Studies in the History of the English Language VII: Generalizing vs. Particularizing Methodologies in Historical Linguistic Analysis 9783110494235, 9783110494501

Citation preview

Don Chapman, Colette Moore, Miranda Wilcox (Eds.) Studies in the History of the English Language VII

Topics in English Linguistics

Editors Elizabeth Closs Traugott Bernd Kortmann

Volume 94

Studies in the History of the English Language VII Generalizing vs. Particularizing Methodologies in Historical Linguistic Analysis

Edited by Don Chapman Colette Moore Miranda Wilcox

ISBN 978-3-11-049450-1 e-ISBN (PDF) 978-3-11-049423-5 e-ISBN (EPUB) 978-3-11-049174-6 ISSN 1434-3452 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. 6 2016 Walter de Gruyter GmbH, Berlin/Boston Cover image: Brian Stablyk/Photographer’s Choice RF/Getty Images Typesetting: RoyalStandard, Hong Kong Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of contents Colette Moore and Don Chapman 1 Introduction

I

Particularizing and generalizing for written records

R. D. Fulk A philological tour of HEL

11

Donka Minkova From stop-fricative clusters to contour segments in Old English Stefan Dollinger On the regrettable dichotomy between philology and linguistics: Historical lexicography and historical linguistics as test cases

29

61

II Particulars of authorship Seth Lerer The history of the English language and the history of English literature

93

Xingzhong Li “Of harmes two, the lesse is for to chese”: An integrated OT-Maxent approach 107 to syntactic inversions in Chaucer’s verse Mark Davies and Don Chapman The effect of representativeness and size in historical corpora: An empirical 131 study of changes in lexical frequency

III Particulars of communicative setting Peter J. Grund Seeing is believing: Evidentiality and direct visual perception verbs in 153 Early Modern English witness depositions

vi

Table of contents

Susan Fitzmaurice Sincerity and the moral reanalysis of politeness in Late Modern English: 173 Semantic change and contingent polysemy Marina Dossena Something to write home about: Social-network maintenance in the 203 correspondence of nineteenth-century Scottish emigrants

IV Particularizing from words Betty S. Phillips Words swimming in sound change

225

John G. Newman Plural marking in the Old and Middle English -nd stems feond 239 and freond Lynn D. Sims From Shakespeare to Present-Day American English: The survival of 263 ‘get + (XP) + gone’ constructions Index

283

Colette Moore and Don Chapman

Introduction 1 Overview The opposition between generalizing and particularizing methodologies is an enduring tension in examining source material in English language history. Indeed, it is a methodological dilemma in any field of study: should one focus on the large or the small? Does investigating the forest help us to understand the trees or does examining the trees aid in our models of the forest? Surely, as we know, it is both. Scholarship recognizes divisions between these kinds of study, the micro and macro, and acknowledges that there are important critical methods for zooming in and zooming out on the object of investigation. This volume incorporates different takes on these questions of scale as they relate to the historical study of the English language. In language study, the types of data used and the assumptions that motivate their collection and interpretation determine the contours of the analyses. To examine methodology in linguistic research, then, is to pursue the opportunities and the limitations posed by the different subfields and investigational techniques of linguistics (Coffin, Lillis, and O’Halloran 2010; Dörnyei 2009; Ender, Leeman, and Wälchi 2012; Johnson 2008; Litosseliti 2010). For inquiry in the history of the English language, these methodological questions take slightly differing forms (articulated in Bergs and Brinton 2012: 1421–1657; Biber and Reppen 2015; Busse 2012; Nevalainen and Traugott 2012; Taavitsainen and Fitzmaurice 2007). Sample methodological questions include: how should historical English research integrate statistical models? What kinds of evidence are necessary for diachronic conclusions? How should linguistic evidence be contextualized through the cultural and material features of text production? This volume continues the conversation on the methodological dynamics in examining the historical record of English: methodologies of the general and the particular. Generalizing perspectives permit linguistic features to be removed from context and to inform collective conclusions; these include lexicography, corpus studies, and theoretical linguistics. Particularizing approaches, on the other hand, underscore the defining role of context in considering these linguistic features and include philological perspectives, historical pragmatics, discourse analysis, and certain kinds of grammatical and lexical approaches to language change.

