composite predicates such as take a walk) and contentful semantics (e.g. other composite predicates such as take offence at).

• The issue of what is ‘regular’ and ‘irregular’ (and how something that was irregular at one point comes to be regular later in the history of a language) is at the heart of debates surrounding the relationship between reanalysis and analogy (see Fischer Reference Fischer2007 and De Smet Reference De Smet2009).

• Frequency – including changes in frequency – has been of importance to usage-based approaches to the structure of language (see much of the work of Joan Bybee, especially Bybee Reference Bybee2007, Reference Bybee2010). This has come to be of particular importance as work in sociolinguistics has influenced theories of language change, with the adoption of particular kinds of quantitative methodologies in studies of language change (see Chapter 2 by Hilpert and Gries in this volume).

• Textual evidence (whether original manuscripts or computerized corpora) provides most of the data for work in historical linguistics. Crucially, the importance of both cotext and context (see Bergs and Diewald Reference Bergs, Diewald, Bergs and Diewald2009) for understanding how new grammatical forms develop is of significant importance for usage-based models.

• Establishing what speakers know unites practically all fields of linguistic enquiry. Work in historical linguistics interfaces with work on language acquisition and language contact, and both of these topics have been addressed within a constructional model.

Thus the kinds of questions relevant to (English) historical linguistics show considerable overlap with those that have driven aspects of work in construction grammar from its outset. More recently, the issue has arisen as to what precisely a ‘diachronic version’ of construction grammar would look like, and what questions in historical linguistics a constructional model could answer. For example, one topic which has been hotly debated in recent years has been the relationship between grammaticalization and constructional change (Noël Reference Noël2007, Gisborne and Patten Reference Gisborne, Patten, Narrog and Heine2011, Fried Reference Fried, Hoffmann and Trousdale2013, Hilpert Reference Hilpert2013, Traugott and Trousdale Reference Traugott and Trousdale2013). Another issue has concerned how change might be modelled within a particular variant of construction grammar (e.g. Fried Reference Fried, Bergs and Diewald2008). In what follows, I present an overview of some of the issues which appear to be particularly pertinent to work on the history of English, and use only English case studies as a source of data. There will be minimal reference to methodological issues, given that this topic is covered elsewhere in the present volume (see Chapter 22 by Traugott in this volume), but it is important to underline the close interplay between the methods adopted and the research questions which drive particular projects. Quantitative and qualitative approaches are complementary, not competing, and both provide insights into how constructions emerge, change, and fade over time.

4.3 Constructional change

In this section, an overview of different kinds of constructional change is provided. Section 4.3.1 is concerned with the emergence of and change to word-formation schemas in English; section 4.3.2 is concerned with similar features associated with grammatical constructions such as cleft constructions. In section 4.3.3, the focus is on loss, and what happens to isolated micro-constructions when a schema disappears over time.

4.4 Comparisons with other accounts of change and future directions

As discussed in section 4.3, some researchers have proposed ways in which a constructional model of language (change) has advantages over other models, and as noted in the first section of this chapter, there are certain ways in which the very fundamentals of construction grammar set it apart from other frameworks. While some differences (on modularity, and on the precise relation between use and structure, for example) are likely to remain contentious for some time, there are other ways in which some constructional approaches to change and some generative approaches have independently reached similar conclusions.Footnote ⁵ One such area concerns the relative importance of reanalysis and analogy in change. Both Traugott and Trousdale (Reference Traugott, Jucker and Taavitsainen2010) and Roberts (Reference Roberts, Traugott and Trousdale2010), for example, privilege reanalysis above analogy. There also appears to be convergence on what it means to say that change is gradual. For example, Traugott and Trousdale (Reference Traugott and Trousdale2013) recognize that constructionalization involves a sequence of changes, but that each individual micro-step is discrete; this appears to be consistent with the nature of upwards reanalysis in a generative model of change which relies on a clausal hierarchy in which category distinctions are very fine-grained (Cinque Reference Cinque1999, Roberts Reference Roberts, Traugott and Trousdale2010).

In terms of future directions, there are many possibilities. One concerns the relationship between micro-constructions and the schemas with which they are aligned, and the degree of granularity at which changes occur. For instance, there is general consensus that change begins in constructs (understood as tokens, attested instances of use), and that a systematic change involves the creation of a new micro-construction (low-level types). But if constructions exist in a taxonomic network, how far do the effects of change spread (both in terms of extensions to other micro-constructions, and in terms of the more general schemas that sanction micro-constructions)? It is recognized that constructional templates vary in their degree of specificity but to what extent and in what way are the more abstract templates affected by change at a micro-constructional level, and how would this be measured? Here it is likely that the kind of quantitative work associated with (diachronic) collostructional analysis (see, e.g., Hilpert Reference Hilpert, Nevalainen and Traugott2012b, Stefanowitsch Reference Stefanowitsch, Hoffmann and Trousdale2013) will shed some light on the effects of change. A related issue is the extent to which patterns which could be brought under a single schema are indeed categorized as such by speakers, or whether speakers treat such relations as a kind of family resemblance. This is connected with the kinds of claims made regarding differences between it-cleft and pseudo-cleft constructions above, where the former is treated as subschema, and the latter as a set of separate micro-constructions. The issue of relationship between change in the mental representation of the individual speaker (i.e. the constructional knowledge characterized by an idiolect) and the change in the ‘linguistic network of a community of speakers’ (Traugott and Trousdale Reference Traugott and Trousdale2013: 22) also needs to be considered in greater depth.

The relationship between a formal model of change and a constructional model of change was briefly addressed above. Formal models tend to have an advantage over non-formal models in terms of the preciseness of the representation of the grammar. Some constructional models (e.g. Sign-Based Construction Grammar; Sag Reference Sag, Boas and Ivan2012, Michaelis Reference Michaelis, Hoffmann and Trousdale2013) have a specific formal representation, and one possible future direction is to see whether and how some of the changes described in the literature on constructional change could be modelled in that framework. Fried (Reference Fried, Bergs and Diewald2008) has articulated some principles of formal constructional change as applied to aspects of the history of Czech in a related constructional model; there appear to be fewer such descriptions of changes in English.

In this final section I have had rather less to say about English and more about more general issues of historical linguistics. There are clearly many other areas of English grammar which could be explored using a constructional model. Indeed, there already have been many such studies on English and related languages (e.g. on raising to subject and raising to object (Noël and Colleman Reference Noël and Colleman2010); on passive and copular constructions (Petré Reference Petré2014); causatives (Hollmann Reference Hollmann2003); on future constructions (Hilpert Reference Hilpert2008), to name just a few). Clearly the greater the number of different case studies, the more hypotheses can be tested across different data sets. Furthermore, both in terms of synchrony and diachrony, construction grammar has had rather less to say about phonological change than about change at any other level.

4.5 Conclusions

Many of the principles that are shared across different variants of construction grammar have been explored from the perspective of language change, and particularly in terms of changes affecting English constructions. Some of these changes have occurred at one ‘level’ in the construction (e.g. changes affecting the semantics of a lexical construction, or a grammatical schema); others have involved the creation of new constructions (including word-formation schemas, and information-structuring constructions like clefts). These topics have been approached using quantitative and qualitative methods; both methodologies have provided rich insights into the nature of constructional change, the relationship between constructional change and other accounts of change, and some of the ways in which English has evolved over time. New directions might include a more precise formalization of aspects of change, research into ongoing changes, particularly in new varieties of English that have been the product of substantial contact between speakers.

5.1 Introduction

In this chapter, I will outline the basic philosophy behind generative grammar and how that view has an impact on the relationship between generative grammar and English historical linguistics. I provide an overview of the shifts that generative grammar has undergone in the last sixty years, from a system with highly specific phrase structure rules to a system of feature checking. The discussion will mostly be limited to morphosyntactic change and will only briefly mention phonological change. I will finish with a discussion of the advantages and the disadvantages that this model offers to linguists working in the history of English and vice versa.

Generative grammar has its beginnings in the late 1950s with the work of Noam Chomsky and emphasizes innate linguistic knowledge, or Universal Grammar. Children use their innate knowledge and, on the basis of the language they hear spoken, also known as the E(xternalized)-Language, come up with a grammar, also known as the I(nternalized)-Language (see Chomsky Reference Chomsky1986: 19–24). Generative grammar focuses on the ability of native speakers to speak and understand grammatical sentences.

Since the focus of generative grammar is on the linguistic knowledge present in the mind of a native speaker, many generativists think that it is impossible to study historical stages of a language because native speakers of, for instance, Old English are unavailable. This lack of native-speaker intuitions is the main reason behind the criticisms one encounters using data from Old, Middle, and later English in a generative framework.Footnote ¹

Because generative grammar has evolved in the last sixty years, I will provide some background on early work but then focus on the two most recent approaches, namely the Principles and Parameters (P&P) model of the 1980s and 1990s and the Minimalist Program of the 1990s to the present. Principles hold true for all languages and include, for instance, the basic shape of phrases (with a head and a complement), the requirement that movement is local, and that sentences are hierarchically structured rather than linearly. Parameters account for cross-linguistic differences and are to be set depending on the evidence the learner is faced with. If, for instance, the language the learner hears places the object before the verb, the parameter is set as Verb-final.

The P&P approach to historical linguistics examines differences that exist in the syntax of different stages, such as variant word order and the presence or absence of Verb-movement. A change in word order is seen as a parametric change from Object–Verb (OV) to Verb–Object (VO). Old English has OV in (1) because the object medoful ‘cup’ precedes the verb ætbær ‘brought’ but Modern English has VO in (2) because the verb precedes the object. This change is accounted for by a resetting of the headedness parameter from Verb-final to Verb-initial.

1. (1)

   þæt	hio	Beowulfe …	medoful	ætbær
   that	she	Beowulf …	meadcup	brought
   ‘that she brought the cup to Beowulf.’ (Beowulf 623–4)

1. (2) She brought the cup.

In Minimalism, the emphasis is very different. This model focuses on how the lexical and grammatical items differ cross-linguistically in their features. For instance, in Modern English, declarative sentences have to have a Subject precede the Verb and this is accounted for by a feature (named the EPP feature on the T(ense) head). The OV/VO difference is also accounted for by means of features, as will be shown in section 5.5.

The outline of this chapter is as follows. In section 5.2, I examine the basic philosophy of generative grammar and the implications for historical syntax. In section 5.3, I consider early generative grammar and, in section 5.4, I discuss its early instantiations and look at the P&P approach, perhaps the most popular approach (even today) among practitioners of English historical generative syntax. In section 5.5, more recent models are discussed, features and cartography. Section 5.6 provides a short assessment on how generative grammar and English historical linguistics can work together fruitfully.

5.2 Generative grammar and historical linguistics

In this section, I first discuss the general generative attitude towards historical linguistics. I also note that mainstream English historical linguists do not always see generative grammar as helpful.

Generative grammar is interested in how a child acquires a grammar on the basis of the language the child is exposed to. If the language the child hears has changed or is changing from that which the parents/caregivers grew up with, the child will have a different input and may come up with a grammar (I-Language) different from that of the preceding generation. Generative grammar proposes principles that account for how a child constructs an internal grammar. Generative linguists are most interested in language change if this change happens in connection with the learning process. External change modifies the linguistic input, the E-Language available to the child, and the real interest is in how the child deals with this in terms of parameter resetting. If children hear more Verb–Object (VO) sentences than Object–Verb (OV) ones, a change possibly due to linguistic contact, they will assume that the word order is VO and set their parameter for Verb-headedness as head-initial rather than as head-final.

There are of course many ways in which language change can provide insights into the (generative) language faculty. For instance, changes that have not yet been attested may show restrictions imposed by Universal Grammar and, in turn, frequently attested changes that are reanalyses of the input by the child shed light on the language faculty. One example of a frequent change is that prepositions are reanalysed as complementizers, e.g. for and after in the history of English and not the other way round. This is a unidirectional change that gives us insight into the language faculty. Thus, (3a) is earlier than (3b) and (4a) earlier than (4b). The prepositional use of (3a) and (4a) is typical of Old English, whereas the first instances of complementizer use is in Early Middle English.

1. (3a)

   hlynode	for	hlawe
   made.noise	before	mound
   ‘It made noise before/around the gravehill’ (Beowulf 1120)

1. (3b) I would prefer for John to stay in the 250 class. (British National Corpus ED2 626)

1. (4a) Ercenberht rixode æfter his fæder

   ‘E. ruled after/following his father’ (Anglo Saxon Chronicle A, anno 640)

1. (4b) After she'd hung up, she went through into the kitchen. (British National Corpus GWO1402)

Note that the change from preposition to complementizer shows some intermediate stages, e.g. one where the entire prepositional phrase is preposed (see van Gelderen Reference Gelderen2011b).

Possibly because of the early emphasis on introspection and grammaticality judgements by a native speaker, work in historical generative syntax was not mentioned or referred to in Chomsky. The exception is the work on phonological change in Chomsky and Halle (Reference Chomsky and Halle1968) and that was most likely due to Halle's interests (as evidenced in Halle Reference Halle1962). Since Chomsky has set the agenda for generative linguists for at least fifty years, it has been ‘less popular’ to pursue historical linguistics, although there have been many exceptions.

Thus, there has always been a group of generative linguists interested in historical change, arguing that such change gives a special insight into the innate language faculty. Work by Closs (later Closs Traugott), King, Kiparsky, Klima, Lakoff, and Lightfoot testifies to that. Starting in 1990 in York, generative historical linguists have come together through the DiGS (Diachronic Generative Syntax) conference which is devoted entirely to diachronic generative syntax. Many of the DiGS conferences had selected papers appear, e.g. as Battye and Roberts (Reference Battye and Roberts1995), van Kemenade and Vincent (Reference Kemenade, van Kemenade and Vincent1997), Pintzuk et al. (Reference Pintzuk, Tsoulas, Warner, Pintzuk, Tsoulas and Warner2000b), and Crisma and Longobardi (Reference Crisma and Longobardi2009), and I return to these publications later.

As mentioned above, generative syntax has typically relied on introspective data, i.e. asking a native speaker for grammaticality judgements, because of the model's emphasis on the internalized grammar. These grammaticality judgements have involved complex constructions, requiring the hearer to judge whether the sentence in (5) is grammatical or not.

1. (5) ^*What did you see the man that ate?

The intended meaning of this ungrammatical sentence is to ask what the man ate; a grammatical alternative is (6).

1. (6) What did the man you saw eat?

Generative grammar assumes that words such as what move from the position they occupy as subject or object of a particular verb – in (5) and (6) what is the object of the verbs ate and eat respectively – to the position they occupy to indicate that the sentence is a question. In (7), the original position of what is indicated by an underscore; what is moved out of a relative clause.

1. (7)

   *What did you see the man [that ate __]?

   Relative Clause Island

The ungrammaticality of (5) shows that what cannot be moved from the relative clause. Ross (Reference Ross1967) identifies structures out of which movement is not possible and calls them ‘islands’. Since, in (5), what is taken from a relative clause, the construction is called a Relative Clause Island. Compare (5) to sentences such as (8) and (9): what is moved from a main clause in (8) and from a complement clause in (9). These movements result in grammatical sentences unlike the movement in (7).

1. (8) What did the man [you saw] eat __?

1. (9) What did you say [that the man ate __]?

Sentences such as these have never been taught to (native) speakers of English and yet they have clear grammaticality judgements. This ability to judge grammaticality then is due to their I-Language, constrained by principles of Universal Grammar.

Allen (Reference Allen1977), working with non-electronic texts, finds instances relevant to islands. In (10), the wh-word hwaet has been moved from the position of the subject of the predicate sie forcuðre to the beginning of the sentence.

1. (10)

   Ac	hwaet	saegst	ðu	ðonne	ðaet	__	sie	forcuðre
   But	what	say	you	then	that		be	wickeder
   ðonne	sio	ungesceadwisnes?
   than	be	foolishness
   ‘But what do you say is wickeder than foolishness?’
   (Boethius 36.8, from Allen Reference Allen1977: 122)

This movement results in an ungrammatical sentence in Modern English when it leaves that adjacent to the gap. (Some speakers of English do not find these sentences ungrammatical in Modern English and Fischer (Reference Fischer2007: 41) suggests the ungrammaticality may be a feature of written English.)

The absence of native speakers of Old English to check sentences like (10) and the reliance on written texts and (now) electronic corpora kept some generative grammarians from going into historical linguistics. Many generative linguists that went into historical linguistics have embraced work with the Penn Corpora of Historical English, including the Penn-Helsinki Parsed Corpus of Early Modern English, the Penn Parsed Corpus of Modern British English, the York–Toronto–Helsinki Parsed Corpus of Old English Prose, the Penn-Helsinki Parsed Corpus of Middle English. The Corpus of Historical American English is tagged for parts-of-speech and non-parsed texts are available from the Dictionary of Old English and the Middle English Compendium, to name but a few electronic resources.

A practical consequence of the uneasy relationship between generative grammar and historical linguistics is that generative conferences and journals do not see historical linguistics as a crucial component to their enterprise of understanding the faculty of language. Apart from the DiGS conference, mentioned above, other venues are generative or historical but not both. For instance, NELS (North East Linguistic Society) is a prominent generative conference and rarely includes historical work. If many generative linguists are sceptical about historical linguistics, the reverse is also true. Many (non-generative) historical linguists often see generative work as data-poor, over-theoretical, and not very insightful. This can be seen from the impact, or rather lack thereof, of generative linguists on mainstream historical linguistics conferences. For instance, conferences such as ICHL (International Conference on Historical Linguistics) and ICEHL (International Conference on English Historical Linguistics) will have generative papers and occasional plenary addresses using that framework, but generative grammar is a minor framework.

The field of English historical linguistics, as evidenced by the programmes of ICEHL and SHEL (Studies in the History of the English Language), is very theoretically diverse. The fields of pragmatics and discourse studies have really taken off and constraints from information structure are being used to account for syntactic variation, in particular for word order variation and pronoun choice. Many syntacticians of English historical linguistics have broadened their interest from syntactic issues to pragmatics. Ans van Kemenade (e.g. Reference Kemenade, Hinterhölzl and Petrova2009), Bettelou Los (e.g. Reference Kemenade, Milićev, Jonas, Whitman and Garrett2012), Susan Pintzuk and Ann Taylor (e.g. Pintzuk and Taylor Reference Pintzuk and Taylor2011), and Augustin Speyer (e.g. Reference Speyer2008) now focus on the relation between word order and information structure. See also Meurman-Solin et al. (Reference Meurman-Solin, López-Couso and Los2012) and the special issue of the Catalan Journal of Linguistics (2011) edited by Montserrat Batllori and Lluïsa Hernanz.

Having introduced the general emphasis of generative grammar, I now turn to some specific work in the next section.

5.3 Transformational generative grammar

In this section, I discuss some early generative approaches to language change for which King (Reference King1969) provides a good overview. This work involves both phonological change and syntactic change but I will focus on the latter, in particular Closs's work on English modals.

Closs (Reference Closs1965), Klima (Reference Klima1965), Kiparsky (Reference Kiparsky1965), and Chomsky and Halle (Reference Chomsky and Halle1968) emphasize learning as the cause of change. The latter authors state that ‘speakers are by and large unaware of the changes that their language is undergoing’ (1968: 250) and that, where adults can only add or delete minor rules, children can reorganize the system. This view goes back to Halle (Reference Halle1962: 64, 66–7). Closs (Reference Closs1965: 415) concludes that ‘language changes by means of the addition of single innovations to an adult's grammar, by transmission of these innovations to new generations, and by the reinterpretation of grammars such that mutations occur’.