2

Colette Moore and Don Chapman

One common aspect to the chapters of the volume, then, lies in their attention to evidence: the ways that research in historical English language study must recognize the realities of the written record of early English. In studying historical stages of the language, we obviously depend upon this written record, and this dependence forces upon us an awareness of the contingencies involved in working with our sources, which becomes part of the craft of doing historical linguistic research. This volume pursues the scope that those contingencies can take, ranging from the very close philological concerns of examining a single word or construction in a single manuscript to the larger abstract ways of creating and using corpora that seek to represent their consolidated written records while allowing for statistical analysis. In all investigations, one sees a tension between the particularizing of the philological method and the generalizing of linguistic claims: this volume explores the tension between these poles. While the constraints of the written record restrict our view of the past and the conclusions that we can draw about its language, then, they also force us to be aware of the particularities of language in ways that are fruitful for scholarship. In fact, even in techniques that would seem to be generalizing in orientation, such as corpus linguistics, researchers use particularizing techniques to build focused corpora and to focus on contextual questions. The articles that follow propose approaches that combine questions of the particular with generalizations about language change: blending the large and the small, the philological with the linguistic.

2 The themes of the volume The sections of the volume address from different angles the tension between these kinds of particularizing and generalizing impulses in historical linguistic analysis. The questions that each chapter invokes share an attention to scope – how much can be abstracted from the written record without losing information that is important to the analysis? Or posed another way, in what ways are our arguments contingent on the construction of the written record?

2.1 Particularizing and generalizing for written records The first section of the volume centers the conversation by defining and addressing philological approaches and how they might intersect with or build upon generalizing quantitative tools like searchable dictionaries and corpora.

Introduction

3

R. D. Fulk outlines the issues of this volume with “A philological tour of HEL”, which sets out the philological component of studying language change: “the study of the extralinguistic contexts of linguistic data or the relation between contexts and data”. Fulk presents the recent history of how philology has been perceived and points out crucial ways that philology needs to be considered to make data meaningful for other linguistic methods. He argues that the written records themselves contain textual information that has not always been considered significant (e.g. capitalization, insertions), but that can be salient to interpreting the language that we examine in historical linguistics. This is a crucial reminder of a tension that is always present, but not always explicit in our analysis. In “From stop-fricative clusters to contour segments in Old English”, Donka Minkova exemplifies the close, philological work that Fulk describes: she discusses the development of affricates in the English consonantal system, using metrics and spelling from many sources to tease out what the written record suggests about the evolution and presentation of consonant sounds. Her inquiry shows that the evolution of affricates influences how we might understand constrastiveness and distinctiveness in phonological reanalysis. Stefan Dollinger’s “On the regrettable dichotomy between philology and linguistics: Historical lexicography and historical linguistics as test cases” concludes this section by investigating the relationship between philology and linguistics in historical lexicography. He incorporates a distinction between the work of “literary philologists” and “linguistic philologists” and posits that lexicography has been largely built on the work of literary philologists. It is not surprising, then, that he calls for more of the systematic, generalizing work typical of linguistics. Yet Dollinger also calls for more philology as well, offering an example of a dictionary entry that would have been improved had it noted the publication site of the citation. He argues that historical dictionaries would benefit from more careful attention to both philology and linguistics, and that the close, particularizing work of dictionaries provides useful data for other, more generalizing approaches.

2.2 Particulars of authorship The second section of the volume investigates particularizing insights as they emerge from studies of authors and authorship. These chapters illustrate and explore the tension between the generalizing of linguistics and the particularizing of philology with regard to individual authors and their purposes. Studies of individual authors, we see, may yet make more general statements about the