Klima (Reference Klima1965: 83) formulates a model of generative language change emphasizing the discontinuous nature of change and reanalysis by the learner.Footnote ² I adapt it as Figure 5.1 which is also based on Andersen (Reference Andersen1973). It includes changes that adults can make (minor innovations) and that are then the input for the new generation. Figure 5.1 expresses that the exposure to a language triggers a grammar in an individual that is referred to as I-Language which, in turn, produces an E(xternal)-Language.

Figure 5.1 Model of language acquisition

(based on Andersen Reference Andersen1973)

Closs (Reference Closs1965), Kiparsky (Reference Kiparsky1965), Lakoff (Reference Lakoff1968), Closs Traugott (Reference Traugott1972), and Lightfoot (Reference Lightfoot, Anderson and Jones1974, Reference Lightfoot1979) use this model in various forms. Their explanations depend on the then current model of phonology and syntax. The model of phonology is fairly abstract with a large number of rules and the syntax has a phrase structure component and a set of ordered transformations. Most change was seen as change in the phonological and transformational rules, either by rule loss, rule reordering and addition, and restructuring/simplification.

Closs (Reference Closs1965) is, of course, the basis of much later work on auxiliaries, both in terms of the data as well as the analysis. She is concerned with the phrase structure rules and how the shape of AUX is different in Old English and suggests ways to account for the difference. The phrase structures for Old English are formulated as in (11); MV stands for main verb, Vt for a transitive verb, Vi for an intransitive one (with ‘move’ indicating a movement verb), PrP and PP for Present and Past Participle, and ‘env’ for ‘environment’.

1. (11)  

   

Closs thus argues that modals are a class separate from main verbs, as they are in Modern English, but that the main verb is in final position (as can be seen from the NP–V rule). She also accounts for the Old English have–be auxiliary split: transitive verbs have a perfect with have and intransitive ones have one with an auxiliary be. The main difference between Old and Modern English is that in Old English, in addition to the main verb and a modal, only one other auxiliary can appear. The AUX rule stipulates that.

She then looks at Middle English and argues that by the thirteenth century the word order is as in Modern English where the AUX precedes the main verb and the T and M precede the other auxiliaries. This is represented by (12) where the lexical verb also precedes the object NP. Other major changes that occur by the late sixteenth century: have is the generalized perfect for all verbs except those of movement, the spread of auxiliary or periphrastic do and the progressive passive. This gets us the following rules for Early Modern English.

1. (12)  

   

Closs chronicles the types of changes: reversal of the order, loss of restrictions, addition and loss of formants, and ‘finally…really radical changes of system membership, e.g. when do, which was a member of the lexical system, gave rise to an operator in the syntactic system’ (p. 412). Her final conclusion is that ‘language changes by means of the addition of single innovations to an adult's grammar, by transmission of these innovations to new generations, and by the reinterpretation of grammars such that mutations occur’ (p. 415). This conclusion fits perfectly with the model presented in Figure 5.1.

Lightfoot (Reference Lightfoot, Anderson and Jones1974), also focusing on modals, formulates the phrase structure rules for the modals a little differently from Closs (Reference Closs1965) but they are similar enough that I need not reproduce them. Unlike Traugott, Lightfoot argues that modal verbs are full verbs in Old English because they bear number agreement and act like main verbs in negative placement and inversion. After Old English, a number of seemingly isolated changes take place in pre-modals, namely the gradual loss of having an object and of inflection, that cause the child to assume a different I-Language from that of the earlier generation. Lightfoot calls this reanalysis a ‘radical change in the deep structure’ and a ‘radical restructuring’ (1974: 234). One of the advantages of generative analyses is to capture the relationships among phenomena that are not otherwise explainable. By assuming the modals and infinitival to and pleonastic do occupy a separate position around 1400, i.e. an I(nflection) or T(ense) position, one can explain the introduction of otherwise unrelated phenomena such as split infinitives, single modals, and the complementary distribution of modals, do, and to (see also van Gelderen Reference Gelderen1993).

The contemplation of the rapid changes in modals leads Lightfoot to adapting a principle from phonological change to syntax. Kiparsky (Reference Kiparsky and Dingwall1971) argues that, if the phonological representation of e.g. writer is /rayt-ər/ (the I-Language form) but the phonetic form is [rajder] (the E-Language form), there may be a reanalysis of the phonological form so that it resembles the phonetic form more closely. He calls this principle the Opacity Principle. If we need many rules to derive the phonetic form, the phonological form is opaque and the learner will make it more like the phonetic representation. Lightfoot adapts the Opacity Principle to syntax but formulates it in a positive way, namely that the grammar avoids opacity and favours transparency. He calls this principle the Transparency Principle and it explains that once, ‘the category membership of pre-modals became opaque…the grammar moved to avoid such opacity’ (1974: 244).

In this early period of the development of the generative framework, the change from verb to auxiliary, just described, is not identified as a special process. This process can be referred to as grammaticalization and had been recognized by von der Gabelentz (Reference Gabelentz1901) and Meillet (Reference Meillet1912). Grammaticalization is often defined as a change of a lexical item to a (more) grammatical one through a loss of semantic features. A typical example of grammaticalization was given in (3) and (4) where a spatial preposition for and a temporal preposition after are reanalysed as conjunctions introducing clauses. Even though Traugott's data present prototypical examples of grammaticalization (a verb being reanalysed as a modal auxiliary) and although she mentions the ‘jump’ of do from a lexical to grammatical class, the term grammaticalization does not appear till much later in her work (e.g. Traugott Reference Traugott, Lehmann and Malkiel1982).

Concluding, early generative approaches concentrate on formulating phrase structure rules such as those in (11) and (12). The next period focuses on principles and parameters, is more comparative, and becomes interested in how to account for grammaticalization.

5.4 Principles and Parameters

Parameters have been used in generative grammar since the so-called Principles and Parameters approach (Chomsky Reference Chomsky, Anderson and Kiparsky1973). As mentioned above, principles are valid for all languages but parameters need to be set and are therefore the locus of change. In this section, I first explain a little more what principles and parameters are in the P&P framework. I then look at work in the history of English that focuses on the headedness parameter and on Verb-movement.

Early examples of parameters include determining if a language has pro-drop (Rizzi Reference Rizzi1982), its headedness (e.g. Verb-last or Verb-first), and whether wh-elements move or not. Pro-drop is the cover term for a set of related phenomena centring on the absence of the subject in a finite clause. Thus, Italian is pro-drop but English is not. Whether the head of a phrase is initial or final (a VO language is head-initial and an OV language head-final), is an important way to characterize a language, with Modern English being head-initial and Old English (mostly) head-final, as we have seen in (1) and (2). Wh-movement characterizes Germanic languages, the various stages of English included, but this is not universally true, with Chinese being an example of non-movement.

Important work in the period from the mid 1970s to the early 1990s is done on word order and headedness. Canale (Reference Canale1978) argues that Old English changes its headedness around 1200 when it changes from an OV to VO language. Van Kemenade (Reference Kemenade1987) examines the word order in Old English and devises a phrase structure with a V-final VP and an I(nflection) position outside of the S(entence). She identifies two changes in the history of English, a change from OV to VO around 1200, echoing Canale, and a change in the position of the I(nflection). She argues that Verb-movement (to second position) becomes more limited in scope and is different from Verb-second in modern Verb-second languages, such as German or Dutch. Exploring the position of auxiliaries in relation to verbs leads to an examination of the inventory and order of the functional categories. Functional categories express aspect, mood, tense, pragmatic and semantic role, and are also known as closed categories. Work on determiners (Philippi Reference Philippi, van Kemenade and Vincent1997), complementizers (van Gelderen Reference Gelderen1993), the position of the I(nflection) (Pintzuk Reference Pintzuk1991), and auxiliaries (Warner Reference Warner1993) is the result.

A much more sophisticated model of word order starts to take shape. For instance, van Kemenade recognizes that wh-elements and negatives in initial, pre-V position trigger absolute V-second but that, with topicalized elements, the subject pronoun can precede the verb resulting in V-third, as (13) shows.

1. (13)

   Ðas	þing	we	habbað	be	him	gewritene…
   these	things	we	have	about	him	written
   ‘These things we have written about him.’ (Chronicle E, 1087, 143,
   van Kemenade Reference Kemenade1987: 110)

A slightly updated tree for (13) is given in (14); I have kept the pronoun as a clitic to C, as in van Kemenade. This tree also shows the general clausal structure assumed from the mid 1980s on, with a C(omplementizer) Phrase and an I(nflection) Phrase (the latter replaced by T(ense) Phrase later on).

1. (14)  

   

This work provides the seed to the current shift by van Kemenade and others, e.g. Los (Reference Los, Meurman-Solin, López-Couso and Los2012), to connecting word order and information structure. Topicalized elements, such as Ðas þing in (13), bring about very different structures from wh-elements, which are focus elements. In terms of information structure, a topic is often old information and a focus may present new information.

Constructions such as (13) also lead Kroch and Taylor (Reference Kroch, Taylor, van Kemenade and Vincent1997) to investigate the regional variation in the Verb-third construction and find that the south deviates more from the strict Verb-second pattern. Speyer (Reference Speyer2008) revisits the issue and argues that the choice between Verb-second and Verb-third is determined by a requirement to avoid two focused elements next to each other.

In the early 1990s, many linguists start looking at the category of I and C cross-linguistically and at Verb-movement (also known as inversion). Pintzuk (Reference Pintzuk1991) looks at the position of I in Old English, Roberts (Reference Roberts1993) at Verb-movement, auxiliaries, and I in French and English, Kiparsky (1995) at V-to-C movement in early Indo-European and Germanic, Eythórrson (Reference Eythórsson1995) at early Germanic Verb-movement, and van Gelderen (Reference Gelderen1993) and Los (Reference Los2005) look at the status of infinitives and the position of to in English.

Battye and Roberts (Reference Battye and Roberts1995), in the first volume to come out of the earlier-mentioned DiGS conferences, review the diachronic work done within the P&P framework. The title of that volume includes ‘Clause Structure’ and that is very apt since much of the work around this time is on formulating the clausal skeleton of CP, IP, and VP, i.e. parametric differences in V-to-I and V-to-C, in pronouns and clitics. In van Kemenade and Vincent (Reference Kemenade, van Kemenade and Vincent1997), entitled Parameters of Morphosyntactic Change, we can see the beginning of the shift towards Minimalism with its emphasis on morphosyntactic features. Most of the papers continue to be more in the P&P framework, as do those in Pintzuk et al. (Reference Pintzuk, Tsoulas, Warner, Pintzuk, Tsoulas and Warner2000b). Typical topics are null arguments, negatives and polarity, position of subjects, the change to VO order, and Verb-movement.

This is a period that becomes influenced by the cartographic model which aims to provide a unique position for each functional category. Following these cartographic ideas, Roberts (Reference Roberts, Brandner and Ferraresi1996) argues that clitics, such as the subject we in (13), are in the Spec of a Fin(ite) Phrase and that the V is in a Fin(ite) head position. I have shown this in (15), which uses a CP split into a Top(ic) Phrase and a Fin(ite) Phrase.

1. (15)  

   

In wh-questions, the wh-element is in the Spec of a Focus Phrase and the V is in the head of this Focus Phrase. Note that there is no evidence for such an expanded main clause CP in northern dialects, as shown by the examples in Kroch and Taylor (Reference Kroch, Taylor, van Kemenade and Vincent1997: 321). In the northern glosses, the verb is always in second position even with topics.

In Old English, most subordinate clauses are Verb-final. Hence, these show no evidence of V-to-I movement although, once in a while, there is one that looks as if it has a topic with a verb moved to I, as in (16), from Pintzuk (Reference Pintzuk1991).

1. (16)

   þæt	his	aldres	wæs	ende	gegongen
   that	his	life	was	end	come
   ‘The end of his life had come’ (Beowulf 822, from Pintzuk Reference Pintzuk1991: 187)

Pintzuk (Reference Pintzuk1991), of course, says that verbs always move to I but that the position of I is variable.

Thus, in the 1990s, historical generative syntax continues to be concerned with clause structure and the parameter of Verb-movement. It starts to pay attention to information structure and to use cartographic structures in the late 1990s.

5.5 Current generative grammar: cartography and featuresFootnote ³

At the moment, the emphasis in the Minimalist Program is on principles not specific to the language faculty, but to ‘general properties of organic systems’ (Chomsky Reference Chomsky and Belletti2004: 105), labelled third factor principles in Chomsky (Reference Chomsky2005, Reference Chomsky, Sauerland and Gärtner2007). For instance, Chomsky (e.g. Reference Chomsky and Belletti2004, Reference Chomsky, Sauerland and Gärtner2007) and Richards (Reference Richards, Heck, Müller and Trommer2008) attribute as little as possible to the role of parameters and to Universal Grammar in general. Apart from this move away from language-specific principles, a second change is that Minimalist parameters consist of choices of feature specifications as the child acquires a lexicon. All parameters are lexical and determine the linear order of the words; therefore, they account for the variety of languages. As Pintzuk et al. (Reference Pintzuk, Tsoulas, Warner, Pintzuk, Tsoulas and Warner2000b: 7) put it, ‘the lexicon…must be the locus of syntactic change’.

Both developments, the move towards general principles and towards parametric features, make it easier to account for grammaticalization, a very frequent change that was left unmentioned in early generative work and one that I will discuss in this section. First, I will say a little more on the cartographic approach whose goal it is to map the order of the functional elements precisely and hence the name (with the main works being Rizzi Reference Rizzi and Haegeman1997 and Cinque Reference Cinque1999). This approach was introduced in the previous section because it emerges from the P&P model. It provides an answer to the problem of how grammatical categories, such as the various auxiliaries and adverbs, are ordered. Cinque comes up with a Functional Hierarchy, as in (17), for a subset of cases.

1. (17)

   Tpast	Tfut	Moodir	Modnec	Modpos	ASPhab	ASPrep
   once	then	perhaps	necessarily	possibly	usually	again
   ASPfreq
   often
   (from Cinque Reference Cinque1999: 107)

Roberts’ tree in (15) represents a cartographic representation of the highest layer of the clause, namely how clitics and topics are ordered in the expanded CP domain. Other diachronic work using this approach is, for instance, that by Wood (Reference Wood2003) on the DP and van Kemenade (Reference Kemenade, van Kemenade and Vincent1997) on the CP layer. More could be done in the domain of modal verbs and adverbs as they grammaticalize. Under the Roberts and Roussou (Reference Roberts and Roussou2003) and van Gelderen (Reference Gelderen2004) approaches to grammaticalization, modals and adverbs would move from right to left on the Hierarchy in (17).

The more Minimalist change in direction has been to focus on features. Here, I discuss two such initiatives: one to account for grammaticalization in terms of features and a second to account for word order that way.

Van Gelderen (Reference Gelderen2011b) argues that grammaticalization can be understood as a change from semantic to formal features. An Old English verb such as wolde ‘wanted’ in (18a) has [volition, expectation, intention] features. These features are simplified when will is reanalysed as having the grammatical feature [future] in (18b).

1. (18a) Ta þreo kingess…forenn till Herode king To witenn whatt he wollde.

   ‘The three kings went to King Herod to know what he wanted.’ (Ormulum 6571, from the Middle English Dictionary)

1. (18b) I'll never forget the judge saying that Lindy would be put into jail for life…(COCA 2012 news)

Subsequently, verbs like go and want go through this same reanalysis as future markers in (19) and (20).

1. (19) Let's get inside. It's going to rain. (COCA 2011 fiction)

1. (20) We have an overcast day today that looks like it wants to rain. (Nesselhauf Reference Nesselhauf2012: 115)

Biberauer and Roberts (Reference Biberauer, Roberts and Eythórsson2008) in examining the shift in word order from OV to VO rely crucially on a feature that has been called the EPP-feature. This feature (in languages that have it) is responsible for making a subject move to sentence initial position. If the T(ense) position bears this feature, a D head will move to T or a DP will move to the specifier of the TP, as is the case in Modern English. Languages can also have a VP or vP satisfy the EPP feature. This is the option chosen in Old English: the VP is moved rather than stranded as it is in Modern English. Thus, Biberauer and Roberts derive the Old English word orders by raising V to the light verb v, VP raising to the specifier of vP, and vP raising to the specifier of TP, as in (21) for sentence (22).

1. (21)  

   

1. (22)

   gif	heo	þæt	bysmor	forberan	wolde
   if	she	that	disgrace	tolerate	would
   ‘if she would tolerate that disgrace.’
   (ALS [Eugenia]: 185.305, from Pintzuk and Taylor Reference Pintzuk and Taylor2011)

To be more precise, in (22), the V forberan ‘tolerate’ first moves to little v and the VP with just the object in it moves to the specifier of the vP, resulting in the OV order. Then, the subject is combined (also known as ‘merged’) to a second specifier position in the vP and the entire vP is moved to the Specifier of the TP.

Various other word orders can be derived as well depending on whether the entire vP is moved to the specifier of the TP or just the subject. In the latter case, we derive (23).

1. (23)

   Martianus	hæfde	his	sunu	ær	befæst
   Martin	had	his	son	earlier	established
   ‘Martin had earlier established his son’
   (ALS_[Julian_and_Basilissa]: 184.1049, from Pintzuk and Taylor Reference Pintzuk and Taylor2011)

In short, current generative historical linguists are using both the cartographic model and feature checking mechanisms to account for grammaticalization and word order variation. In the next section, I provide a more general evaluation of what generative grammar and English historical linguistics can contribute to each other.

5.6 Generative grammar and English historical linguistics

English historical linguistics is most beneficial for generative grammar, in my mind, when it can provide evidence for recurring patterns of change, such as the frequent grammaticalizations. Because written sources in English go back at least 1,500 years, we can see these changes recur, as in the case of the emergence of future markers. These then provide a window on the acquisition process, so crucial to generative grammar. If different varieties all change in the same direction (all renew in a similar way), there must be a universal force driving that.

English historical linguistics, in its turn, benefits by the novel questions generative grammar asks, such as what is the role of certain features in Old English. We have had recent innovative work on changes in word order and information structure that would not have been done if generative grammar had not provided the cartography of the clause (and the nominal). This has given rise to a renewed examination of data from older stages. For Old English, it has meant a closer look at the kinds of topicalizations and the structure of Verb-third constructions. There is thus a lot of opportunity for English historical linguistics and generative grammar to have a mutually beneficial influence.

5.7 Conclusion

In this chapter, I have sketched the basic philosophy behind generative grammar and how that view impacts the relationship of generative grammar with English historical linguistics. In the last sixty years, there has been a shift from a system with highly specific phrase structure rules to a system of feature checking.

Generative grammar is concerned with the linguistic knowledge in the mind of a speaker and it focuses on how children acquire their internal grammars based on the available evidence. In this model, the child is the main reason for (internal) linguistic change. If language change can be shown to be regular, this gives us a window on the language faculty. Lexical words grammaticalize to grammatical ones and this can be accounted for if it is the child that extrapolates grammatical features from lexical ones during language acquisition. Language acquisition in this model is discontinuous and linguistic change need not be gradual: a parameter can be set differently by the next generation, with major changes as a result. Currently, the emphasis on features makes it possible to also consider gradual change, such as that characterized by changes in future marking and the change from preposition to complementizer.