4

Colette Moore and Don Chapman

history of English, and general statements will always depend upon the assemblage of these individuals. Seth Lerer’s “The history of the English language and the history of English literature” directly addresses the importance of literary authors in the analysis of historical records. His view of philology complements the linguistic/philology divide present in Fulk’s and Dollinger’s chapters. Lerer argues that writers, aware of the value of innovation and aware of their own use of language, introduce innovations that make their way into general language to varying degrees. Here he is not simply championing “literary philology” on its own terms, but rather pointing out its value for linguistic investigations into linguistic change, particularly for those changes that might go unnoticed because they are assumed to be literary and therefore somehow not “linguistics”. Writers use language change as part of their repertoire and can serve as agents of linguistic change. Further, our models of literary authorship construct and depend upon notions of linguistic innovation. The second chapter in this section, “‘Of harmes two, the lesse is for to chese’: An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse”, by Xingzhong Li, illustrates the tension between the particularizing of philology and the generalizing of linguistics in its investigation of Chaucer’s metrics. Xingzhong Li’s analysis uses optimality theory and maxent grammars as a methodology to analyze syntactic inversion in Chaucer’s verse, building different metrical, phonological, prosodic, and semantic constraints for the prevalence of inversion. This emphasis on the particulars of authorial choices in verse means that his analysis is partly philological in orientation, even as its use of Optimality Theory invokes generalizing theoretical methods. It is through analysis of the work of a single author that the force of the stylistic choices can be felt. Finally, Mark Davies and Don Chapman push at the question of the importance of authorial context from the other direction in their examination of largecorpus study: “The effect of representativeness and size in historical corpora: An empirical study of changes in lexical frequency.” Corpora necessarily abstract away most particulars of the written record, but historical corpora usually try to record and control for the most important contexts of the written record, such as date of composition, genre of composition, social demographics of the writer, purpose of the writers, and so on. Davies’s and Chapman’s investigation, however, asks whether the principle of representativeness in aspects of the context might be a factor of corpus size; their investigation of lexical frequency over time suggests that a massive corpus like Google Books might not need to be representative in order to be useful as evidence for lexical change. As such, it presents the opposite end of the cline for considering the defining importance of authorship: Google Books offers not fine-grained attention to authors, but

Introduction

5

rather the sweeping perspective that becomes possible through authorial amalgamation and the large-scale data mining that that permits.

2.3 Particulars of communicative setting These notions of philology – that it establishes data for linguistic investigations or accounts for communicative need – are visible in methodologies associated with historical sociopragmatics. Such methodologies illustrate philology’s benefit to linguistics in another way that Fulk lists: the pragmaphilology which “describes the contextual aspects of historical texts, including the addressers and addressees, their social and personal relationship, the physical and social setting of text and production and text reception, and the goal(s) of the text” (Jacobs and Jucker 1995: 11). The chapters in this section all address different kinds of communicative context: the pragmaphilological context of genre and setting, the sociolinguistic context of social networks, and the cultural context of societal pressures and changing values. The material, cultural, sociolinguistic, and pragmatic environments for written texts influence the kinds of questions that researchers can ask of their records and the kinds of answers that they find. Peter J. Grund’s “Seeing is believing: Evidentiality and direct visual perception verbs in Early Modern English witness depositions” exemplifies the importance of communicative purpose: it uses a corpus of witness depositions to investigate the use of direct visual perception verbs like see, observe, perceive as markers of evidentiality in English. The pragmaphilological approach finds visual perception verbs to be a pragmatic strategy for marking certainty, emphasis, and credibility. The study indicates that generalizing methods like corpus research are helpfully complemented by particularizing considerations such as pragmatic force in investigations of language variation and change. Susan Fitzmaurice’s “Sincerity and the moral reanalysis of politeness in Late Modern English: Semantic change and contingent polysemy”, examines communicative need in its cultural setting; she argues that some terms are subject to competing cultural and political pressures that define, reshape, and reanalyze words. Fitzmaurice’s discussion of the cultural place of politeness in eighteenth-century England examines the semantic evolution of the word politeness and posits a “contingent polysemy” of varying senses in the use of the term that is influenced by shifting relations to notions of sincerity. This portrait of individuals and social groups shaping words to fit their sociopolitical ends provides an intriguing complement to authorial-driven language change (discussed, for example, by Lerer). “Something to write home about: Social-network maintenance in the correspondence of nineteenth-century Scottish emigrants” is Marina Dossena’s investigation of the communicative setting in emigrant letters. Her study examines

6

Colette Moore and Don Chapman

pragmatic variation in politeness, addressing terms, endearments, description of shared memories, and promises of future visits; it considers these positive faceenhancing methods as reinforcing and maintaining social network connections between old world and new. The influence of external events can be seen in the focused analysis of a particular text type, and changes in style in the correspondence reflect the communicative needs of emigrants.

2.4 Particularizing from words Particularizing can come in the form of extra-textual context, as in the previous section, or it can come in the form of closer analysis of the elements of textual composition. Previous chapters have shown changes to the words themselves (e.g. Fitzmaurice polite, Dollinger ice, Lerer man); the contributions in this section note the importance of words in changes that go beyond the vocabulary. The role of individual words in language change has been increasingly acknowledged and investigated in such approaches as lexical diffusion and grammaticalization; words, we see, are critical segments that influence the evolution of the phonological and syntactic structure of a language. Betty S. Phillips’s chapter, “Words swimming in sound change”, invokes a particularizing context for sound change: that vowel shifts always occur in the context of words and, conversely, that lexical diffusion always occurs within phonological environments. Examining several vowel examples, the research establishes the primary role of frequency and the individual to sound change. Her argument for lexical diffusion revisits an old question left from the shift from nineteenth-century philology to twentieth-century linguistics, namely whether sound change occurred all at once (the neo-grammarian hypothesis) or whether it proceeded through the lexicon. In “Plural marking in the Old and Middle English -nd stems feond and freond”, John G. Newman also looks at lexical diffusion (or diffusion by wordclass) but for morphological rather than phonological change. His chapter examines token frequencies of the -nd Old English nouns and finds analogical diffusion for the adoption of -s as the plural marker. The plural -s did not diffuse in a uniform chronological way: both feond and freond first showed -s plural marking in Old English, but then they showed other kinds of pluralization in early Middle English and did not fully shift to the -s plural until late Middle English. The inflection of particular words is influenced by generalizing trends in pluralization (like umlaut or zero plural marking), and this analysis examines the mechanisms of the relationship between the two.

Introduction

7

Finally, in “From Shakespeare to Present-Day American English: The survival of ‘get + (XP) + gone’ constructions”, Lynn D. Sims focuses on another way that language change proceeds via the lexicon, namely through grammaticalization of verbs in constructions like “get thee gone”. The research traces the construction ‘get + (XP) + gone’ through corpora and dictionaries to contextualize the shift in usage as it expands in function in Early Modern English. The usage later contracts in British English, but persists in some dialects of American English.

3 Conclusion This volume results from the eighth meeting of the Studies of the History of English Language Conference (SHEL-8), held 26–28 September 2013, at Brigham Young University in Provo, Utah. The conference presented a fruitful opportunity for scholars to come together over inquiries into historical English language research, continuing and developing lines of inquiry from previous meetings. The chapters in this subsequent volume develop a common theme that emerged from the conference: the ways that scholars use the written record to understand language change. Written documents make possible historical language scholarship, and their contours and particulars determine and constrain what we can learn about the English of the past. These articles each present different angles on how to approach language change through written records of early English, and they collectively illustrate the tensions of moving along the cline of particularizing and generalizing perspectives. Just as artists have tiny brushes for detail work and large brushes for sweeping strokes, scholars of the history of the English language have sharp qualitative tools for close analysis and broad-spectrum quantitative and theoretical tools for larger-scope examination. The art of historical linguistic research employs them both.

References Bergs, Alexander & Laurel Brinton (eds.). 2012. Historical linguistics of English: An international handbook, vol. 2. Berlin: De Gruyter Mouton. Biber, Douglas & Randi Reppen. 2015. Introduction. In Douglas Biber & Randi Reppen (eds.), The Cambridge handbook of English corpus linguistics, 1–8. Cambridge: Cambridge University Press. Busse, Beatrix. 2012. Historical text analysis: Underlying parameters and methodological procedures. In Andrea Ender, Adrian Leeman & Bernhard Wälchli (eds.), Methods in contemporary linguistics. 285–308. Berlin: De Gruyter Mouton.