6.1 Introduction: the nature and purpose of philological study

What is meant by the word ‘philology’ is by no means universally agreed upon, though a commonly held conception of the matter is that it is an aggregate of the various modes of inquiry required for the editing of texts in extinct languages (see, e.g., Gumbrecht Reference Gumbrecht2003: 1–4), an endeavour that has for most of its history had as its chief aim recovering authorial versions of texts altered in the course of transmission (though alternative views are mentioned below). Philology may thus involve historical and comparative linguistics and the study of manuscripts (including palaeography, codicology, the study of how manuscripts are related to one another, and scribal practices), orthographic systems (including orthoepy), poetic metre, rhyme, translators’ practices, and numismatics, among other concerns. In connection with English historical linguistics, many of these same concerns are brought to bear by the philologist, and thus, the philological component of historical linguistic study may be regarded as the extralinguistic contexts of linguistic data, or the relation between contexts and data. Linguistic features of historical texts may accordingly be influenced by such extralinguistic contextual factors as the genre and communicative purpose of the text; conventions employed such as orthographic standards, which may be in competition with authorial habits and the residue of intermediate textual states in the process of a text's transmission, involving such concerns as dialect translation and updating of archaic linguistic forms by agents like editors, compilators, and scribes; the text's intended audience, which may demand attention to such concerns as social distance, gender, authority, and age; errors introduced in the process of reproducing a text, for example by scribes and typesetters; and the material features of the medium in which the text is transmitted, such as the space available in a book or document, which may result in purposeful omissions, increased density of abbreviation, and so forth.

In historical language study, philology is thus deployed not in the service of textual editing but of mediating between the demands of linguistic methodology and the limitations that beset the records of prior states of the language available for linguistic analysis, since those records do not directly and unproblematically represent earlier linguistic stages of English. For example, nearly all the surviving prose from the later Anglo-Saxon period is preserved in the West Saxon dialect, roughly half of which is commonly called ʻÆthelwoldianʼ or ʻÆlfricianʼ and is characterized by relative uniformity of orthography and morphology, along with some distinctive and characteristic lexical choices. It is apparent that this linguistic variety is fairly artificial and bookish, having been promulgated by Æthelwold, Bishop of Winchester (963–84), and his prolific student Ælfric as a way of standardizing the language. Of the remaining prose, the considerable majority, to various degrees, gives the impression of heterogeneity in orthography, morphology, and lexis. Although the reason for this is not universally agreed upon, it seems likely that it is a result of the history of individual texts, many of which appear to have been composed in other dialects and only later ʻtranslatedʼ into West Saxon, with varying degrees of thoroughness (see, e.g., Fulk Reference Fulk, Fitzmaurice and Minkova2008, Reference Fulk, Jurasinski, Oliver and Rabin2010, Reference Fulk, Denison, Bermúdez-Otero, McCully and Moore2012, with references). Such ʻtranslationʼ from one dialect to another, resulting in a mixture of dialect features in a single text, is undeniable in connection with English texts from the later Middle Ages: see LALME 1: 12–23, with extensive discussion of this and other sources of textual mixture. Epigraphic traditions not associated with manuscript production may provide a more realistic impression of spoken medieval varieties of English (see Colman Reference Colman1984). Although no medieval text is unambiguous in nature as a witness to an earlier state of the language, the varieties of recorded Late West Saxon just described illustrate well the sorts of historical and cultural concerns that differentiate medieval sources from the kinds of data available for present-day languages, and which it is the philologist's concern to make explicit in connection with diachronic studies of English.

6.2 The rise of philology and the fields it originally encompassed

The current role of philology in historical linguistics was moulded by the historical forces that shaped linguistics as a discipline. The modern study of earlier stages of English may be said to have begun in earnest with Matthew Parker, Archbishop of Canterbury (1559–75), who set out to instruct himself and others of his circle in the Old English language, for the purpose of demonstrating that certain practices of the newly formed Church of England regarded as heretical by the Roman Catholic Church, such as clerical marriage and the translation of Scripture into the vernacular, were also practices of the Anglo-Saxon Church, and that the Anglican Church thus represented a return to a more original state of Christianity. The translation and dissemination of Old English texts therefore served an immediate political purpose. Parker's method of learning Old English was of a philological nature, relying upon Old English translations of Latin, glossed texts, and the Latin grammar (in Old English) of Ælfric, working out the grammar and lexicon of the unknown language by reference to the known one.

Although work on Old English continued apace after Parker's day, it was not until the appearance of the Thesaurus of George Hickes (1703–5) that the study of earlier stages of the language began to assume the appearance of a philological endeavour in the current understanding of the phrase. Hickes perceived the importance of numismatics to the study of the language, and he included a section on Anglo-Saxon coins in the work. He also made an attempt to distinguish varieties of Old and Middle English, differentiating Anglo-Saxon from ʻDano-Saxonʼ, ʻNortmanno-Saxonʼ, and ʻSemi-Saxonʼ (see Cain 2010). Most important, however, was his recognition of the value of comparative language study, since he drew direct comparisons between Old English and Gothic and Icelandic, and he included grammars of Old High German and Old Icelandic in the Thesaurus. It would be many years before the English language was again presented in such a thoroughly comparative framework. Hickes's Thesaurus remains of importance today chiefly because its second volume, compiled by Humphrey Wanley, is a catalogue of Anglo-Saxon manuscripts in British archives (at a time when studying Old English texts usually meant working from manuscript, so few editions had appeared), many of which manuscripts have subsequently undergone loss or damage, so that Wanley's descriptions and his incipits and explicits remain vital sources of information. Wanley's careful work laid the foundation for what remains a vital philological enterprise: the cataloguing and detailed description of manuscript sources (for bibliography, see Wright Reference Wright2016a and Reference Wright2016b).

Yet the most important development in the rise of philology as a discipline was the discovery of the Indo-European family of languages and the realization that they must descend from a common tongue. The beginning of Indo-European studies is usually dated to a 1786 lecture by Sir William Jones in which he pointed out similarities among Greek, Latin, and Sanskrit, with related observations about Gothic, Celtic, and Persian, though it is plain now that Jones was not the first to recognize similarities of this sort, some observations having been made as early as the sixteenth century (see Murray Reference Murray2015). The first full flowering of Indo-European language studies was Franz Bopp's comparative grammar of the Indo-European languages (1833–52), a systematic study on scientific principles (despite some missteps, such as Bopp's attempt to bring Malayo-Polynesian and the Caucasian languages into the Indo-European family). What made possible the study of language history on a scientific basis was the discovery of sound laws – rules for phonetic change comparable to the laws of mechanics or chemistry – the earliest significant example of which was the description by Jacob Grimm (Reference Grimm1822, in some respects anticipated by Rask Reference Rask1818) of regular correspondences between the sounds of the Germanic languages and of other Indo-European languages (e.g., f in English father and fee corresponding to p in Latin pater and pecu), pointing to a systematic set of sound changes in prehistory now referred to as Grimm's law or the first sound shift (see Prokosch Reference Prokosch1939: 47–60). Diachronic linguistics was thus established as a science and became itself a model for other scientific endeavours, most notably in the way Charles Darwin, in theorizing about natural selection, conceived of the development of species as analogous to the development of languages (see Alter Reference Alter1999).

As long as the focus of linguistic inquiry remained on diachrony, the term ʻphilologyʼ could be used to designate the field of historical linguistics, as it commonly did in the nineteenth century. In these early days of philology there was no very notable disjunction between those aspects of it devoted to language study and those to textual criticism. The great linguists of the nineteenth century were mostly editors, as well: Grimm, for example, edited a variety of medieval texts (not to mention the work for which he is now best known, the fairy tales that he compiled with his brother Wilhelm); Rask edited the two eddas; and Eduard Sievers, the most brilliant figure in the school of Junggrammatiker dominating the field in the latter half of the century, edited the Heliand and the Old English Genesis B and formulated the metrical principles on which the editing of early Germanic poetry depends vitally to this day. Similarly, the philological journals of the day were plainly devoted to both linguistics and literature, the two types of studies being mixed indifferently in almost any published issue of periodicals like Archiv für das Studium der neueren Sprachen und Literaturen(1846–), Transactions of the Philological Society (1854–), Zeitschrift für deutsche Philologie (1869–), Beiträge zur Geschichte der deutschen Sprache und Literatur (1874–), Zeitschrift für deutsches Altertum und deutsche Literatur (1876–), Anglia: Zeitschrift für englische Philologie (1877–), Englische Studien (1877–1944), Nordisk Tidsskrift for filologi (1877–), Arkiv för nordisk filologi (1883–), Neuphilologische Mitteilungen (1899–), Journal of English and Germanic Philology (1903–), Modern Philology (1903–), Studies in Philology (1906–), Maal og minne (1909–), Neophilologus (1915–), English Studies (1919–), Philological Quarterly (1922–), Acta Philologica Scandinavica (1926–88), Studia Neophilologica (1928–), and other philologically inclined journals. It will be observed from this list, which includes just one British journal (Transactions of the Philological Society), that it was in the German and Scandinavian academy that the principles of philological study were first elaborated, the Dutch and American journals being later additions. The British attitude towards the discipline in fact was generally hostile for much of the time that philology dominated language study elsewhere in the Germanic-speaking world (see, e.g., Momma Reference Momma2013).

6.3 The varieties of work subsumed within philology

A perusal of these same journals also reveals the range of epistemologies involved in the practice of nineteenth- and early twentieth-century philology, though that range can perhaps be illustrated more effectively by a selection of some of the better-known philological achievements and instances of philological discovery in connection with the history of English. Among the most familiar of these is the remarkable hypothesis of Sievers (Reference Sievers1875) that lines 235–851 of the Old English poem Genesis (the portion now referred to as Genesis B, and which had already been recognized as different in style from the rest of the poem) is translated from a Continental Saxon original (for an account, see Doane Reference Doane1991: 3–8). The hypothesis was dramatically confirmed in 1894, when a fragment of an Old Saxon versified Genesis was discovered in a manuscript in the Vatican Library, and its first twenty-five lines were found to correspond closely to lines 790–816 of the Old English poem. Sievers's hypothesis was based upon observations of similarity between Genesis B and the Old Saxon Heliand in the use of poetic formulae, vocabulary, and shared phrases, and it demonstrates not only the power of philological reasoning but also its utility in respect to linguistic investigation, since the strong influence of Old Saxon upon the language of Genesis B indicates that the poem is not to be used as linguistic evidence for Old English poetic language.

Sievers (Reference Sievers1885, Reference Sievers1893) was also the chief architect of our understanding of Old English poetic metre, though of course his analysis is indebted to earlier work. He arrived at his conclusions on the basis of comparison to Old Norse verse, of painstaking observation of patterns of stress and alliteration, and of abstract reasoning aided by a profound knowledge of etymology. Even at their first formulation his principles were rigid enough to reveal ways in which the recorded forms of Old English poems must differ from the forms they took when initially composed. His findings have thus proved an indispensable aid to the editing of poetic texts, and they have shed important light on the problem of determining the date and dialect of origin of Old English poems, which are almost uniformly preserved in late copies much altered in the course of scribal transmission. The analysis of Sievers in turn enabled a discovery by H. Kuhn (Reference Kuhn1933) about the relation between syntax and stress that is perceptible only on the basis of poetic metre: words of variable stress, such as pronouns, monomorphemic adverbs, and finite verbs, are stressed unless they appear in the group of syllables that comprise the first unstressed position in the clause. The phenomenon is relatable to Indo-European clausal patterns and lends support to common views about the dating of Old English poems.

The identification and localization of the dialect of the early or mid ninth-century Old English gloss on the Vespasian Psalter is a remarkable accomplishment. Since the psalter itself is Kentish in origin, it was at first thought that the gloss must be, too. Sweet (Reference Sweet1882) was the first to recognize that the dialect must be an archaic variety of the Midland dialect known as Mercian, a conclusion he reached on both codicological grounds – an existing Mercian charter seems to have been written on a blank leaf of the book that was later separated from the manuscript – and linguistic, by comparison to the language of known Mercian charters of an early date. Restriction of the provenance to the westernmost portion of Mercia was established by Tolkien (Reference Tolkien1929), due to his discovery of remarkable similarities between the language of the psalter gloss and of a group of Middle English texts (the ʻKatherine Groupʼ) associable with the medieval diocese of Hereford, and Millett (Reference Millett1992: 219) would narrow the relevant area to southern Shropshire. Such a localization accords well with certain features derivable from modern dialect surveys.

The localization of Middle English dialects has made tremendous strides in the past half century due chiefly to the research undertaken in conjunction with the production of the Linguistic Atlas of Late Mediaeval English (LALME; see Chapter 7 by Horobin in this volume). One reason for this progress is the development of the so-called fit-technique, whereby unlocalizable texts are assigned a geographical position by triangulation on the basis of features shared with texts of known provenance (see Benskin Reference Benskin and Riddy1991). The result of such careful attention to diatopic issues is some remarkable findings about the localization of Middle English texts, such as the ability to discern in the B-text of Piers Plowman a set of linguistic features discoverable on the basis of poetic alliteration and associated with a closely circumscribed area of western Worcestershire, including the Malvern Hills, where the poet is known to have spent his youth (see Samuels Reference Samuels and Smith1988). What is remarkable about this is that Piers Plowman is to be found in so many manuscripts copied at so many different locations in which local features were added to the text, though no surviving manuscript was made in western Worcestershire.

Determining the pronunciation of Elizabethan English is obviously a prerequisite to its phonological analysis. Since that pronunciation is recoverable (to the extent that it is recoverable at all) only by philological means, this is an area of study in which the value of philological investigation to linguistic analysis is particularly plain. Some of the most detailed studies have focused on the pronunciation of a single individual, most prominently William Shakespeare (see Cercignani Reference Cercignani1981, with a critique of prior research going back to 1861 on pp. 2–21). There is a certain irony in this practice of studying a single speaker, since the pronunciation of no single Elizabethan is fully recoverable, especially an individual like Shakespeare, whose native dialect was not that of London, though he worked in the city, and who may have employed pronunciations other than his own for dramatic or comic effect. Still, the study of one individual's speech, when tempered with what can be determined about the pronunciation of others, reduces the effect of linguistic heterogeneity that inevitably arises in the study of such a cosmopolitan setting as London c.1600. The evidence for Shakespeare's pronunciation is both internal and external. The most important internal evidence is to be derived from rhyme and metre in his poetry, though puns and spellings, it has sometimes been argued, are relevant. External evidence is of various sorts. It includes what is known about antecedent states of the language (as divined on the basis, once again, of internal and external evidence), the evidence of the language in modern times, and, what is most important, the witness of early grammarians and rhetoricians as to the pronunciation of the language in the sixteenth and seventeenth centuries (treated in detail by Dobson Reference Dobson1968). These last include, for example, works of orthoepy, which are tracts intended to prescribe correct usage, devise phonetic descriptions, or promote spelling reform, publications which began to appear in the sixteenth century, when the expansion of the merchant class and the commercial success of the printing press had created a market for such works and made their dissemination economically feasible. Likewise in early modern studies, philological concerns have aided immensely in studies of authorship. For example, in an influential essay, Forker (Reference Forker and Howard-Hill1989), on the basis of an examination of style, idiom, vocabulary, and spelling, is able to confirm the widespread belief that three pages added to the sole manuscript of the drama Sir Thomas More, by Anthony Munday and others, must be in the hand of Shakespeare.

What these examples illustrate is the great variety of concerns addressed by philological inquiry. Such a variety of concerns naturally draws on a wide field of resources, so that the pursuit of philology very commonly draws on rather diverse aids to analysis, such as historical concordances and dictionaries, catalogues, and indexes of primary sources (manuscripts, early printed books, specific collections, or genres, etc.), historical dialect atlases, handbooks of poetic form, historical grammars and manuals of etymology, manuscript facsimiles, and bibliographic reference works, among many others, including an ever-increasing array of electronic resources. In the face of such diverse aims and aids, it is not possible to provide comprehensive instruction in philological methods in any very concise way. The best way to learn such methods is to examine individual cases like these and study them for what they demonstrate about how philology is practised.

6.4 The relation of philology to English historical linguistics and literary studies

As these examples illustrate, many kinds of evidence, both linguistic and extralinguistic, are involved in philological discoveries, and those discoveries tend to be of service both to historical linguistics and to textual criticism. It was thus inevitable that the term ʻphilologyʼ should at first have encompassed many areas of study besides diachronic language study. But the turn towards synchrony as the primary focus of linguistic research in the twentieth century eventually made it plain that philology and historical linguistics were not synonymous, since the latter incorporated methods developed in synchronic linguistics that were of no great relevance to the textual principles that remained central to the philological enterprise. The particularizing and contextualizing aims of philology in fact came to seem, to some, an impediment to the practice of historical linguistics on the basis of the same generalizing and abstractly inclined aims that prevail in the study of living languages (see, e.g., Hogg Reference Hogg1992a: vii–viii).

The conflict may be illustrated by the example of a study by Toon (Reference Toon1983: 90–118) in which orthographic variation between <a> and <o> before nasal consonants is taken to evidence the rise and decline of a sound change in various Old English dialects, though in texts of the later Middle Ages and in modern dialect surveys rounding of this vowel is attested securely only in the West Midlands. To analyse the data as attesting to the progress of such a change is to treat written records as if they were direct representations of speech, when in fact it is likely that a number of contextual factors have conditioned the spelling in these early documents, including the cultural influence exerted by the politically dominant West Midland kingdom of Mercia, which plainly affected the language of these documents in other ways, and uncertainty about how best to represent sounds for which there was no distinctive character in the alphabet that the Anglo-Saxons inherited from Roman and Irish missionaries. This latter consideration is particularly relevant in the present instance, because if, as is generally assumed, the vowel in question was /ɔ/, which occurred only before nasal consonants, it is possible that in the earlier stages of the Old English orthographic tradition there had not yet developed the convention of using <o> to represent this sound that was neither /ɑ/ nor /o/. That this is the correct interpretation is suggested by the word ōnettae ‘anticipated (?)’, which appears in the earliest relevant document, a document that does not show any instances of <o> from a before nasal consonant and thus is assumed by Toon to antedate the change. This word ōnettae appears to show compensatory lengthening of initial /ɔ/ in ^*anhǣtidæ (Luick Reference Luick1914–40: §110 Anm. 2), indicating that rounding of the vowel before nasal consonants had already taken place. These are perhaps among the reasons why Kitson (Reference Kitson and Fisiak1995: 102) calls Toon's ʻa mad bookʼ; see Lowe Reference Lowe2001 for specific criticisms of methodology.

What the example illustrates, then, is how considerations that are not strictly linguistic in nature but which fall under the rubric of philology are indispensable to the linguistic analysis of historical records. To approach such records the same way a linguist would approach an objective transcript of a sample of a living language is to incur the danger of engaging in historical and linguistic fallacies. The problem is a serious one because it raises certain obstacles to the application of many modern linguistic methods to the analysis of earlier stages of English, with the result that to some it has seemed that English historical linguistics is mired in the methods of the nineteenth century and makes wholly inadequate use of the sorts of insights to be derived from newer methodologies. The failure to apply the insights of sociolinguistics in connection with social stratification has been particularly noticeable (see, e.g., Toon Reference Toon, Koopman, van der Leek, Fischer and Eaton1987, Hogg Reference Hogg, Kastovsky and Bauer1988: 188) up until the advent of sociohistorical and, later on, historical sociolinguistic methodology (see, e.g., Romaine Reference Romaine1982, Tieken-Boon van Ostade Reference Tieken-Boon van Ostade1987, and Nevalainen and Raumolin-Brunberg Reference Nevalainen and Raumolin-Brunberg1996a, Reference Nevalainen and Raumolin-Brunberg2003; see also Chapter 1 by Romaine and Chapter 17 by Nevala in this volume).

A result of the philological obstacles raised to the analysis of earlier stages of the language using current linguistic methods is that in the course of the twentieth century philology grew increasingly marginalized as a methodology. Whereas it had traditionally allied the fields of linguistics and textual and literary criticism, it came to play a vastly reduced role in the former, and in regard to textual criticism some of its prevailing tenets, such as the value of stemmatics (determination of the genetic relations of manuscripts or other media) and the rationale of textual emendation, came under increasing attack; for an account of the debates in textual criticism, see Greetham (Reference Greetham1999). But at the same time that philology grew ever more estranged from the field of linguistics, its role as the chief method of literary study came to an end as literary criticism came to be dominated by modes of analysis derived from British aestheticism, at first particularly (in Britain) the so-called practical criticism and (in America) the New Criticism, which regarded textual criticism as ancillary to aesthetic concerns (see, e.g., Graff Reference Graff1987). Philology thus remained most firmly tied to textual editing, though it never relinquished its stake in both linguistics and literary analysis, which remained indispensable to its operations.

6.5 New directions for philology

Under these circumstances, countless excellent editions of early texts continued to be produced on the basis of traditional philological methods, but a small though extremely vocal minority of philologists attempted to develop a new basis for philology that could be reconciled with the synchronic analysis of early texts rather than the prevailing model anchored in textual reconstruction. The most self-conscious of these attempts was a collection of articles, a self-characterized ʻcall to armsʼ, proclaiming the establishment of the so-called New Philology (Nichols Reference Nichols1990), a set of practices centred on the conviction that the study of textual variance as an expression of cultural conditions coeval with the manuscripts in which they occur is a more fitting object of philological scholarship than textual reconstruction. The New Philology was thus to be intensely manuscript-centred, since the manuscripts themselves, in all their textual and extra-textual aspects, are the key to unlocking pre-modern language and culture. The linguistic area in which such an approach seemed likeliest to be fruitful was identified as pragmatics, though grammar was also identified as a major topic for exploration on the assumption that the grammar of (quasi-)oral discourse, as represented by medieval texts, is not as rule-bound as the grammar of literate discourse (see Fleischman Reference Fleischman1990).

The New Philology never achieved the status of an established methodology, for a variety of understandable reasons, but its promise of a way to wed philology to modern linguistic methods was not to be ignored, and many of its premises endure by their incorporation into current manuscript studies and into varieties of historical dialectology and pragmatics, as detailed below. But the reinvigoration of philology that was one of the stated aims of the New Philology has been furthered more effectively by the development of digital resources. This may at first seem an ironic development, since the view of many at the dawn of the age of digitized resources was that the advent of digital editions presaged the end at last of philology by placing the traditional functions of the editor in the hands of the ultimate user of the edition. For example, Szarmach (Reference Szarmach, Keefer and O'Brien O'Keeffe1998: 107) remarked: ʻUltimately, the computer is likely to subvert the authority of the editor, who will become an inputter of text, a programmer if lucky, but never again the sole arbiter of meaning.ʼ Conner (Reference Conner, Keefer and O'Brien O'Keeffe1998: 115) concurred: ʻThere are many ways to examine a text, and the electronic edition must try to make as many of them available to the scholar as possible. Thus the editor becomes the technician who makes it possible for the scholar to assemble the information from the edition he or she would like to explore.ʼ There are several reasons why such a revolution did not occur, but the most relevant one in the present context is that the kinds of resources made available by digitization are of considerably greater use to editors than to the readers of editions. For example, most digital editions include both a digital facsimile of one or more manuscripts plus an edited text, but glosses and textual commentary are linked to the edited text rather than the facsimile, so that the experience of using a digital edition differs only in detail from that of using a conventional critical edition in conjunction with a facsimile. The utility of digitization to editors, as opposed to end users, is well illustrated by a project outside the realm of English studies, Skaldic Poetry of the Scandinavian Middle Ages (Clunies Ross et al. Reference Rossouw and van Rooy2012). Here more than fifty contributing editors work under the direction of six general editors, preparing editions of skaldic verse in both print and online versions, many of the materials necessary for which are made available on the project web site, including manuscript images, transcripts of the manuscripts, dictionaries in electronic databases, concordances, digital scans of major editions and reference works, and detailed editorial guidelines, among many others. From projects like this one it should be apparent that digitization has the potential to breathe new life into the field of philology simply by rendering its practice less daunting and inconvenient. But ʻpotentialʼ is perhaps not the right word, since many editors can testify that digital resources have already contributed enormously to the compilation of editions that have appeared only in print.

The emphasis on manuscript text and incidentals championed by the New Philology dovetailed with an important and influential feature of the LALME project: its insistence upon the graphemic, rather than phonological, basis of dialect variance. It is not, of course, that all graphic variation is void of phonological significance, but some of it is (e.g., the alternation of <gh> and <ȝ> in words of identical derivation in Middle English manuscripts), and so methodological consistency demanded that LALME reflect a dialectology of purely graphemic variation. The confluence of such a dialectology of signs with the rising impulse toward a version of philology based on manuscripts has resulted in a strong infusion of philological concerns in the areas of historical dialectology, pragmatics, and corpus linguistics. In regard to historical dialectology, this is evident in regard to the ongoing revisions to and digitization of LALME and the related dialect projects based in Edinburgh (see Institute for Historical Dialectology, currently Angus McIntosh Centre for Historical Linguistics).

In the recently developed area of historical pragmatics, philological concerns have come to play an increasingly prominent role – so much so that a term has been coined for this area of study, ʻpragmaphilologyʼ, which ʻdescribes the contextual aspects of historical texts, including the addressers and addressees, their social and personal relationship, the physical and social setting of text production and text reception, and the goal(s) of the textʼ (Jacobs and Jucker Reference Jacobs, Jucker and Jucker1995: 11; see also Chapter 3 by Mazzon in this volume). Because pragmaphilology was developed long after literary criticism ceased to be dominated by philological methods, differentiating its concerns from those of literary criticism has been a matter of some concern (see, e.g., Taavitsainen and Fitzmaurice Reference Fitzmaurice2007: 22–5, Pakkala-Weckström Reference Pakkala-Weckström, Jucker and Taavitsainen2010: 219). Yet most studies in this area are concerned with the same sorts of issues that have historically dominated the field of pragmatics and are not common topics in literary criticism. For example, Pakkala-Weckström (Reference Pakkala-Weckström, Jucker and Taavitsainen2010), in surveying studies of Chaucer in the pragmaphilological mode, lists their concerns as speech acts (insults, threats, and promises), forms and pronouns of address, personal affect features (including oaths, exclamations, modal auxiliaries, and deictics), and politeness strategies.

One similarity between classical philology and pragmaphilology is their connection to textual editing, since it is an express concern of historical pragmatists that their findings be taken into account in the production of critical editions (see, e.g., Busse and Busse Reference Busse, Busse, Jucker and Taavitsainen2010: 248). Yet the connection to textual editing is not an uncomplicated one, since a particular aspect of pragmaphilology that moves beyond the traditional bounds of linguistic pragmatics in the direction of some of the historical concerns of classical philology and the New Philology is its emphasis on the use of manuscripts or diplomatic texts rather than edited texts (see Lass Reference Lass, Dossena and Lass2004b). The rationale for this preference is the observation that manuscript texts are closer to authentic representations of historical language varieties than edited texts, since edited texts are interpretative. Editors alter manuscript readings, admit some scribal changes and omit others, supply modern punctuation, and, especially in the case of Old English, alter word or morpheme divisions (see Dury Reference Dury, Dossena and Fitzmaurice2006: 195–9). Even if the unaltered manuscript readings are not unmediated examples of natural language, they are closer to the source than any edited version. Some of the obstacles to using manuscript texts the way living informants are used in synchronic studies were pointed out above, in connection with the discussion of the artificiality of medieval scribal conventions, but the more relevant concern is probably not whether manuscripts are somehow more ʻauthenticʼ but whether the differences between edited and manuscript texts have sufficient impact upon pragmatic studies to justify the considerable effort and philological expertise required to work with manuscript texts.

One study that demonstrates with particular clarity the value of manuscript study in the context of historical pragmatics is a paper by Stenroos (Reference Stenroos2010) that addresses the question whether William Langland, the author of the Middle English Piers Plowman, used ye (as opposed to thou) as a singular pronoun, the way vous may be used in French. Prior studies have sometimes held that singular ye in this text is due to scribal interference. The issues are complex, since Piers Plowman is preserved in at least three versions in more than fifty manuscripts, and so a great deal of philological spadework is required to allow Stenroos to reach the conclusion that scribal alteration of the pronouns of address was minimal, and Langland was as apt as Chaucer to use singular ye.

Since much research in historical pragmatics is conducted on the principles of corpus linguistics, it is natural that scholars concerned with fidelity to manuscript conditions should want to use corpora in which the details of those conditions are encoded. Considerable discussion has in fact been devoted to the problems and possibilities associated with the compilation of philologically suitable corpora and editions: see, e.g., Bailey Reference Bailey, Curzan and Emmons2004, Dury Reference Dury, Dossena and Fitzmaurice2006, Kytö et al. Reference Kytö, Walker and Grund2007, along with Honkapohja et al. Reference Honkapohja, Kaislaniemi, Marttila, Jucker, Schreier and Hundt2009, Kytö et al. Reference Kytö, Grund and Walker2011b, and Marttila Reference Marttila2014.

Attention to manuscript features is a prominent aspect of a number of corpus encoding projects. One encoding project that is particularly thorough in marking manuscript features is the Middle English Grammar Corpus (MEG-C). For example, expunction (indication of deletion by the insertion of points) is indicated by the switches <exp>…</exp>, with similar switches for rubrics, underlining, crossed-out text and rubbed-out text, and for various kinds of additions to the text, among them interlinear and marginal insertions, as well as for different causes of illegibility. While the main text is transcribed in capital letters, the expanded letters of abbreviations are lower case, and the flourishes at the ends of words that in Latin texts indicate abbreviations but in Middle English texts notoriously may sometimes be nothing more than decoration are recorded with a tittle (∼). Capitals in the manuscript itself are indicated with a preceding asterisk (^*), and extra-large capitals by a pair of asterisks. There is thus a great deal of philological information encoded in the project's transcripts, since this may be relevant especially to the study of pragmatics in older written sources. Some other projects that encode manuscript features are Digital Editions for Corpus Linguistics (DECL, Honkapohja et al. 2012); the Corpus of Scottish Correspondence (CSC, Meurman-Solin Reference Meurman-Solin, Lenker and Meurman-Solin2007); and the Málaga Corpus of Late Middle English Scientific Prose (Miranda García 2011).

Corpus linguistics thus points the way to facilitating increased attention to the influence of extralinguistic factors on historical texts (see Chapter 8 by López-Couso in this volume). Even with the advances marked by corpus linguistics, obstacles remain to the unqualified application of linguistic methods to historical texts, such as variation in the principles and degree of precision employed by editors and compilers of corpora to record the text of their primary sources, the inevitable loss of information whenever a handwritten text is transcribed in machine-readable form, and problems in equating the analysis of written texts with that of spoken language. Still, recent developments plainly hold forth the promise of a reinvigorated role for philology in historical linguistics.

6.6 Conclusion

Although philology has probably never been in any real danger of dying out altogether as a methodology, the establishment of linguistics and literary criticism as separate disciplines certainly lessened its influence, and it continually faces the problem of replenishing the ranks of competent philologists, given how many different epistemologies must be mastered for its effective prosecution. Yet the successful integration of philological concerns into historical dialectology, pragmatics, and corpus linguistics in relation to handwritten and printed sources from the Middle Ages to very recent times is reason for optimism about the future of philological pursuits, as are comparable developments in medieval and early modern studies without a particular linguistic focus, since there digital resources likewise have rendered philological interests more accessible and convenient to pursue.

Despite persistent calls by prominent literary theorists for a ʻreturn to philologyʼ as a way of reinvigorating the study of language and literature (see, e.g., de Man Reference de Man1986: 21–6, Patterson Reference Patterson and Van Engen1994, Gumbrecht Reference Gumbrecht2003, Said Reference Said2004: 57–84; cf. Ziolkowski Reference Ziolkowski2005), such a development seems unlikely to occur. Yet philology still has a vital role to play in all the scholarly fields it once encompassed, and particularly in the area of English historical linguistics, both in its more traditional form as a method for dealing with the obstacles to the linguistic analysis of historical data and in its more recent employment as a means to access long-neglected aspects of historical texts through manuscript study.

7.1 Introduction

Printing was not introduced into England until the final quarter of the fifteenth century; before then books were copied by hand and are known today as ‘manuscripts’. As a consequence, the vast majority of sources for studying Old and Middle English are in manuscript form. Because of their individual and often idiosyncratic status, manuscripts must be treated differently from early printed books, which were generally reproduced in large numbers of very similar copies and therefore present fewer problems. As a consequence, this chapter will focus primarily on the issues involved in analysing manuscripts, although there is some discussion about the specific problems encountered when dealing with printed texts at the end of the chapter. In this chapter, then, I will consider the various scholarly approaches that are appropriate for dealing with these types of evidence, and examine the resources available to researchers wishing to exploit them.

7.2 Why work with manuscripts?

Since many Old and Middle English works are conveniently available in modern editions, handily accessible on the shelves of most research libraries, you may wonder why it is necessary to consult manuscript materials at all. The reason is that modern editions are not straightforward reproductions of a medieval original; all modern editors make interventions designed to make the text more accessible to a modern audience. The extent to which a modern editor intervenes when reproducing a medieval work depends on the kind of editorial approach adopted. There are essentially three types of edition, each of which treats the original source in very different ways (Greetham Reference Greetham and Gibaldi1991, Reference Greetham1994, McCarren and Moffat Reference McCarren and Moffat1998). The most useful type of edition for the historical linguist is the diplomatic edition, which aims to represent the manuscript source as closely as possible. Rather than trying to reconstruct an authorial original from all surviving copies, a diplomatic edition seeks to supply a close reproduction of a single witness. An example of the diplomatic edition is the series of editions of the various manuscript witnesses to the Early Middle English work Ancrene Riwle/Wisse, published by the Early English Text Society (Day Reference Day1952, Wilson Reference Wilson1954, Tolkien Reference Tolkien1962, Mack and Zettersten Reference Mack and Zettersten1963, Dobson Reference Dobson1972, Zettersten Reference Zettersten1976, Zettersten and Diensberg Reference Zettersten and Diensberg2000). Rather than issuing a single, conflated, edition of this complex text, the EETS commissioned a series of diplomatic editions of each of these important witnesses, enabling readers to analyse them as manuscripts in their own right and carry out their own collations and comparisons across witnesses. These editions are most useful because of their principle of fidelity to the manuscript witness, although the editorial tendency to introduce modern punctuation, expand abbreviations, combined with the necessity of altering the text's layout, mean that the diplomatic edition will never fully replace consultation of the original manuscript, or a high-quality facsimile reproduction. One further important caveat worth bearing in mind when working with diplomatic editions is that editors tend to differ in the extent to which they consider it acceptable to interfere with the text they are editing. While modern editors tend to restrict this kind of tinkering to a minimum, editions produced in the nineteenth and early twentieth centuries are often considerably more interventionist.

Related to the diplomatic edition is the ‘best-text’ edition, which chooses one of the surviving manuscripts as its base text and follows it closely in establishing its text. This edition differs from the diplomatic edition in allowing an editor to alter his base text by drawing upon other witnesses where his base manuscript is lacking text or is clearly inferior. The degree of such editorial intervention is often dictated by the audience for which the edition is intended. Editions designed for a general or undergraduate readership will typically involve modernization of spelling and the introduction of modern punctuation at the very least, whereas an edition intended for a scholarly audience is more likely to preserve the spelling of its base manuscript. A good example of a best-text edition is Norman Blake's (Reference Blake1980) edition of Chaucer's Canterbury Tales, based closely upon the ‘Hengwrt’ manuscript (National Library of Wales MS Peniarth 392D), preserving its spelling and grammar, but modernizing its punctuation.

The edition that causes the most difficulties for the historical linguist is what is known as an ‘eclectic’ or ‘open’ edition. While other types of edition are based upon a single manuscript witness, this edition selects each reading from all of the surviving witnesses on the basis of its individual merits, regardless of its manuscript support. The key distinction here is that, while a best-text editor sticks closely to his base manuscript unless it is clearly erroneous or deficient, the eclectic editor frequently departs from the base manuscript in favour of readings from other sources. A key principle behind the selection of individual readings is known as ‘lectio difficilior’, meaning that the editor identifies the ‘more difficult’ reading, that is, the reading that is least likely to be the result of scribal error and consequently most likely to be authorial. This is supplemented by consideration of other factors such as consistency of grammar, lexis, authorial linguistic habits, and metrical features. Where best-text editors place their faith in a particular witness to be more reliable than others, eclectic editors place greater confidence in their own ability to determine a correct reading from a collection of spurious ones. An example of eclectic editing can be found in the three volumes of the Athlone edition of William Langland's Piers Plowman; while each volume has a base manuscript text from which the edited text's spelling and grammar are derived, the editors frequently introduced readings from other witnesses where these are considered to be superior (Kane Reference Kane1960, Kane and Donaldson Reference Kane and Donaldson1988, Russell and Kane Reference Russell and Kane1997).

Electronic editions of individual texts or manuscripts are becoming more common, especially of the works of major canonical writers such as Chaucer and Langland, of important manuscript anthologies, such as the Exeter Book (Muir Reference Muir2000) and the Auchinleck Manuscript (Burnley and Wiggins Reference Burnley and Wiggins2003), or utilitarian sources, such as witness depositions (see ETED Reference Kytö, Grund and Walker2011 and Chapter 27 by Walker in this volume) or recipe collections (see Marttila Reference Marttila2014). The Canterbury Tales Project has published CD-ROM editions of individual parts of the work, covering the General Prologue, Miller's Tale, Wife of Bath's Prologue, and Nun's Priest's Tale (Robinson Reference Robinson1996, Reference Robinson2004, Solopova Reference Solopova2000a, Thomas Reference Thomas2006). Rather than privilege a single witness, or reduce the numerous witnesses to a single conflated text, these editions offer diplomatic transcriptions of all extant manuscript and pre-1500 printed witnesses to the work. These transcriptions stick very closely to the manuscript text, preserving scribal punctuation and retaining abbreviation marks rather than expanding them. The Piers Plowman Electronic Archive differs from the Canterbury Tales Project in focusing exclusively on single-witness format editions; seven CD-ROM editions of manuscripts of Langland's poem have been released so far. There are a further two CD-ROM editions issued by the Canterbury Tales Project which focus on single witnesses. The first of these is a digital facsimile of the Hengwrt manuscript, complete with accompanying transcription of the full text of this manuscript, as well as that of another important early witness, the Ellesmere manuscript. The second full-text edition presents the two editions of the Canterbury Tales issued by William Caxton (1476, 1482) (see Bordalejo Reference Bordalejo2003). This last provides valuable evidence for the study of early printed texts, a surprisingly neglected area. Access to data for the analysis of early printed books has been significantly facilitated in recent years by the availability of Early English Books Online (EEBO), a resource which provides access to images of complete copies of some 125,000 texts printed between 1475 and 1700. In addition to these scanned images, EEBO also offers fully searchable transcriptions of approximately 70,000 of these printed works. But, while this is a useful collection of electronic data, it does not represent a balanced or representative selection of data and cannot, therefore, be used in the same way as a corpus assembled according to modern scholarly standards (see Chapter 8 by López-Couso in this volume).