8

Colette Moore and Don Chapman

Coffin, Caroline, Theresa M. Lillis, & Kieran O’Halloran. 2010. Applied linguistics methods: a reader: systemic functional linguistics, critical discourse analysis and ethnography. London: Routledge. Dörnyei, Zoltán. 2009. Research methods in applied linguistics: Quantitative, qualitative, and mixed methodologies. Oxford: Oxford University Press. Ender, Andrea, Adrian Leeman, & Bernhard Wälchli (eds.). 2012. Methods in contemporary linguistics. Berlin: De Gruyter Mouton. Jacobs, Andreas & Andreas H. Jucker. 1995. The historical perspectives in pragmatics. In Andreas Jucker H. (ed.), Historical pragmatics. Pragmatic developments in the history of English, 3–33. Amsterdam & Philadelphia: John Benjamins. Johnson, Keith. 2008. Quantitative methods in linguistics. Malden, MA: Blackwell. Litosseliti, Lia (ed.). 2010. Research methods in linguistics. London: Continuum. Nevalainen, Terttu and Elizabeth Traugott (eds.). 2012. Rethinking evidence. The Oxford handbook of the history of English. 19–156. Oxford: Oxford University Press. Taavitsainen, Irma and Susan M. Fitzmaurice. 2007. Historical pragmatics: What it is and how to do it. In Fitzmaurice, Susan M. & Irma Taavitsainen (eds.), Methods in historical pragmatics, 11–36. Berlin: de Gruyter.

I Particularizing and generalizing for written records

R. D. Fulk

A philological tour of HEL 1 Introduction The aim of the present study is to shed light on various of the issues in the sometimes fraught relations between philology and English linguistics, in the process illustrating the roles that philology plays in the study of the history of the language. Firstly (§2), competing definitions of philology will be examined to determine which is most relevant to English historical linguistics, and, secondly (§3), reported notions about the scientific status of linguistics vis-à-vis philology will be considered. Following thereupon (§4), it will be demonstrated what untenable conclusions commonly result when linguistic studies are insufficiently informed by philological concerns. As will be argued ultimately (§5), excluding philological considerations from historical linguistic analysis is, in any case, a futile endeavor, since the data upon which such analysis is constructed can be derived only by philological methods. A summary, with conclusions, is offered in section six.

2 Defining philology Definitions of “philology” vary widely; several are discussed below.1 The present study, however, concerns the role of philology in English historical linguistics, and in this context, the most immediate obstacle to defining philology is that there is little agreement how linguistics and philology are related. A survey of linguists (as well as members of the American Speech and Hearing Association) conducted by Winters and Nathan (1992) found widely divergent opinions, including the view that philology is the ancestor of linguistics, that it is a data source 1 This study will not touch on those aspects of philology that have no direct bearing on the history of the English language. Thus, “philology” is here rather narrowly defined, despite the imprecision of the definition ultimately offered. The reiterated call for a “return to philology” as a means of reinvigorating literary studies, a trend initiated by de Man (1982; reprinted in de Man 1986: 21–26), is reviewed critically by Ziolkowski (2005) and Fulk (2016: 107). See also Warren (2013) for a recent array of perspectives by various hands on the present and future of philology. For a history of philological study from Greek antiquity to the early twentieth century, see Turner (2014), and specifically on English philology in the nineteenth century, see Momma (2013).

12

R. D. Fulk

for linguistics, that the two interact mutually, that there is no relation, that philology is a branch of linguistics, and, conversely, that linguistics is a branch of philology. The survey also confirmed the impression which the authors gathered from the literature that contempt for philology is not uncommon among linguists, since nearly ten percent of respondents provided “[a]nswers that can only be described as abusive” (1992: 357). Such animus may perhaps be explained in part by misconceptions about the nature of philological inquiry, as suggested by the disagreements just cited as to how philology and linguistics are related. Yet even among philologists there is no consensus about how to define the field. In a recent concise account of philological methods, the present writer identified some of the commonest definitions (Fulk 2016). According to one of these, philology is an aggregate of the various modes of inquiry required for the editing of texts in extinct languages (see, e.g., Gumbrecht 2003: 1–4), an endeavor that has for most of its history had as its chief aim recovering authorial versions of texts altered in the course of transmission. [. . .] Philology may thus involve historical and comparative linguistics and the study of manuscripts (including paleography, codicology, the study of how manuscripts are related to one another, and scribal practices), orthographic systems (including orthoepy), poetic meter, rhyme, translators’ practices, and numismatics, among other concerns. (2016: 95)