The advent of electronic methods for digitizing manuscripts and making them available via CD-ROM or the World Wide Web has begun to transform the ways in which historical linguists access and analyse original manuscripts. While access to original documents via printed facsimiles or editions is limited in various ways outlined above, digitization of manuscripts is transforming scholarly access to original sources. But while this is a development to be gratefully welcomed by scholars, there are certain reservations which need to be considered. Digitization projects tend to be driven by librarians rather than scholars; consequently, the end results may not be best suited to the research needs of an individual researcher. Libraries may choose to focus their energies on manuscripts of particular cultural significance, or those which are visually particularly impressive, rather than on documents of particular linguistic significance. Where libraries do aim to provide digital surrogates of complete collections rather than just the best-known instances, these often encompass a highly diverse selection of manuscripts, containing works of various periods in a variety of different languages. An important and very valuable exception to this policy is the John Rylands Library in Manchester, which has produced a focused resource by bringing together a digital collection of its Middle English manuscripts, accompanied by physical descriptions of the manuscripts and their contents. Because of funding constraints, libraries often make available images of individual folios of particular manuscripts, rather than complete codices, restricting the kind of analysis that is possible. Even more useful are examples of digitized manuscripts where digital photos are accompanied by electronic transcripts of the text. As with any edition of a manuscript, these transcripts are most useful where there has been very little editorial intervention. Unlike printed editions, electronic transcripts accompanied by digital photographs of the manuscript can be checked for accuracy and for fidelity to the original. Another huge advantage over conventional printed editions is that electronic transcripts are searchable, allowing a researcher to capture all instances of individual forms under analysis. As we will see below, one of the limitations imposed by working with manuscript sources is that it is often necessary to rely on sampling of data; with electronic analysis it becomes possible to interrogate the text in its entirety with complete accuracy, a form of analysis simply not possible when working by hand.

7.3 Working with manuscripts

Working with original manuscripts will in most cases require travel to the repository in which a codex is held to consult it. For a useful discussion of how to approach a repository, and helpful suggestions concerning important preparatory work, see the relevant section in Clemens and Graham (Reference Clemens and Graham2007: 71–81), Introduction to Manuscript Studies. This is less problematic in the case of early printed texts, which frequently survive in multiple copies; in such cases all that is needed is access to a major research library. In some instances it will be possible to employ a printed facsimile in place of the original, but only a small proportion of the surviving corpus of medieval manuscripts has been published in this format. Where facsimiles are available, they must be used with caution, since they frequently do not faithfully preserve certain physical features of a manuscript, such as its original dimensions or its use of coloured inks.

For those wishing to work on manuscripts containing Middle English verse texts, an invaluable bibliographical resource is the New Index of Middle English Verse (Boffey and Edwards Reference Boffey and Edwards2005). An additional tool which surveys the same material is available as an online open-access database, compiled and edited by Linne Mooney, Daniel W. Mosser, and Elizabeth Solopova: the Digital Index of Middle English Verse. Although both resources set out to cover the same manuscript materials, the digital edition is as yet incomplete; however, once complete it will ultimately have the advantage of allowing rolling revision, correction and updating. Both of these resources are based upon, and ultimately replace, the earlier The Index of Middle English Verse (Brown et al. Reference Brown and Robbins1943) and its Supplement (Brown and Robbins Reference Brown and Robbins1965). Middle English prose texts which have been made available in modern printed editions are catalogued by Lewis et al. (Reference Lewis, Blake and Edwards1985). Manuscript copies of prose texts are surveyed in the ongoing Index of Middle English prose project, whose individual volumes supply complete listings of prose contents in Middle English manuscripts by repository. To date, twenty volumes in this series have been published, comprising the major British libraries, as well as smaller and lesser-known collections, e.g. Yorkshire Libraries (Pickering and Powell Reference Pickering and Powell1989), Midland Libraries (Edden Reference Edden2000), Parisian Collections (Simpson Reference Simpson1989), and Scandinavian Collections (Taavitsainen Reference Taavitsainen2007 [1994]). These volumes enable researchers to identify texts by their incipits (opening words of a text), and to locate additional witnesses of a particular text. For researchers interested in scientific and medical writings, an important resource is the Scientific and Medical Writings in Old and Middle English: An Electronic Reference (eVK) by Linda Ehrsam Voigts and Patricia Deery Kurtz (Reference Voigts and Kurtz2000), an electronic database published on CD-ROM, which contains records of texts extant in almost two thousand manuscripts.Footnote ¹ A further resource covering Late Middle English scientific materials is the Malaga Scientific Prose corpus, which covers the contents of manuscripts dealing with medicine, botany, and pharmacopoeia held in the Hunterian Collection at Glasgow University Library and the Wellcome Library in London. Because individual volumes deal exclusively with either verse or prose, and only with texts written in Middle English, these surveys do not supply a complete inventory of a manuscript's contents. For researchers whose interests lie in an individual manuscript rather than a particular text, there are various resources which offer a physical description of the codex, a complete listing of its contents, and a bibliography of relevant published works. An important starting point is a library catalogue, which generally offers descriptions and listings of contents of all manuscripts in a particular collection. However, not all libraries have published catalogues; in some cases where catalogues are available these are outdated and, at best, provide only skeleton descriptions. In recent times, some of the larger repositories, including the British Library and the Bodleian Library, have begun the process of updating their catalogues and making them available online. Because such projects are extensive and expensive, they are progressing slowly. The Parker Library in Cambridge, which houses one of the largest and most important collections of Anglo-Saxon manuscripts in the world, has published electronic facsimiles of all of its manuscripts, complete with an online catalogue. However, the catalogue itself remains essentially that compiled by M. R. James (Reference James1909–13). Another important published resource is a modern critical edition of a work or works found in the manuscript. While editions vary as to the amount of detail supplied concerning manuscript witnesses, all modern editions will supply some account of the manuscript's date, contents, and further references, which should be consulted before beginning an analysis. Editions published by the EETS, one of the most important series of Old and Middle English editions, include brief descriptions covering features such as date, physical dimensions, materials, contents, layout, and provenance. The standard resources for beginning research into Old English manuscripts are N. R. Ker's Catalogue of Manuscripts Containing Anglo-Saxon (Reference Ker1990 [1957]), and Helmut Gneuss's Handlist of Anglo-Saxon Manuscripts (Reference Gneuss2001, Reference Gneuss2003, Reference Gneuss2011). Descriptions of medieval manuscripts in British Libraries outside the major repositories can be found in the four volumes compiled by N. R. Ker, Medieval Manuscripts in the British Isles, with a further volume containing Indexes and addenda edited by I. C. Cunningham and A. G. Watson (1969–2002). For the post-medieval period there are the four volumes of Peter Beal's Index of English Literary Manuscripts (1980–93), covering the period 1450–1700. This will be supplemented by the ongoing Catalogue of English Literary Manuscripts, which will offer a freely accessible online database comprising descriptions of thousands of manuscripts representing the work of more than 200 authors for the period 1450–1700. For listings of early printed material, researchers should consult the English Short Title Catalogue (ESTC), which is a comprehensive listing of early printed books, newspapers, and printed ephemera from the period between 1473 and 1800. The database contains more than 450,000 entries, representing the collections from more than 2,000 libraries from around the world, and is available online via an institutional subscription.

7.4 Dating and localizing sources

One of the problems of dealing with manuscript sources is that they seldom contain dates or places of production: information which we are accustomed to finding in printed texts. There are some exceptions to this: chronicles that consist of a series of annalistic entries preserve dates, although we cannot always assume that the date of an entry is necessarily the date at which the entry was written. A good example of this is the Peterborough Chronicle, which was written in distinct periods: the so-called First Continuation covers the period up to 1131, while the Final Continuation, stretching from 1132 to 1154, was written retrospectively in 1154 (Clark 1970). But in most cases surviving manuscripts are undated. To make matters more complex, there is often a considerable time lag between the composition of a text and the date on which a particular manuscript was copied. Which of these dates should be the basis of an analysis of the language of that particular manuscript? An awareness of the difficulties of precise dating of sources can be particularly important when considering texts that straddle period boundaries. For instance, there are a number of texts copied in the twelfth and thirteenth centuries which are copies of Old English originals; should these be considered to be instances of Old or Middle English? Several late Middle English texts survive only in copies produced in the sixteenth or seventeenth centuries. A good example of this is the single extant witness to the poem The Destruction of Troy, Glasgow University Library MS Hunter 388; this manuscript was copied in the mid-sixteenth century, although the text it preserves was composed c.1400 (Panton and Donaldson Reference Panton and Donaldson1869, 1874). Drawing boundaries that divide up the history of English into distinct historical periods is a controversial activity which is made considerably more complex once one takes into account the evidence for the dating of the sources upon which these scholarly entities are constructed (Kitson Reference Kitson and Fisiak1997, Lass Reference Lass, Taavitsainen, Nevalainen, Pahta and Rissanen2000a).

Another difference between manuscript and printed books is the lack of explicit evidence for a manuscript's place of production. This information is of particular significance for the study of regional variation; without such information one can study a manuscript's evidence for dialect usage, but be unable to make a statement as to which dialect is being investigated. Even in cases where a manuscript's provenance can be established, this may not have any bearing upon the scribal dialect it preserves. The Vespasian Psalter (British Library Cotton Vespasian a.1) can be confidently located in Canterbury in the ninth century, when an interlinear gloss was added in Old English (S. H. Kuhn Reference Kuhn1965). But the dialect of the glossator can be shown to be Mercian rather than Kentish, reminding us that scribes, like manuscripts and texts, were mobile. There are some exceptions to the lack of an established provenance: legal documents, parliamentary records, memorandum books of local trade and religious guilds, ecclesiastical charters, and so on. However, documents of these kinds frequently pose problems of their own that limit their usefulness as linguistic documents: they are frequently short, formulaic, and are often written in highly conservative forms of language.

7.5 Dealing with variation

Another issue associated with manuscripts is that, unlike printed books, where a single text may be reproduced in numerous identical copies, manuscript books vary from each other to a greater or lesser extent. The works of the major Middle English literary authors John Gower, Geoffrey Chaucer, and William Langland are found in over fifty manuscript copies each (for details see the relevant entries in Boffey and Edwards Reference Boffey and Edwards2005). Anonymous texts such as Cursor Mundi, Prick of Conscience, and Speculum Vitae survive in large numbers of copies. These witnesses typically span a period of some hundred years from the date of composition, and have been copied throughout the country in a wide range of regional dialects enabling scholars to map their transmission through a study of their dialects (Lewis and McIntosh Reference Lewis and McIntosh1982, Beadle Reference Beadle, Laing and Williamson1994). More recently, attempts have been made to exploit hypertext technology to provide ‘cultural mappings’ of the 183 surviving copies of the Middle English prose Brut, produced between the fourteenth and seventeenth centuries, bringing linguistic evidence into dialogue with provenance, textual affiliation, readership, and patronage (the Imagining History project). While some of these extensive manuscript traditions preserve copies that are closer in time and location to the authorial original, there are very few instances of medieval works which survive in manuscripts copied by the author himself, known as a ‘holograph’ manuscript; nearly all such manuscripts are at least one stage removed from the author's original copy. The small number of instances of authorial holographs from the medieval period includes Bodleian Library MS Junius 1, containing The Ormulum, written by its author, an Augustinian canon named Orm from Bourne, Lincolnshire, between 1180 and 1190 (Burchfield Reference Burchfield1956, Parkes Reference Parkes, Stanley and Grey1983; see the Orrmulum Project by Nils-Lennart Johannesson). British Library MS Arundel 57 contains the Ayenbite of Inwyt, containing a colophon stating that its author, Dan Michel, wrote it in his ‘oȝene hand’ in Canterbury in 1340, when he would have been at least seventy years old (Gradon Reference Gradon1965–79, Hanna Reference Hanna and Kelly2011). Fifteenth-century holographs include manuscript collections of his own verse by the London poet Thomas Hoccleve and copies of works by the hagiographer and Augustinian friar of Lynn, Norfolk, John Capgrave (Lucas Reference Lucas1997, Burrow and Doyle Reference Burrow and Doyle2002). While these texts might seem ideal candidates for linguistic analysis, they represent but a small proportion of the extant corpus of medieval manuscripts. Furthermore, they represent a limited range of dialects, London, Norfolk, Lincolnshire, and Kent, leaving large sections of the country completely unrepresented. Literary texts of the Old English and Early Middle English periods tend to survive in smaller numbers of copies than ones composed in the later Middle English period (Laing Reference Laing1993). The Early Middle English period also suffers from a patchy coverage of texts; the majority of surviving manuscripts can be localized to the Western or Eastern counties, leaving the central Midlands poorly attested (Laing Reference Laing, Taavitsainen, Nevalainen, Pahta and Rissanen2000).

Another issue that needs to be confronted when analysing manuscripts concerns the scribes’ treatment of the language of their exemplars. In a seminal essay published in 1963, Angus McIntosh observed that a scribe setting out to copy an exemplar could approach the task in one of three ways:

1. 1 He could copy his exemplar exactly: carrying out a ‘literatim’, i.e. letter-by-letter, transcription of the original.

2. 2 He could translate the language of his copytext into his own dialect.

3. 3 He could do something in between types 1 and 2, thereby producing a mixture of his own forms and those of his exemplar.

Subsequent work in the field of Middle English dialectology by Michael Benskin and Margaret Laing (Reference Benskin, Laing, Benskin and Samuels1981) revealed that Early Middle English scribes, who were more accustomed to copying a fixed language such as Latin, and to writing in a more formal script, were more likely to copy literatim, preserving the language of their copytexts with considerable fidelity. A good example of a literatim scribe is found in British Library MS Cotton Caligula A.ix of the Owl and the Nightingale. This text was copied in the late thirteenth century by a single copyist but contains two completely distinct spelling systems, with abrupt switches from one system to the other. The only possible explanation for this scenario is that the scribe was copying literatim an exemplar that was written by two different scribes with two different spelling systems (Stanley Reference Stanley1960).

In the later Middle English period it appears to have been more common for scribes to translate the language of their copytexts into their own usages, taking us further away from the language of the exemplar and the original from which it ultimately derives. While this copying practice tends to obscure all traces of an author's own usage, it does shed considerable light on a wider range of Middle English varieties than if all such copyists simply preserved the spelling of the authorial text. An important exception to this generalization concerns the fifteenth-century manuscript tradition of Gower's Confessio Amantis, a work composed in the late fourteenth century. Gower employed an idiosyncratic dialect, comprising a mixture of Kentish and Suffolk features (Smith Reference Smith and Echard2004), which was deliberately preserved by later copyists throughout the fifteenth century (Smith Reference Smith, Samuels and Smith1988a). By contrast, the manuscripts of works by Chaucer and Langland have been more thoroughly translated into a variety of scribal varieties, obscuring much of the evidence for the authors’ own usages, but providing us with a wider range of scribal usages than would have been the case if all such scribes had merely reproduced those of their originals (Horobin Reference Horobin2003, Samuels Reference Samuels1985).

7.6 Approaches to manuscripts

A particularly important development in the use of manuscript evidence for the study of linguistic variation in Middle English was the Middle English Dialect project, which culminated in the publication of A Linguistic Atlas of Late Mediaeval English (LALME). The LALME project began in the 1950s and was published in 1986; an electronic version of LALME has recently been released online, hosted by the Institute for Historical Dialectology at the University of Edinburgh.

The Middle English Dialect project based its analyses on original manuscripts where possible, although in many cases it relied on modern editions. Modern printed sources utilized in the LALME survey include editions of historical documents, such as wills, deeds, charters, and collections of letters, often published by local history societies. Because these texts were edited for historians rather than for linguists, and because they were frequently published in the nineteenth or early twentieth centuries, they are often inconsistent in their fidelity to the language of the originals. But reliance upon edited texts rather than original manuscripts was not limited to historical documents. Because of the difficulty of getting access to original sources, the LALME editors frequently turned to edited versions of literary works as well. Works which survive in diplomatic editions in parallel text format were particularly well suited to this kind of use. A good example of a text treated in this way is the Cursor Mundi, which was edited by Richard Morris for the EETS in 1874. This edition prints four important manuscript copies of the work in parallel, as diplomatic transcripts with minimal editorial intervention.

As a further compromise, necessitated by the length of some Middle English works, the LALME editors analysed tranches of text rather than texts in their entirety. In the case of the Cursor Mundi, for example, the chosen extract comprises the first 2,000 lines of the poem. The most obvious difficulty with this manner of analysing the sources is that it supplies only a partial account of the linguistic features of a particular manuscript. It may be that this partial account is entirely representative of the remainder of the manuscript, but it is also possible that it is not. Additional forms not recorded in the opening 2,000 lines might occur at a later point, while forms found in the opening section might fall out of use later in the text. We saw above, in discussing McIntosh's typology, that scribes were prone to change the way they copied during the transcription process. While the linguistic profiles make no attempt to give exact counts for particular forms that they record, they do supply an indication of relative frequency of usage, placing less-common forms in single brackets and especially rare forms in double brackets. Although this is little more than a rough way of indicating frequency of usage, it is highly likely that such frequencies would change if the entire text were to be analysed.

If the localizations are to be considered reliable then it is important that the data for each linguistic profile be accurate in their totality. A recent reconsideration of the LALME handling of original sources with reference to British Library MS Cotton Nero A.X, the unique surviving witness to Sir Gawain and the Green Knight and other alliterative poems by the same author, by Ad Putter and Myra Stokes (Reference Adams2007), has shown the extent to which the LALME linguistic profile misrepresents the language of the original manuscript through omissions and inaccuracies, leading the authors of this study to conclude that ‘LP 26 is so imprecise that it can hardly be trusted to provide more than a rough indicating of the scribe's locale’ (p. 487). Another significant limitation in the LALME profile concerns the very partial reporting of context and distribution. For instance, under the item SHE (third-person singular pronoun), LALME LP 26 reports the majority form ho, and a minor variant scho is given in double brackets. But what this indication of frequency fails to note is that five of the six instances of scho are found to cluster in a short stretch of text within Gawain. This distribution is significant because it strongly suggests that the forms are not those of the scribe, but that they were inherited from an exemplar, one which is likely to be different from that used for the other poems in the Cotton Nero manuscript.

Once again, we see that the difficulties presented by the availability of original sources led to a number of compromises. Instead of using only original materials, the LALME editors were compelled to draw upon edited texts, including ones which were not especially faithful to the linguistic features of the original manuscript. The sheer scope of the project also led to compromises in the amount of data to be analysed, which resulted in linguistic profiles which provide only a partial account of the variant forms attested in a particular source, and only a relative indication of their frequency and distribution.

These limitations were partly necessitated by the form of the original sources and difficulties of access; the successor project to LALME, the Linguistic Atlas of Early Middle English (LAEME), has attempted to provide a much more comprehensive overview of the earlier Middle English period. Because of the more restricted corpus, LAEME has produced an extensive electronic corpus of the surviving textual materials, rather than basing its own survey on a selective analysis. As a result, the LAEME project has made available an electronic corpus of Early Middle English texts that has been explicitly produced for linguistic analysis, preserving features of spelling, punctuation, abbreviation, and even, in some cases, distinctions in letter forms. In addition to providing the transcriptions, the LAEME editors have also applied a series of tags, encoding lexical meaning and grammatical function, for every word in their corpus, allowing it to be searched by part of speech and word, as well as by spelling form. Having completed this project, the LAEME team has now returned to the LALME project and released an online version, with a number of revisions and corrections of the data presented in the original publication. Although revision of linguistic profiles has necessarily been piecemeal, given constraints of time and funding, these have generally set out to address some of the perceived shortcomings described above, replacing profiles that were based upon untrustworthy editions with ones taken directly from manuscripts and separating out profiles which represent the work of more than one scribe.

Since the publication of LALME, scholars have sought to build upon and extend this valuable resource. Researchers at the University of Glasgow and the University of Stavanger are constructing an electronic corpus of Middle English texts based upon localizations provided in LALME. To date the corpus comprises 500,000 words of Middle English, transcribed directly from manuscript or from a printed facsimile. The ultimate goal of this project is the compilation of a Middle English Scribal Texts Archive (MESTA), which will comprise a series of corpora, with texts arranged according to a wide range of linguistic and non-linguistic criteria. Rather than focusing simply on geography as a means of understanding linguistic variation, this project will consider a broader range of criteria for classification, including codicological phenomena such as script, decoration, mise-en-page, and other textual factors.