As explained in that study, this view of present-day philology is a consequence of disciplinary history. In the nineteenth century the term “philology” encompassed linguistics as a whole, as well as literary criticism. With the rise of structuralism in the early twentieth century and the concomitant displacement of diachronic concerns from the center of linguistic study, linguistics was established as a discipline unto itself.2 Likewise, with the development of methods that valued aesthetic concerns over matters of language study, literary criticism also established a separate identity. Thus bereft of two of its major components, philology soldiered on with the textual concerns that were still its exclusive province. It is particularly in classical studies that philology is often regarded as more or less synonymous with textual editing, but more than half of Winters and Nathan’s respondents (1992: 356) likewise identified philology as in some manner devoted to the study of texts.3 2 It should be said that already in the second half of the nineteenth century a distinction was being drawn between Philologie and Sprachwissenschaft or Linguistik among German scholars, as documented by Koerner (1982). But linguistics remained solidly diachronic in focus at the time. 3 Interestingly, Mańczak (1990) mounts the argument that linguistics is as much the study of texts as philology is, and in a more concrete sense than is suggested below in section five.

A philological tour of HEL

13

The other most prevalent understanding of philology is that it is in essence historical linguistics. Indeed, “historical linguistics” is one definition of “philology” offered in the OED, and more than a third of Winters and Nathan’s respondents identified it as such (1992: 356–357). The reason is, once again, the shift of focus to synchronic phenomena among most linguists beginning in the early twentieth century, creating the impression that historical language study is to be associated with nineteenth-century methods. It is to be noted, however, that this identification of philology with historical linguistics is purely an anglophone phenomenon. As Koerner (1982: 404–405, 408–409) remarks, French philologie and German Philologie retain more of the older, literal sense of philologia as ‘love of learning and literature’, and they thus pertain to wider cultural concerns. A third view of the matter was offered by the present writer in the study cited above: “[T]he philological component of historical linguistic study may be regarded as the extralinguistic contexts of linguistic data, or the relation between contexts and data” (Fulk 2016: 95).4 The fundamental problem with diachronic linguistics is that the researcher, if dealing with language use before the invention of sound recording, does not have direct access to oral language use but must glean data from textual media of uncertain reliability. For any number of reasons, an inscribed, written, or printed text may not reflect actual usage as it would be encountered outside the textual medium. It may be that a written text represents the last copy of many in a series in which successive copyists have attempted to update archaic language and alter forms foreign to their own dialect but have succeeded only imperfectly, with the result that the text represents a mixture of dialect forms and older and newer vocabulary, as is commonly the case in regard to texts in Middle English (see, e.g., McIntosh, Samuels & Benskin 1986: 1.12–1.23). Or it may be that the language used in a text is artificial, in the sense that it conforms to a conventional or imposed literary or professional standard, as is the case in regard to most Old English poetry, which is recorded in a koine comparable to that of Greek epic poetry (see, e.g., Sisam 1953),5 and in regard to most legal texts in English of any period (see, e.g., Mellinkoff 1963). Or it may be that the syntax of a text is artificial, since it imitates the word order or grammatical constructions of a 4 Compare Blockley (1999: 6): “Linguistics is the purity of the unseen, almost unseeable things for which texts are only the evidence. Philology is everything else” that pertains to language. 5 Here the term koine is used in the sense applied to it by Anglo-Saxonists, referring to a common poetic language incorporating features of more than one dialect, just as Homeric poetry is composed in Ionic, with a mixture of forms from Aeolic and Mycenaean. The term is potentially confusing, since it does not refer specifically to the Greek Koine of the Hellenistic Period, though it, too, even if Attic at base, was an artificial, mixed dialect.

14

R. D. Fulk

language from which it is translated (see, e.g., Mitchell 1985: §§3814–3845; Timofeeva 2010). A number of other factors may also bring into question the extent to which historical texts represent ordinary linguistic usage, such as scribal error, manuscript damage, accommodations made for the sake of meter and rhyme in poetry, and printers’ alterations in the course of typesetting. It is factors such as these that sometimes render it impossible to apply the methods of synchronic linguistics to the language of historical texts. It is also factors such as these that are matters of intense interest to philologists.