Researchers at the University of Birmingham exploited the LALME localizations in a different way, in order to describe the ‘literary geography’ of the West Midlands area. The Manuscripts of the West Midlands project produced A Catalogue of Vernacular Manuscript Books of the English West Midlands, c.1300–c.1475. The completed resource includes descriptions and images of 150 manuscripts whose scribal dialects were located by LALME to the West Midland counties of Gloucestershire, Herefordshire, Shropshire, Staffordshire, Warwickshire, and Worcestershire. By bringing together both linguistic and palaeographical evidence, this resource opens up new possibilities for an integrated approach to the study of language and script.

The analysis of the outputs of prolific scribal copyists is another valuable method of analysing linguistic and palaeographical variation in combination. Research in this field builds upon Jeremy Smith's groundbreaking work on an important copyist of Middle English literary manuscripts (known as ‘Scribe D’) (Smith Reference Smith, Samuels and Smith1988b); the recent completion of the Late Medieval English Scribes database (Mooney et al.), freely accessible over the web, has made available palaeographical and dialectological information for several hundred Middle English scribes. Using this database it is now possible to construct detailed linguistic profiles, examining how language is conditioned by text-type, register, the status of the particular manuscript. In addition, this resource enables scholars to assemble graphetic profiles of the kind first proposed by Angus McIntosh in Reference McIntosh1975, but abandoned because of the difficulties in handling the quantities of data involved in the pre-electronic era. Another important electronic resource is the database of manuscripts entitled The Production and Use of English Manuscripts 1060 to 1220, compiled by Orietta Da Rold, Takako Kato, Mary Swan, and Elaine Treharne. The post-conquest period, which saw many of the most significant changes in this history of English, is surprisingly under-researched; particularly marginalized are copies of Old English works by important prose writers such as Ælfric and Wulfstan, produced after the Conquest, despite the important evidence they preserve about the way Old English linguistic conventions were preserved or reorganized for a later audience. A focus on literary manuscripts in this period has led to the neglect of a large body of non-literary material, such as legal texts comprising writs, charters, and land grants (Treharne Reference Treharne2012). While such texts can present a specific set of challenges for linguists, because of their textual transmission, length, and formulaic nature, they have been shown to preserve valuable evidence for contemporary usage as well as for the reception of Old English in the eleventh and twelfth centuries (Lowe Reference Lowe2001).

The availability of electronic editions of complete manuscripts enables scholars to extend the LALME analyses of individual manuscripts comprising multiple texts. Above I noted the limitations imposed by the LALME requirement to sample manuscripts, rather than analyse them in their entirety. Now that complete manuscripts have begun to be digitized it has become possible to analyse a scribal output on the basis of an exhaustive linguistic profile. The recent digitization and transcription of the entirety of the ‘Vernon’ manuscript (Bodleian Library, MS Eng. Poetry a.1) has enabled a more fine-grained analysis of the linguistic output of the two scribes responsible for copying the vast collection of texts it comprises than was possible in the two linguistic profiles offered in LALME (Scase Reference Scase2011, Horobin and Smith 2011). The opportunities offered by electronic analysis of complete texts for scribal identification were demonstrated by Alison Wiggins, who used complete scribal profiles to argue that Scribes 1 and Scribes 6 of the Auchinleck manuscript were indeed distinct copyists (Wiggins Reference Wiggins2004, see also Runde Reference Runde, Cloutier, Hamilton-Brehm and Kretzschmar2010). Analysis of scribal profiles of complete transcriptions of manuscripts of Jacob van Maerlant's Scolastica, a Middle Dutch translation/adaptation of Peter Comestor's Medieval Latin Historia scholastica by K. H. van Dalen-Oskam (Reference Dalen-Oskam2012) has implications for scholars working with digitized editions of medieval English manuscripts. Jacob Thaisen (Reference Thaisen2013) has applied similar techniques to copies of early witnesses of Chaucer's Canterbury Tales to determine the likely number of exemplars on which these extant copies were based. Developments in the discipline of historical pragmatics have also drawn upon manuscript evidence, especially features of mise-en-page and layout. Analysis of the organization of the manuscript page – the integration of headings, paragraph markers, the provision of marginal glosses, and the use of red ink and white space – has shed light on the ways in which a manuscript may communicate visually with the reader (see the essays collected in Pahta and Jucker Reference Pahta, Taavitsainen and Pahta2011 and Chapter 28 by Moore in this volume). Close reading of the provision and organization of visual devices, such as speaker markers, can inform our understanding of authorial attempts to communicate the form of a text, as well as the ways scribes and readers responded to such efforts (Echard Reference Echard1997, Pearsall Reference Pearsall, Matsuda, Linenthal and Scahill2004). Study of layout and mise-en-page can also be usefully applied to non-literary manuscripts. In her study of handwritten business correspondence in the nineteenth century, Marina Dossena has emphasized the importance of calligraphy, ornamental lettering and the generous provision of white space as politeness and ‘face-enhancing’ moves which extend beyond strictly linguistic choices (Dossena Reference Dossena, Pahta and Jucker2011). Work in the field of historical sociolinguistics has drawn heavily upon collections of correspondence, since these can provide autograph evidence of a known individual's linguistic usage. But such work has often relied upon edited collections rather than manuscripts, with the consequence that they focus on grammar, lexis, and pragmatics, rather than orthography and phonology. A good example is a study of letter collections of the late medieval and early Tudor periods by Nevalainen and Raumolin-Brunberg (Reference Nevalainen and Raumolin-Brunberg2003), which draws upon the Corpus of Early English Correspondence (CEEC), a collection of 2.7 million words representing 6,000 letters scanned from 96 edited collections. The increasing availability of freshly edited letter collections, such as Alison Wiggins's diplomatic edition of 234 original letters to and from Bess of HardwickFootnote ² (c.1521/2–1608), will make reliance upon untrustworthy sources less necessary. Similar projects have been undertaken under the aegis of the Centre for Editing Lives and Letters at University College London.Footnote ³

A neglected feature of both manuscripts and early printed books of considerable significance for both linguistic and literary studies concerns authorial, scribal, and compositorial punctuation practices. Malcolm Parkes's (Reference Parkes1992) magisterial overview of the history of western punctuation, Pause and Effect, remains the major survey of this field; a more recent study of the use of parentheses by John Lennard (Reference Lennard1991) provides a similar diachronic overview of a single punctuation feature. However, we continue to lack detailed interrogation of specific usages in manuscript and printed books which would enable a more nuanced understanding of the development in what Parkes has termed the ‘grammar of legibility’. Small-scale studies have been carried out on the punctuation of major literary writers (e.g. Killough Reference Killough1982 on Chaucer, McKenzie Reference McKenzie, Salmon and Burness1987 and Warren Reference Warren, Salmon and Burness1987 on Shakespeare), or on specific manuscripts (e.g. Solopova Reference Solopova and Fein2000b on British Library MS Harley 2253), but we lack a more comprehensive, diachronic overview covering both literary and non-literary practices. More recently, attention has been paid to the pragmatic functions of punctuation in specific varieties (e.g. Smith and Kay Reference Smith, Kay, Pahta and Jucker2011 on Older Scots), or specific textual traditions (e.g. Parkes Reference Parkes, Oguro, Beadle and Sargent1997 on Nicholas Love, Mirror of the Blessed Life of Jesus Christ). A recent monograph on methods of marking speech in Middle English manuscripts and early printed texts has demonstrated the value of drawing upon both traditional linguistic features in combination with punctuation practices, in tracing the diachronic development of modes of speech marking in both manuscript and print (Moore Reference Moore2011).

7.7 Manuscript to print

The transforming impact of printing for the establishment of a standard variety of English is generally assumed, but has received surprisingly little detailed study. Relevant work in this area has been focused on individual textual traditions, such as Caxton's editions of Reynard the Fox (Blake Reference Blake1965), Chaucer's Canterbury Tales (Horobin Reference Horobin2001), Chaucer's Romaunt of the Rose (Caie Reference Caie, Pahta and Jucker2011), or Caxton's printing practices more generally (Fisher Reference Fisher1996). The development of new resources which will enable manuscripts to be compared with printed versions of the same text, such as the Malory Project, which brings together the ‘Winchester’ manuscript of Malory's Morte Darthur with Caxton's printed edition of 1469–70, will offer new possibilities for the study of the linguistic implications of the shift from manuscript to print (Kato and Hayward). Printed books of the sixteenth and seventeenth centuries are a particularly neglected area, although detailed work on the printed tradition of Spenser's Shepheardes Calender demonstrates what can be achieved through careful analysis of this extensive body of primary material (Rutkowska Reference Rutkowska2013). Recent attention to the spelling practices of printed copies of early modern dramatic works has supplemented the findings of authorship attribution studies. Douglas Bruster (Reference Bruster2013) has argued that variant spellings found in the Additional Passages attested by the 1602 quarto of Thomas Kyd's play The Spanish Tragedy, lend support to the theory that these lines were authored by Shakespeare. Laurie Maguire and Emma Smith (Reference Maguire and Smith2012) suggest that the First Folio text of Shakespeare's All's Well That Ends Well (1623) preserves contractions and orthographical preferences typical of Thomas Middleton, leading them to propose the play was the product of collaboration between the two playwrights. Unusual spellings witnessed by the Shakespeare First Folio has led Saul Frampton (Reference Frampton2013) to propose that the edition was subjected to metrical, stylistic, and linguistic intervention from the Italian lexicographer and translator of Montaigne, John Florio. The spelling habits of early modern printers more generally have been assessed by N. E. Osselton (Reference Osselton1963, Reference Osselton, Blake and Jones1984); linguistic standards among printers of the eighteenth century have been surveyed by Ingrid Tieken-Boon van Ostade (Reference Tieken-Boon van Ostade, Fisiak and Krygier1998). More recently, Anita Auer has embarked on a collaborative project, Emerging Standards: Urbanisation and the Development of Standard English, c.1400–1700, designed to investigate the development of standard English from the perspective of urban vernaculars of major regional centres.Footnote ⁴ This project will involve producing a corpus of manuscript and printed materials from 1475 to 1700, and analysing it according to factors such as date, text-type, social stratification, and migration patterns. So, while working with original documents is a necessary feature of all historical linguistic research, it presents a number of distinct challenges which must be given detailed consideration before a research project is carried out. While it is possible to use an edited text as a substitute for the original document, this should only be done following a detailed assessment of the nature of the edition and its manner of representing the original and is not an appropriate substitute for the original in many cases. The reliance upon edited texts is particularly problematic for research into orthography, phonology, punctuation, and historical pragmatics. Yet, the increasing availability of manuscript facsimiles and diplomatic transcriptions in electronic form is making reliance upon such sources unnecessary. Particularly neglected areas of research in this field concern the impact of printing upon the written language: what models were adopted by early printers and how standardized were their practices? While scholars have carried out studies of individual printers, or on the relationship between manuscript and print of particular texts, we lack any large-scale investigation into the linguistic usages of the early printers. The developing field of historical pragmatics also raises a number of questions which will be of great interest in future work on manuscript sources, concerning the communicative functions of features of layout and ordinatio. We also lack detailed studies of individual punctuation practices, as well as a clear diachronic overview of developments in usage. The increasing availability of electronic corpora, which bring together large quantities of such data in electronic form are transforming the kinds of research questions that can be asked of such data; this is the subject of the following chapter (Chapter 8 by López-Couso in this volume).

8.1 ‘Old’ material in new formats

The use of textual material as a window into the linguistic system of past stages of the language has a long tradition in English historical linguistics. Classic examples of the use of what Svartvik (Reference Svartvik and Facchinetti2007: 12) calls ‘language corpora BC, i.e. “before computers”’, can be found in the work by James Murray for the Oxford English Dictionary and in Otto Jespersen's A Modern English Grammar on Historical Principles (1909–49), where the various linguistic structures being discussed are conveniently and generously exemplified with quotations from the literary works of renowned authors.

Although English historical linguistics has always been heavily anchored in textual evidence, the last three decades or so have witnessed an increasing interest in the compilation of structured and systematic collections of texts from earlier periods of the language, mostly in computerized form. The availability of ‘old’ material in new formats, including not only electronic corpora, but also electronic dictionaries and online collections of texts, which provide quick and easy access to a large amount and a wide variety of data, has undoubtedly stimulated new research methods and approaches (see, e.g., Szmrecsanyi Reference Szmrecsanyi, Krug and Schlüter2013a and Chapter 2 by Hilpert and Gries in this volume) and has enabled scholars to ask new questions and to reconsider old questions in a different light.

In this context, this chapter offers a compact overview of the kind of material available to students and researchers interested in the analysis of variation and change in the history of English. Given the limitations of space, rather than providing detailed descriptions of the most influential English diachronic corpora and of all their potentialities for research,Footnote ¹ the chapter can only attempt to give a small-scale survey of a selection of these resources (see section 8.2.2), placing these corpora in the context of the history of English, highlighting their advantages and weaknesses, and offering some examples of the kind of data which have been obtained from them. Although the emphasis is on the long diachrony from Old English to Present-day English, sources for the study of recent and ongoing change in contemporary English are considered as well (see section 8.2.3). The chapter also mentions in passing some major concerns related to the design, compilation, and development of historical corpora, among others the issues of corpus size, the restriction of the available data mostly to the written medium, and the lack of sociolinguistic information about many (particularly) early texts.Footnote ²

In addition to corpora, this chapter also pays attention (though in a highly summary fashion) to other types of electronic material which can be employed as corpora or be used to complement the data provided by corpora (see section 8.3). These include, among others, electronic dictionaries and atlases, full-text collections, and online text archives and repositories of texts from different periods, which, though not necessarily designed for linguistic analysis, offer a wealth of possibilities for investigating language variation and change.

The chapter closes with some comments on the future of electronic corpora (see section 8.4), including some suggestions for the compilation of further historical corpora representing, for example, regional grammatical differences and specific registers in various historical periods.

8.2 Computerized historical corpora

8.3 Electronic collections and other digital resources

In addition to computerized corpora (see section 8.2), there has been an exponential growth of other electronic collections and of online resources available to researchers and teachers of English historical linguistics. These include such diverse materials as electronic editions of individual texts and authors, online collections and repositories of texts (see section 8.3.1), electronic dictionaries, thesauruses, and lexicons (see section 8.3.2), linguistic atlases (see section 8.3.3), and even the World Wide Web and recently developed online applications such as Google Books Ngram Viewer (see section 8.3.4). The usefulness of these resources is definitely beyond doubt. Some of them can be used as – or under the guise of – historical corpora (e.g. the Oxford English Dictionary). In turn, digital archives like Early English Books Online (see below) and the Project Gutenberg may provide sources from which researchers can create their own tailor-made corpora which suit best their needs and purposes.Footnote ¹³ On other occasions, these resources can serve as a helpful complement to corpora and compensate for some of their deficiencies. Thus, for example, textual databases, some of which contain millions of words of running text, can be used as an effective means of complementing the information on low- and medium-frequency linguistic phenomena provided by small stratified corpora (see Curzan and Palmer Reference Curzan, Palmer, Facchinetti and Rissanen2006).

8.4 Looking ahead

As shown in sections 8.2 and 8.3, the study of language variation and change in English today relies on an impressive range of high-quality corpora and other electronic and online resources. It seems, however, that despite the considerable progress made over the last couple of decades, the need and interest for the creation of further materials of the kinds surveyed in this chapter is still alive.

The demand for more and better resources is particularly strongly felt in certain research areas in English historical linguistics. The development of new specialized corpora and corpus-like resources is called for in domains like cross-linguistic and interdisciplinary studies (see Kytö Reference Kytö, Bergs and Brinton2012: 1523), multilingualism and other contact phenomena (see Kytö and Pahta Reference Pahta, Nevalainen and Traugott2012: 131–2), and the history of regional varieties of the language. As regards regional variation, although much has been accomplished with the development of dialect-based corpora such as HCOS, CIE, and COHA, among others (see section 8.2.2.2), and electronic atlases such as LAOS (see section 8.3.3), there is still room for improvement. Kortmann and Wagner (Reference Kortmann, Wagner and Hickey2010: 270–1, 290–1), for instance, complain about the lack of sufficient and satisfactory material for the study of non-standard morphosyntax in various periods of the language, and make a plea for ‘appropriate data necessary for adding the many missing pieces to our puzzle of grammatical variation among dialects in the history of English’ (2010: 291). Very little is known also of the historical development of New Englishes, given the scarcity of suitable data. It seems, however, that resources like the 91-million-word Corpus of Legislative Council Proceedings from the Hong Kong Hansard (1858–2012) can offer interesting insights into the evolution of the lexicon of Hong Kong English (see Evans Reference Evans2015) and may serve as a possible model to investigate the history of other varieties of English worldwide.

We further need corpora for specific registers or genres in various historical periods for which extensive documentation in the form of a specialized corpus is still lacking. This is the case, for example, of legal English. Although both the HC and ARCHER contain law texts (see section 8.2.2), the size of the legal components in these two diachronic corpora is too small for a detailed linguistic analysis of legal English across time.Footnote ²⁴ In the near future, the ARCHER material will be complemented by a corpus of British English legal opinions, the Corpus of Historical English Law Reports (CHELAR) which is being compiled by the research group Variation, Linguistic Change, and Grammaticalization at the University of Santiago de Compostela, and which will allow research into the development of this particular text-type from the early seventeenth century to the present day (see Rodríguez-Puente Reference Rodríguez-Puente2011).

The rapid increase of internet facilities, the technological advances in corpus compilation and annotation, and the development of ever more sophisticated software tools for the analysis of corpus data offer promising prospects for research in English historical linguistics in the next few decades. It is indispensable, however, that the historical corpus linguist be ready to combine both quantitative and qualitative approaches to the study of language variation and change, paying attention not only to the unquestionable relevance of statistical corpus findings yielded by structured third-generation mega-corpora, but also to careful philological analysis and close reading of individual examples from more specialized and balanced corpora.

9.1 Introduction

When William Caxton established his printing press in London in 1476 or even when Dr Johnson published his Dictionary in 1755, the phrase ‘listening to the past’ did not make sense except on a metaphorical level. As the present chapter will argue, this has changed, and it is time historical linguists took this change into account in their work. We have had sound recording of the human voice for more than a century – since 1877, to be precise. From the early part of the twentieth century technical quality rose to a level which makes linguistic analysis possible. A large but sometimes badly documented and geographically dispersed body of relevant data has accumulated, and a hundred years is certainly a decent time depth for the study of phonetic change, and the history of the spoken language more generally, on the basis of authentic audio data and in real time.

In order to assess the potential of audio recordings as linguistic data, we have to clarify the relationship between sound and speech, which is trivial only as long as we are dealing with the medium of transmission of the linguistic signal. Koch and Oesterreicher (Reference Koch and Oesterreicher1985) were among the first to systematically explicate the properties of spoken language beyond the level of the medium, coining the terms ‘medial orality’ and ‘conceptual orality’ to capture the different aspects of the phenomenon. With regard to the medium, the distinction between spoken and written (or signed) language is clear. The signal is acoustic in the former case, but visual in the latter. In a ‘medium’ perspective, the value of historical audio recordings therefore resides in the fact that they preserve past acoustic signals, which for the study of past pronunciations is usually preferable to reconstructions based on indirect evidence such as puns, rhymes, orthoepist literature, or other metalinguistic commentary. We can profit from direct access to historical sound irrespective of the discourse function or the sociolinguistic status of the recorded utterance in question.