3 Philology, linguistics, and science It is in acknowledgment of such uncertainties in connection with the materials available for the study of premodern languages that Labov (1994: 11) offered his much-cited definition of historical linguistics as “the art of making the best use of bad data”. This is an artful definition. Firstly, referring to historical linguistics as an art appeals to the notion discerned in the linguistic literature by Winters and Nathan (1992: 353, 363–364) that philology is unscientific. (See also Koerner 1982: 404.) This suggests one explanation for the animus expressed by some linguists toward philology, as illustrated in Figure 1.

Figure 1: One model of the source of linguistic animus toward philology. “CHAOS” = “a dark / Illimitable Ocean without bound, / Without dimension, where length, breadth, and heighth, / And time and place are lost,” etc. (Milton 1667: Book II, ll. 891–894).

Secondly, applying the term “bad data” to the linguistic content of historical texts ensures that the nature of historical linguistics will be observed from the perspective of linguistics rather than, say, of history or philology. To the philologist, by contrast, the contextual impediments to straightforward linguistic analysis

A philological tour of HEL

15

are at least as interesting as those linguistic conclusions that can legitimately be drawn, and for philological purposes they thus are to be regarded, in and of themselves, as good data. The notion that philology is unscientific, according to the findings of Winters and Nathan (1992: 364), is to be associated with the idea that philology privileges data over theory. What is meant by “theory” here cannot be what is normally meant by the word in the experimental sciences, where it refers to a set of structured assumptions that can never be proved correct but can be rendered virtually certain if it explains sufficient phenomena without being refuted by counterevidence, but which must remain capable of invalidation by such means. (For discussion, see Fulk 1992: 6–24, Branchaw 2014). In this sense philology can be called highly theoretical, involving the constant formation of hypotheses involving inductive reasoning (as in all the social sciences), for instance about the original dialect and date of texts altered by scribes in the course of transmission (as described in §2 above), about etymologies, and about the phonetics of Early Modern English, to the extent that its sounds can be reconstructed on the basis of early orthoepic tracts and rhymes in verse, among other factors. Rather, what is meant by the idea that philology is atheoretical is undoubtedly that it is not governed by schools of thought like Structuralist Theory, Principles and Parameters Theory, and Optimality Theory, theories that arise, evolve, and then, instead of becoming disciplinary fixtures like quantum theory and evolutionary theory, give way to newer approaches, seemingly passing out of fashion the way varieties of literary theory and other non-scientific theories do. Yet language is a human behavior, and research in neither Optimality Theory nor Generative Theory is conducted on the empirical grounds normally practiced in the behavioral sciences.6 The resulting incapability of such theories of being invalidated makes the notion that philology is, by comparison, unscientific, seem odd, at least to a philologist, though perhaps also to other behavioral scientists. That science in the view of some linguists is a matter of supposed purity of method is especially plain in the realm of phonology, since it is not an uncommon view among phonologists that resorting even to analogy to explain certain historical developments is a betrayal of science, which instead demands a purely phonological explanation. For example, the viewpoint that morphological explanations are not scientific explanations is expressed in especially 6 On the mathematical, non-empirical basis of Generative Theory, in which probability is tied to elegance of formulation rather than the results of experimentation, see Tomasello (1995: 131, 134–136).