In Koch and Oesterreicher's terms, what is oral ‘medially’ can be in a complex relation to orality as conceived of ‘conceptually’. A news broadcast and an oral-history interview are both examples of spoken language on the medial level. Conceptually, however, news broadcasts are typically based on written texts composed to be read out on air (and in a standard accent), sharing more of their syntactic, lexical, and textual structure with written news reportage in the press than with a face-to-face discussion of political issues. This may be different in the oral-history interview, which may allow spontaneous speech to surface, even though the recording situation is usually semi-formal and biased toward the narrative monologue; in comparison to the news broadcast, the oral-history interview is also far more likely to feature vernacular linguistic forms in addition to standard ones. A good oral-history interview may easily move from a relatively formal, conceptually ‘written’ opening to genuine medial and conceptual orality later on in the discourse. On the conceptual level, the distinction between speech and writing is obviously no longer binary but gradual, and all sorts of hybrid and mixed manifestations are possible.

In giving direct access to the past acoustic signal, historical audio recordings are a definite boon to historical phonology. However, since there is quite clearly also a place for the historical study of spoken language in a wider context and on the plane of ‘conceptual orality’ as defined above, they may be additionally relevant to the study of the lexicogrammatical and discourse features of older stages of spoken, non-standard, and vernacular English – in other words, to the booming fields of historical pragmatics and historical sociolinguistics. Here, however, a damper of sorts is presented by the fact that especially in the early stages of sound recording the material which was produced tended to be of a conceptually written nature. The voices of the great and the famous recorded in formal settings in the early twentieth century can definitely tell us a lot about standard pronunciation in formal oratory, but not necessarily about the lexicogrammatical peculiarities of informal conversation. For this, a fictional representation of dialogue in a 1920s novel may well be the better source.

On the whole, however, the increasing volume and time-depth of our recorded acoustic heritage is excellent news. It has the potential to revolutionize the study of sound change by widening the window for ‘real time’ studies to almost a hundred years in favourable circumstances, and it will complement and enrich the existing database for historical-sociolinguistic and historical-pragmatic studies of English from the beginning of the twentieth century. The advantages will become even more obvious in the future, because technological advances and more inclusive social notions of what is worthy of being preserved for posterity have led to better coverage of informal-conversational and vernacular language. Audio recordings have thus definitely begun to help fill a gap which was sorely felt by some of the great language historians of the past.

Hermann Paul, pioneer of nineteenth-century historical and comparative linguistics, for example, was keenly aware of the tension between the spoken and written modalities of language in synchrony and diachrony:

Most language change originates in the colloquial baseline style which Paul mentions and which is characterized by the following four features: it is spoken, spontaneous, vernacular, and interactive/dialogic. For the greater part of the history of English, however, our reconstruction of developments is usually based on sources which tend to be the exact opposite of this stylistic baseline, namely texts which are written, edited, standardized, and monologic. Ideally, a comprehensive history of a language should be based on spoken data which preserve the signal (medial orality) or, where this is not possible, closely approximate to informal conversational and/or vernacular usage, thus maximizing conceptual orality.

Sections 9.2 and 9.3 will briefly address some of the logistical challenges arising in the language-historical study of sound recordings. Section 9.4 will present one promising source of relevant data, namely the recordings made with the help of British prisoners of war during the First World War by the Königlich Preußische Phonographische Kommission between 1916 and 1918, with a view to assessing its potential for historical-linguistic research. Section 9.5 will refer to some relevant recent and ongoing research projects. This will be complemented by a brief conclusion and an appendix pointing the reader to a selection of resource centres.

9.2 Various types of data: potential and limitations

The year 1877 and, in practical terms, the early twentieth century represent barriers which cannot be pushed back when it comes to the preservation of the acoustic signal. However, this does not mean that historical-linguistic research based on audio data is disconnected from work on older stages of the language. As stated above, there is indirect evidence for the reconstruction of past pronunciations, and conceptually oral, speech-like, or speech-based genres allow us to study other aspects of spoken usage (lexicogrammar, discourse norms) ‘by proxy’, as it were. This is done, for example, in the Corpus of English Dialogues 1560–1760 (Culpeper and Kytö Reference Kytö, Jucker and Taavitsainen2010; for the guide to CED, see Kytö and Walker Reference Kytö and Walker2006). Culpeper and Kytö (Reference Culpeper and Kytö2010) present a thoughtful discussion of the ways in which such writing partially and variably approximates the structure of spontaneous speech. Among the methodologically intriguing issues they raise is the question of whether the best written representation of real speech is found in factual transcriptions such as the trial records of the Old Bailey Corpus (OBC, Huber Reference Huber, Meurman-Solin and Nurmi2007) or in the fictional evocations of speech in novels and plays (Mair Reference Mair, Andersen and Bech2013c; for these corpora, see also Chapter 8 by López-Couso in this volume). Putting the question in this way, of course, disregards another issue, namely the immense variability in the type and quality of dialect representation in literary works (see, e.g., Corrigan Reference Corrigan1996, McCafferty Reference McCafferty2005, or Fennell Reference Fennell2008 for examples from eighteenth- and nineteenth-century Scotland and Ireland).

Another avenue worth exploring is to sift the inexhaustible sands of standard writing for the nuggets of vernacular literacy. Thus, pauper letters and emigrant correspondence have been used in a variety of contexts to reconstruct the history of spoken English (Montgomery Reference Montgomery1995, García-Bermejo Giner and Montgomery Reference García-Bermejo Giner, Montgomery and Thomas1997, Fairman Reference Fairman, Barton and Hall2000, Boling Reference Boling, Smyth, Montgomery and Robinson2006). Self-evidently, none of these written data fully compensate for the absence of the phonetic dimension, as pronunciation and intonation are rendered imperfectly and unsystematically even where an effort is made to represent them through non-standard spellings. For example, the selective availability of conventionalized spellings for informal and vernacular items distorts representation, as does eye dialect as a purely visual index of non-standardness without any phonetic equivalent at all.Footnote ¹

At the risk of some simplification, Table 9.1 displays the potential and limitations of the various types of data available for the historical study of spoken English.

Note that of the six criteria in Table 9.1 it is only ‘physical realization’ that is binary, because it exclusively refers to medial orality. A text is either sound-recorded or written. If the recording is transcribed, the transcription's status is derived and secondary in relation to the spoken original. By the same token, a broadcast reading from a written text (such as Twain's Huckleberry Finn) is derived from – and hence secondary to – the written original.

The five remaining criteria refer to conceptual orality and are therefore gradient. A formal public speech is usually read aloud from a written text, which minimizes the scope for spontaneous improvisation. (Written and spoken) narratives are basically monologic but may contain stretches of reported or enacted dialogue, and so forth. Where such mixed and transitional constellations are particularly likely, this is represented by ‘–/+’ or ‘+/–’ in the relevant boxes in Table 9.1 (with the first symbol in the series indicating the more expected value). For example, the ‘–/+’ value for the Old Bailey texts in the column for ‘vernacular’ indicates that these data were systematically standardized and edited by the reporting clerks but still contain traces of the original spoken vernacular (see Chapter 18 by Beal in this volume; see also Archer Reference Archer, Nevalainen and Traugott2012a: 147–9).

As becomes apparent in the rightmost column of Table 9.1 (‘physical realization’), the phonetic dimension is absent by definition for all types of speech-like writing; we will never have audio recordings of the original trials documented in the Old Bailey Corpus, of nineteenth-century paupers reading their letters aloud, or of Samuel Clemens reading the text of Huckleberry Finn in the pronunciation he considered appropriate. As has been pointed out above, this need not be a problem for many types of investigation into the history of spoken and vernacular English. In the fictional evocation of a spoken text, the lexicon and grammar often preserve important aspects of the sociolinguistic profile of informal and vernacular speech of the time. For example, the choice between thou and you, between third-person singular present -eth or -s, or multiple negation can all be safely studied on the basis of such data (see, e.g., Nevalainen and Raumolin-Brunberg Reference Nevalainen and Raumolin-Brunberg2003 based on correspondence; for thou vs. you in trial proceedings, witness depositions and drama, see Walker Reference Walker2007; for -eth vs. -s in witness depositions, see Chapter 27 by Walker in this volume).

This is patently not so with the phonetic details which elude the English spelling system, and reliability and consistency are generally low even where conventionalized non-standard spellings exist, such as, for example, gonna, gimme, or lemme (for going to, give me, or let me). A contracted pronunciation (e.g. /gɪmɪ/) cannot be ruled out when we read <give me> in a transcription, nor is the presence of the non-standard spelling gimme in the transcription absolute proof that the form was pronounced as a contraction. Lay transcribers generally tend to over-standardize what they hear in the transcription, but they may also assign non-standard spellings to individuals they perceive or stereotype as non-standard speakers. To research the history of such phonetically reduced forms, transcripts are not enough. We need historical audio recordings – and preferably not the public oratory which is easiest to obtain for the first half of the twentieth century, but spontaneous dialogic data of the type illustrated by the Watergate Tapes.Footnote ²

9.3 Working with historical audio recordings: technical, legal, and ethical challenges

The nineteenth century saw a number of experimental attempts at recording voice and music (see Picker Reference Picker2003). As hinted at above, the very early period of sound recording produced a patchy legacy of data, with a quality which is generally too low to warrant linguistic analysis. What is conventionally considered the dawn of the era of mechanical recording of the human voice is the year 1877, when Thomas Alva Edison successfully recorded himself reciting the nursery rhyme ‘Mary Had A Little Lamb’ on his phonograph. The medium for storage being wax-coated cylinders, mass production and mass dissemination were not feasible at first, but became possible when Emil Berliner developed the gramophone, using shellac discs (or records) as the carrier medium from 1888. In the early days, their quality also fell short of what one would like to have for linguistic analysis, but from the early twentieth century, that is, for over a hundred years now, recordings of sufficient quantity and quality have accumulated to warrant investigation.

Until relatively recently few linguists would have been tempted to use old audio recordings as data for their research because of obvious technical obstacles. The development of recording technology has always been very fast, with older storage media becoming obsolete and inaccessible to anyone but specialists within a short while. Just as in the 1930s it was impossible for most linguists to work with 1890s wax cylinders, most present-day dialectologists lack the equipment needed to play 1960s reel tapes or even 1980s cassette tapes. More importantly, damage to the carrier meant the end of the recording. Some practical illustrations of this issue may be found in Allen Reference Allen, Beal, Corrigan and Moisl2007.

As a result, many potential language-historical treasures remained widely dispersed in libraries, archives, and collections – public and private, commercial and non-commercial – all around the world; constituting unexplored ‘hidden depths’, to cite the title of Widdowson (Reference Widdowson1999). This has only really changed recently in the course of the digital revolution, which has produced transportable formats such as ‘.mp3’Footnote ³ and facilitated access through the World Wide Web. This is illustrated very well by the material which will be discussed in section 9.4, which somewhat unexpectedly survived the Second World War and the subsequent Cold War in two Berlin archives, was largely forgotten and only rediscovered and digitized in the 1990s. Parts of it are now made available freely to the public by the British Library. For the first time since their collection, this has created a situation in which the data have become widely accessible to interested researchers again, and many legal issues surrounding access and use have been clarified, too. Readers requiring a comprehensive and detailed survey of the technical and legal challenges of archiving and curating audio data are referred to Linehan (Reference Linehan2001), a collection of essays written by experts in the field who exemplify the issues with a wide range of examples spanning the gamut from early recorded music via oral history and performance to wildlife recordings.

From the ethical point of view, working with historical recordings is generally less problematic than analysing more recently recorded spontaneous conversational data, which requires the subjects’ informed consent before the data are obtained and appropriate measures to protect their privacy when the data are published (see Cameron Reference Cameron2001: 19–25). Obviously, some of the historical data which we may want to analyse were not gathered in accordance with current ethical recommendations in linguistics. The Watergate Tapes are an obvious case in point. However, even in such cases the data will usually have been in the public domain for some time before becoming available for linguistic analysis, so that informant rights are rarely infringed on by linguists for the first and crucial time.

9.4 The recordings of the Königlich Preußische Phonographische Kommission: an early source for the study of real speech in real time?Footnote ⁴

9.4.4 Rhoticity

Ellis (Reference Ellis1890) attests rhoticity for most regions of England in the 1860s, and many traditionally rhotic areas were still described as mainly rhotic when the SED research was conducted in the 1950s. Britain, on the other hand, argues that ‘we can comfortably assume…that the actual area of loss was much greater and had penetrated further into the west and south-west [than can be inferred from SED data] simply because the data come (deliberately) from the most conservative speakers of the community – old rural nonmobile men – and exclude those who are likely to have pushed further the innovatory drive toward r-lessness’ (Britain Reference Britain2009: 130). Rhoticity is one of the variables frequent enough in the data to allow at least rudimentary quantitative analysis. For Norfolk, this yielded a very low rhoticity rate of 2.3 per cent (or 7 out of 302 instances),Footnote ⁹ which characterizes the dialect as a non-rhotic one in practical terms.

By contrast, the Lautarchiv data for the county of Kent, as solidly non-rhotic as Norfolk today, are more highly variable, with 66.7 per cent rhoticity for Sevenoaks and 38 per cent for Hunton. The informant for Sevenoaks was born in the town in 1883 and went to school locally until he became a polisher's apprentice in London. He returned to Sevenoaks, where he lived until the age of 20, and then went to India. The Hunton informant, also born in 1883, grew up in the village, where he became a blacksmith, probably in his father's business. At the age of 17 he left home to join the army for three years, and lived in London for another five after that. In his personal protocol form, Brandl quotes the speaker as saying: ‘If I set my mind to it, I can talk the language of the farmers.’ London was already described as non-rhotic by Ellis (Reference Ellis1890: 58) and we can assume that several years in London must have brought both informants into close contact with non-rhotic speakers.

Instrumental analysis of the data is tempting, but of course beset with a number of problems. Comparison across time requires that we be able to distinguish real change in the sound and distortions introduced by the different types of recording apparatus used at different times. When the original cylinders and shellac discs were digitized in the 1990s and early 2000s, the non-lossless .mp3-format (rather than, for example, the lossless .wav-format) was chosen as the standard, which may be another slight disadvantage in this case. In addition, the PRAAT response is likely to differ between the relatively noisy historical data studied here and present-day studio-quality input. Figure 9.1 presents the measurements of one Norfolk informant's long high back and stressed central vowels (lexical sets GOOSE and STRUT).

Figure 9.1 Measurements for vowels in two sons, produced by Norfolk informant (Fred Eccles), recorded in 1917

Even if comparison across time should be undertaken with caution in view of the caveats formulated above, comparison within the Lautarchiv corpus itself remains a possibility – helped by the highly standardized procedure of data gathering employed. As most informants were asked to narrate the Parable of the Prodigal Son, the phrase chosen here for illustration, two sons, occurs in most recordings. A considerable number of informants also recited the numerals, which establishes a convenient frame of comparison for potentially interesting diphthong variables (five, nine) or rhoticity (four, fourteen, forty), for example.

9.5 The historical linguistics of sound: ongoing projects and research perspectives

As has been mentioned above, some of the Lautarchiv recordings were transcribed, analysed and published in a series of twenty short pamphlets describing British dialects by Alois Brandl himself (Brandl Reference Brandl1926–7). The framework for presentation was a traditional-dialectological one, and the material mainly served the purpose of realistic illustration. In fact, Brandl's most interesting observations on the data are not found in these publications but rather in a personal memoir, where he writes:

This statement is vivid testimony to the power of recording technology, which instils awareness in the linguist of the fluidity and sociolinguistic variability of the vernacular. The fact that it is found in a personal memoir rather than one of the author's scholarly publications, on the other hand, shows that the time was not ripe for him to follow up on this observation in his research – or, in other words, to take the step from a traditional-dialectological to a sociolinguistic-variationist perspective in the study of language variation and change. As the material was largely forgotten later, the new perspective has never been systematically applied to it since.

However, there are several other projects which illustrate the potential of language-historical analyses supported by genuine audio recordings. Harrington et al. (Reference Harrington, Palethorpe and Watson2000, Reference Harrington, Palethorpe, Watson, Hardcastle and Beck2005) is a study of changes in British RP in the course of the second half of the twentieth century which became widely known even outside linguistics because of the high profile of its informant, Queen Elizabeth II. Comparing the Queen's Christmas Broadcasts from the 1950s and the 1980s, the authors show that some of the informant's vowels moved closer to variants associated with younger and less upper-class speakers. Radio news broadcasts produced in Australia during the same period were used by Price (Reference Price2008, Reference Price, Nevalainen and Traugott2012) to demonstrate the decolonization of Australian media language. British RP, the exclusive norm in the early period, gave way to a more local Australian accent. In addition to analysing archival broadcasts, the study profited from the fact that some of the announcers were still available for off-air interviews in their ‘real’ voices at the time the research was carried out.

Both these projects used media data and focused on standard pronunciation. This is different in research on early New Zealand English audio data which were discovered by Elizabeth Gordon and extensively analysed by herself, Jennifer Hay, Peter Trudgill, and others (see Trudgill et al. Reference Trudgill, Gordon, Lewis and Maclagan2000, Trudgill et al. Reference Trudgill, Maclagan and Lewis2003, Gordon et al. Reference Gordon, Campbell, Hay, Maclagan, Sudbury and Trudgill2004). The recordings were produced by the New Zealand Broadcasting Service's ‘Mobile Unit’, who travelled the country from 1946 to 1948 to record interviews with around 300 mostly elderly informants.Footnote ¹⁰ Beyond recording individuals and, possibly, charting diachronic changes in their pronunciation, this broad and heterogeneous demographic base makes it possible to move beyond the idiolect and explore the complexities of variety genesis.

A pioneering vernacular resource with a similarly broad base, though less remote in time, was NECTE, the Newcastle Electronic Corpus of Tyneside English, which brought together two collections of recordings and integrated them into a state-of-the-art digital corpus. The data comprised the Tyneside Linguistic Survey, a sociolinguistic documentation carried out under the direction of Barbara Strang in the late 1960s and early 1970s, and the PVC (Phonological Variation and Change) corpus compiled by James Milroy and co-workers from 1991 to 1994 (see the NECTE website for further details). NECTE was subsequently incorporated into the Diachronic Electronic Corpus of Tyneside English (DECTE), which has added contemporary data in order to produce a monitor corpus for the study of ongoing changes in their historical context (see http://research.ncl.ac.uk/decte/pvc.htm and Corrigan Reference Corrigan, Nevalainen and Traugott2012 for a case study on changes in the GOAT vowel).

Historic early blues recordings (Miethaner Reference Miethaner2000) and ex-slave narratives (Schneider Reference Schneider1989, Poplack and Tagliamonte Reference Poplack and Tagliamonte2001) have helped the study of older African-American English. In fact, a forthcoming collection of studies bearing the significant title Listening to the Past (Hickey forthcoming) shows that there is virtually no major region of the English-speaking world for which hitherto unexplored historical audio recordings are unavailable. As the editor's project website informs, the book aims to look at:

England will be covered by four case studies devoted to Tyneside, Liverpool, London Cockney, and RP, Ireland and Scotland by one each (Dublin and Glasgow). Five contributions will deal with the US, two with Canada, and one each with Jamaica, West Africa, South Africa, Australia, New Zealand, and Tristan da Cunha.

9.6 Conclusion

Historical sound recordings of the human voice have accumulated for more than a century – and in quantities which, for a language such as English at least, constitutes a potentially rich and valuable store of data. Over the past two decades, significant technological advances in the digital conversion, storage, and dissemination of the data have brought them into the reach of the ordinary linguist. As the present chapter has demonstrated, this has led to the beginnings of a boom in research which was to be expected. The most immediate area to benefit from this boom is historical phonology, both of standard and non-standard speech. Careful analysis of historical sound recordings will enable researchers to check existing assumptions for accuracy on new data, to achieve a more precise chronology of known developments and, at least in a small number of instances, will also force unexpected corrections. The cursory analysis of the Berliner Lautarchiv English dialect data carried out for purposes of demonstration here, for example, has shown that the dialect landscape of the late nineteenth and early twentieth centuries presented in sources such as Ellis or the SED is certainly not wrong but involves a significant degree of abstraction from the full extent of variability which – then as now – resulted from complex interaction of regional with social and stylistic variation.

As most early recordings do not provide spontaneous speech (or where they do restrict themselves to certain textual genres such as the folk anecdote or the oral history narrative) the recordings are less likely to benefit the study of historical morphosyntax. Here future generations of historical linguists will still have to fall back on speech-like and speech-based written genres as the most likely source to provide ‘speech by proxy’. In research on recent and ongoing processes of grammaticalization, the one-hundred-year time window opened up by audio data may allow us to trace phonetic reduction of highly frequent chunks ([have] got to → gotta, etc.) in real time. Here as elsewhere, however, in view of the time typically taken by grammaticalization and similar changes: if we want the full picture, we shall have to look back in time further than the ‘watershed’ year of 1877, when Edison's invention made it possible to conserve the evanescent sound of the human voice for posterity.

10.1 Introduction

The grammatical tradition in the Modern English period (1500–1900) has become a thriving topic in historical linguistics, albeit after a period of neglect due to the negative associations of eighteenth-century grammar books and the ‘intimidating’ amount of nineteenth-century material to explore (see Görlach Reference Görlach1999a: 7–8, Beal et al. Reference Beal, Hodson and Fitzmaurice2012: 201–2). The 1970s and 1980s witnessed a change of outlook, with scholars such as Leitner calling for ‘more teaching and research into “grammaticology”, the cross-disciplinary study of grammars’, which, as he put it, is ‘more than desirable’ for the understanding of the history of linguistics (1986: 1334). Pioneering work dates from the early twentieth century, with Kennedy's (Reference Kennedy1927) bibliography, and Leonard's (Reference Leonard1929) and Poldauf's (Reference Poldauf1948) studies. Bibliographic resources and access to evidence improved in the 1960s with Alston (Reference Alston1965) and the English Linguistics facsimile series (Alston Reference Alston1967–73). Michael's (Reference Michael1970, Reference Michael1987) and Vorlat's (Reference Vorlat1975) monographs are arguably the most comprehensive studies of the development of linguistic thought during the Early and Late Modern English periods. On a smaller scale, early studies focused on specific grammar writers such as Robert Lowth (Pullum Reference Pullum1974) and William Ward (Subbiondo Reference Subbiondo1975); or on specific linguistic features like double negation (Tieken-Boon van Ostade Reference Tieken-Boon van Ostade1982), phrasal verbs (Hiltunen Reference Hiltunen1983), and auxiliary do (Tieken-Boon van Ostade Reference Tieken-Boon van Ostade1987). In the 1990s Sundby et al.'s Dictionary of English Normative Grammar (Reference Sundby, Bjørge and Haugland1991) contributed an exhaustive analysis of the proscriptions laid down in two hundred eighteenth-century normative works, and a number of case studies on literary authors and social networks followed (Tieken-Boon van Ostade Reference Tieken-Boon van Ostade1991, Wright Reference Wright, Fernández, Fuster and Calvo1994, Percy Reference Percy and Britton1996). The account of normative linguistics in Tieken-Boon van Ostade (Reference Tieken-Boon van Ostade, Auroux, Koerner, Niederehe and Versteegh2000a) testifies to the development of this field. Easier access via online resources (Early English Books Online, Eighteenth Century Collections Online, the Eighteenth-Century English Grammars database)Footnote ¹ allowed for a change in research perspective and gave the field a boost. Scholars turned to the original materials and to the grammar writers and demonstrated that historical grammars are a source of evidence in their own right. A number of dissertations and monographs have investigated linguistic features such as past and past participle forms (Oldireva Gustafsson Reference Oldireva Gustafsson2002), the subjunctive (Auer Reference Auer2009), and preposition stranding (Yáñez-Bouza Reference Yáñez-Bouza2015a); others have focused on the grammar writers themselves, including Ann Fisher (Rodríguez-Gil Reference Rodríguez-Gil2002), Joseph Priestley (Straaijer Reference Straaijer2011), and Robert Lowth (Tieken-Boon van Ostade Reference Tieken-Boon van Ostade2011). The state of the art on several topics in the field is illustrated in edited volumes (Beal et al. Reference Beal, Nocera and Sturiale2008, Tieken-Boon van Ostade Reference Tieken-Boon van Ostade2008b, Hickey Reference Hickey2010, Yáñez-Bouza and Rodríguez-Gil Reference Yáñez-Bouza and Rodríguez-Gil2013b), conference series (Perspectives on Prescriptivism 2003, 2006, 2009, 2013), and large-scale projects (The Codifiers and the English Language, Leiden University). While the focus has fallen primarily on the seventeenth and eighteenth centuries, the nineteenth century remains largely underexplored, although interest is now growing here too (Michael Reference Michael1997, Görlach Reference Görlach1998a, Anderwald Reference Anderwald2012, the Collection of Nineteenth-Century Grammars).

This chapter aims to illustrate how grammars from the early and late Modern English periods have become a serious object of study in the field of English historical linguistics, not only as sources for the norms of present-day written standard English, but also as evidence of language use, variation, and change. First, their production is contextualized in relation to the process of standardization of English and the descriptivism–prescriptivism continuum (section 10.2). This is followed by a survey of the main approaches in the study of historical grammars, which will highlight the strands to which historical grammaticology can contribute and the various ways in which grammars can be examined. Section 10.3 reports on the influence of precept on usage, the interference of precept in change related to linguistic and sociolinguistic factors, and the descriptive adequacy of grammarians and their grammars. Section 10.4 touches on the potential of less researched areas such as subsidiary content in grammars, grammars attached to other works, and the paratext of grammar books.

10.2 English grammar writing

The production of grammars of English increased dramatically during the four centuries of the Modern English period, though not at a steady pace, as shown in Figure 10.1.Footnote ² The history of English grammar writing begins with William Bullokar's Bref Grammar for English in 1586, one of only two grammars known to have appeared in the late sixteenth century; by the end of the seventeenth century only twenty-one ‘explicitly English grammars’ were in existence (Michael Reference Michael1970: 151). Production in the early eighteenth century is still modest, with fewer than forty new grammars. This is in stark contrast to the rapid growth during the second half of the century, which yielded over 200 works and over 1,600 printings, with remarkable peaks in the 1770s and 1790s. This itself was but a preamble to the ‘hyperactive production’ during the nineteenth century, when an average of eight or nine new works were produced each year: approximately 860 new grammars and over 3,600 printings overall, with salient peaks during the 1840s and 1870–1880s (Michael Reference Michael1997).

Figure 10.1 Production of English grammars per decade from 1586 to 1900

The development of English grammar writing can be said to run parallel to the process of standardization and the changing attitudes towards Latin (for standardization, see Chapter 18 by Beal in this volume). Little attention was given to the grammar of English in the early Modern period, primarily because the term grammar conveyed notions of Latin grammar until well into the eighteenth century (Yáñez-Bouza Reference Yáñez-Bouza2015b). Latin was the only model that early grammarians had to hand, and this was reflected in the approach to the structure of English (trying to fit English into Latin) and in teaching methods (exercises of bad English, parsing exercises, etc.). Following Nevalainen and Tieken-Boon van Ostade's (Reference Nevalainen, Tieken-Boon van Ostade, Hogg and Denison2006: 274–87) account of standardization, we can argue that the need for English grammar writing during the sixteenth and seventeenth centuries responded to attempts to elaborate the functions of the English language against the dominance of Latin, most commonly in vocabulary, spelling, and style, but also in relation to grammar. The beginning of the grammatical tradition was a timely response to the cultural climate created by the Renaissance, the Reformation, and the Humanist movement, which together favoured the emancipation of English from Latin culture and raised the prestige of English in schools (see Vorlat Reference Vorlat1975: 3–6). John Wallis's (Reference Wallis1653) Grammatica Linguæ Anglicanæ marked a new era of interest in the vernacular by pointing out the structural differences between English and Latin, paving the way for a conscious movement of reform ‘which sought to give English a grammar in its own right’ (Michael Reference Michael1970: 210). In the eighteenth century, Beal (Reference Beal2004: 101) notes that the increasing interest in vernacular education was ‘[p]erhaps the most pressing and practical reason for the production of grammars’. The eighteenth century also witnessed the last two stages of the standardization process: codification and prescription. Codification is the phase when rules are laid down in grammars, dictionaries, and other sorts of self-education handbooks regarded by their readers as authoritative manuals; once the language is codified, the prescription stage involves the implementation of the codified norms in written language (see Nevalainen and Tieken-Boon van Ostade Reference Nevalainen, Tieken-Boon van Ostade, Hogg and Denison2006: 282–4). The two phases are said to coexist from around 1745 to the 1770s (Tieken-Boon van Ostade Reference Tieken-Boon van Ostade and van Ostade2008a: 7–8), after which there was a spate of practical and utilitarian grammars targeting new and wider audiences. The shadow of Latin persisted during the eighteenth and nineteenth centuries, especially in the description of parts of speech, but the ‘irregularities’ of the Latin–English mapping were ‘repeatedly denounced’ by nineteenth-century grammarians, and Latin gradually lost ground (Dekeyser Reference Dekeyser1975: 28).

The approach to grammar writing has often been presented as a binary opposition: those who describe actual usage versus those who evaluate usage and prescribe recommendations about how language ought to be used. But we should rather see attitudes ranging along a descriptive–prescriptive continuum. Vorlat suggests a threefold distinction:

In the sixteenth and seventeenth centuries the normative approach prevails, leaning more or less towards the descriptive. The prescriptive approach is first made explicit in Christopher Cooper's Grammatica Linguæ Anglicanæ (Reference Cooper1685), but at that time prescriptivism ‘is in the making, not yet made’ (Vorlat Reference Vorlat1979: 137). The shift from normative to prescriptive grammar becomes apparent towards the mid-eighteenth century when incorrect and improper usage was given more attention and the ideal of correctness led to the practice of imposing norms on language users; it was the beginning of the heyday of the prescriptive era, which, as Vorlat (Reference Vorlat and Schmitter2007: 512) puts it, ‘grew into a mania’ during the nineteenth century. As Bailey notes, attitudes towards grammatical correctness and propriety ‘hardened into ideology’ (Reference Bailey1996: 215) during this century, particularly in the latter half, when it was commonly felt that ‘everyone ought to consult’ grammars and dictionaries (1996: 3). That said, the nineteenth century can be better described as ‘a continual alternating between descriptivism and prescriptivism’ (Dekeyser Reference Dekeyser1975: 4), when the dominance of the latter approach came up against the influence of the New Philology. Linguistic thought had come to be occupied with speculative theories of language discussed in universal grammars. A major influence here was John Horne Tooke's theory of etymological derivations based on the relation of language and mind, a priori reasoning, and the belief that there is only one true primitive meaning for each word (Diversions of Purley, 1786). It is only in the 1830s and 1840s that Benjamin Thorpe (1782–1870) and John Mitchell Kemble (1807–57) introduced from the continent the empirical and a posteriori methods that would be followed by the founders of the Oxford English Dictionary in its incipient stages during the 1850s and 1860s, as well as by historical grammarians such as Robert Gordon Latham in An Elementary English Grammar for the Use of Schools (1843) and Henry Sweet in A New English Grammar (1891–8). The latter half of the nineteenth century then witnessed ‘the transition from old to new, from amateurism to linguistic professionalism’ grounded in a scientific approach to analysis and description (Dekeyser Reference Dekeyser1975: 24).

10.3 Approaches and findings

This section surveys approaches taken in the literature and summarizes the main findings. For reasons of space, only a small number of case studies can be discussed; the Appendix to this chapter complements this account with a more complete list of linguistic features.

10.4 The study of historical grammars beyond morphosyntax

This section draws attention to some under-explored aspects of historical grammars in the belief that further research here can shed light on the understanding of the term ‘grammar’ as well as on the growing importance of grammar knowledge.

10.5 Concluding notes

This chapter has shown that early grammars can help to account for change in the frequency of a particular construction and for the loss of linguistic variation, as not all linguistic changes result from internal forces – changes from below; some come from external, prescriptive norms – changes from above. Sometimes grammatical precepts triggered permanent change, at times they only resulted in temporary change, and on occasions they retarded natural developments; very often they reinforced an existing trend, and in this sense grammars can be studied as evidence of contemporary usage. The study of grammarians’ evaluative comments has the added value of providing insights into sociocultural attitudes towards language in earlier times, such as regarding the social and dialect status of linguistic features. This chapter has also shown that the study of grammars can shed light on aspects beyond morphosyntax, including phonology, rhetoric, spelling, and punctuation. Attention has also been drawn to underexplored areas which have the potential to contribute further to the history of grammar writing and historical linguistics, such as subsidiary grammars and the paratext of grammar books.

Of course, one must also bear certain pitfalls in mind. For instance, historical grammars, as part of the standardization process of written English, were aimed at written language, with the implication that colloquial and dialectal features were often criticized or indeed not recorded. Descriptive adequacy needs to be examined cautiously: grammarians might be discussing their own usage or the usage of the social class they aspired to, and not necessarily common, widespread usage. There might also be a time lag between the time a new use emerges or dies out and the time this is recorded in grammar books, for it naturally takes a few decades for change to be noticed, especially a change which develops slowly. Another important point is the difficulty of assessing the extent to which grammars were actually read, in that education was for the most part a privilege that few could afford, and the kind of schools that the lower and lower middle classes had access to would not always have had the resources to buy many books. Thus, as mentioned in relation to William Clift, there are indications, although not always direct evidence, that certain authors read grammars. For these reasons, the analysis of evidence from grammars is better when complemented with the study of usage corpora.

Historical grammars from the Early and Late Modern English periods have gradually received more attention, and their use as evidence for the study of historical linguistics has slowly but solidly come to be recognized in the disciplines of historical sociolinguistics and normative linguistics. With due caution, then, the study of historical grammars can help to explain language use in the past, and today, can shed light on early and modern conceptualization of linguistic features, and can help us trace language change and (the loss of) variation.

11.1 Introduction

This chapter addresses the question of how historical linguists extract data from the material examined. Material here refers to the actual evidence of language use that we have at our disposal (typically spoken or written texts). The researcher's data comprise the relevant observations on which the empirical analysis is based, and which are extracted from the material. By way of exemplification, let us assume that the research question concerns the substitution of the progressive passive (e.g. the house is being built) for the so-called passival construction, which is active in form but passive in meaning (e.g. the house is building). In this case, the most important data are likely to be attestations of these two constructions as they occur in texts. Those data may allow the researcher to construct a quantitative picture of the change and show how the progressive passive accounts for an increasing proportion of the data over time. If there is material available from different categories of language users (e.g. men and women from different socio-economic groups) or texts (e.g. speech-related and expository genres), the quantitative picture can be refined by the addition of such extralinguistic parameters to the analysis (see section 11.3.1). The linguistic context of the data may also provide important information (see section 11.2.3); for instance, it may be that the progressive passive first appeared in certain types of verb phrase and then spread to other parts of the paradigm.

In addition to examples of the linguistic constructions themselves, important clues may be available in contemporary comments on usage. In the case of the progressive passive, such comments demonstrate that this incoming construction met with near-universal condemnation in normative sources (see, e.g., Anderwald Reference Anderwald2012: 36–9); such opposition may constrain the spread of an innovative feature in at least some genres. I will focus on data that comprise attestations of linguistic features in this chapter, but it should be noted that, in fields such as historical phonology, contemporary comments may be the most important type of data (see section 11.3.2). These two types of data can also complement each other. Dekeyser (Reference Dekeyser1975) investigates number and case relations in nineteenth-century English (e.g. the choice between possessive and objective pronouns as subjects in gerundial clauses like his/him winning the contest in I was surprised by his/him winning the contest). By considering both actual usage and commentary, Dekeyser succeeds in presenting a many-faceted picture of linguistic variation during the 1800s.

The rest of this chapter is structured as follows. Section 11.2 is concerned with general issues relevant to data extraction from historical material, with a focus on electronic corpora; while the account focuses on historical English morphosyntax, large parts of it will also be applicable to other sub-disciplines, such as historical pragmatics and sociolinguistics. Most of the examples that I use to illustrate the account will be taken from Early and Late Modern English. Section 11.3 addresses some special challenges that historical linguists face as regards the nature of the material, phonological analysis, spelling, and corpus annotation. The chapter ends with a brief concluding discussion of possible future developments in the field in section 11.4.

11.2 The identification of data in English historical linguistics

11.3 Challenges for historical linguists

Although most of the example studies chosen in section 11.2 were historical, much of the discussion would be equally valid for an investigation of Present-day English. In this section, in contrast, I address some challenges that are arguably characteristic especially of historical linguistic research: the special nature of historical material, implications for historical phonology, the variable spelling that characterizes many historical texts, and problems involved in compiling and annotating historical corpora.

11.4 Trends and possible future developments

As the present chapter has demonstrated, the textual basis that linguists have at their disposal for the study of the history of English is a source of challenges as well as opportunities. In addition to fulfilling all the methodological requirements of a synchronic study, the researcher must also gain familiarity with the special demands peculiar to the period(s) studied. However, some of the difficulties facing historical linguists also open up possibilities that are not easily available to synchronic researchers; for instance, the lack of a fixed spelling, which creates problems for automatic retrieval using concordancers, also enables historical phonologists to study past speech through the written medium.

As regards possible future developments in data collection in English historical linguistics, two separate but complementary trends can be mentioned. First, an ever increasing range of original documents are being made available electronically. Several digital archives make it possible to access large numbers of primary sources online. These sources can of course be accessed directly as part of a manual process of data collection. However, partly by drawing on such electronic archives, corpus compilers are also assembling increasingly extensive corpora that enable automatic retrieval procedures. Especially for the Early and Late Modern English periods, large numbers of texts are becoming available for corpus-based research for the first time. Techniques like Optical Character Recognition (OCR) are likely to become increasingly important in converting documents to corpus texts.

At the same time, however, the early twenty-first century has witnessed increased interest in returning to the original manuscripts of older texts. Several researchers have noted that there is a need for editions that ‘aim at reproducing the original manuscripts faithfully, and avoid normalizing, modernizing, or otherwise emending the original manuscript texts’ (Kytö et al. Reference Kytö, Walker and Grund2007: 66). For instance, Grund (Reference Grund2006: 106) notes that an edition which incorporates ‘readings from several different manuscript witnesses into the text of the edition’ presents ‘an eclectic or hybrid text for which there is no actual historical witness’. Grund (Reference Grund2006: 119) also raises the issue of whether texts which exist in several witnesses that differ linguistically ‘should be considered manifestations of the same text or independent texts’ in corpus compilation.

Ideally, these two trends may be combined in electronic editions of texts (for examples, see Chapter 7 by Horobin in this volume). Electronic editions can incorporate linguistic variation among witnesses, yet still remain faithful to each individual witness. In addition, provided that the edition comes together with (or is compatible with) concordancers and other forms of corpus-linguistic software, some versions of the texts can be made available in searchable form, with different degrees of normalization; for instance, while one searchable version of the text can be rendered with normalized spelling that would facilitate tagging, retrieval, etc., another can preserve the original spelling of the text. A clear desideratum in this regard would be the development of common standards as regards the software and editorial conventions used for such electronic editions, which would ultimately enable researchers to create their own computerized corpora by combining texts from different electronic editions, each of which has been compiled by – and for – linguists. While the large amount of funding and time required to produce such editions is likely to impose limitations on development, electronic editions of this kind may ultimately combine philological and corpus-linguistic perspectives in a way that will provide researchers with the best of both worlds in terms of data retrieval.