16

R. D. Fulk

plain terms in a rather startling rebuff to advocates of a certain analogical change in Northwest Germanic, to the effect that they “all invent (or adopt from each other) surprising morphological solutions to a straightforward problem of linguistic development, rather than adhering – as I think one must – to established methods of linguistic reconstruction the more tenaciously the more difficult the problem appears to be” (Vennemann gen. Nierfeld 1997: 298).7 If science is what, in the view of some linguists, distinguishes their work from that of philologists, it is striking how many would appear to believe that the aim of language science is to account for phenomena from as insular a perspective as can be achieved, rather than taking into account all the factors that are known to have a bearing upon language use. Some of the respondents to the survey of Winters and Nathan grounded their distaste for philology in the observation that they simply found it “boring” (1992: 364). Although the authors rightly observe that “what one person finds dry and boring [. . .] will prove fascinating to another” (363), it would probably be inaccurate on that basis to assume reciprocal attitudes and suppose that philologists find linguistics uninteresting. After all, philology is heavily dependent on historical linguistics. And yet, given the questionable scientific status of some linguistic theories as approaches to human behavior, as mentioned above, the reluctance of some philologists to embrace particular varieties of linguistic theory is unsurprising. For the present writer, no small part of the appeal of philology is its capability of affording what Ziolkowski (2005: 250) has described in characteristically apt fashion as “the humane experience of reading and making contact with the past that draws many students and general readers to the Middle Ages.” Or to put this in more objective terms, philology (and particularly historical linguistics) enables the recovery of aspects of sometimes quite distant history that are remarkably detailed, so that it has the capability of saying something factual about past events that it would otherwise be impossible to establish. By comparison, linguistic theory tends to be abstract, and what its relation is to human cognition is not always plain. It in fact has been an axiom of Generative Theory that the grammars it produces are not representations of actual human cognition but are only, in a sense, metaphorical (see Tomasello 1995: 139, note 2). It should hardly seem odd, then, that some should prefer a discipline grounded in the (relatively) actual to one whose relevance to anything other than itself is not always plain.

7 For a less doctrinaire expression of a similar attitude, see Krygier (2000: 463).

A philological tour of HEL

17

4 Three case studies of linguistics unsupported by philology Linguists, naturally enough, would like to work with “good data,” data that can be subjected to statistical analysis without any of the qualifications demanded by such extra-linguistic contextual factors as those mentioned in section two above. The insistence of the philologist on the indispensability of such factors in the study of non-contemporary language thus inevitably casts him or her in the role of spoiler – perhaps another reason for the animus modeled in Figure 1. No matter the rancor such a role may provoke, however, the spoiler’s function is one that cannot be dispensed with if the conclusions of linguistic studies are to be valid, because any philologist who takes an interest in linguistic theory is inevitably familiar with scores of linguistic studies, even some by distinguished scholars and appearing in eminent journals, which are rendered worthless by their failure to take philological factors into account. An example of a study vitiated in this manner is Taylor (2005), arguing that OV order is base-generated in Old English, and whereas VO order is derived, its increasing frequency over time indicates the rise of base-generated VO order. The data adduced to support this conclusion, however, are problematic in several respects. The author derives the data from just three texts, Beowulf, The Meters of Boethius, and Ælfric’s Lives of Saints, selected to illustrate change over time, since they were most likely composed a century or more apart, in the given order. All three she refers to as “metrical” (Taylor 2005: 139), though the last does not conform to any meter and resembles verse only in that it is divisible into lines, most of which exhibit alliteration, though not generally in accordance with the alliterative patterns found in verse. Inclusion of The Meters of Boethius is also questionable: regardless of the oft-remarked inferiority of the Meters’ verse form, the text is the result of turning a prose translation of the Latin original into verse by means of minimal alterations, so that it ought to be assumed that the Meters exhibit the influence of prose syntax (on which see Suzuki 2010 in connection with verbs). It is not explained why the study was restricted to alliterative texts, and to just three of them. This restriction is unfortunate, because the word orders studied are constrained in verse by metrical considerations. The metrico-syntactic work of Kuhn (1933) shows that finite verbs receive less stress than nouns in Old English (as in many other Indo-European languages), confirming the apparent implications of a metrical regularity known as Sievers’ Rule of Precedence (on which see Fulk 1992: 182, note 23). Because non-finite verbs are sometimes unstressed, presumably they, too, received less

18

R. D. Fulk

stress than nouns, though more than finite verbs. This explains why, in a construction like wel bið þæm þe mot / æfter deaðdæge drihten secean ‘it will be well for the one who may seek the Lord after [his] day of death’ (Beowulf 186b– 187),8 the noun drihten ‘Lord’ must precede the infinitive secean ‘seek’, since it, and not the infinitive, must bear the alliteration. The higher rate of exceptions in the Meters is presumably due to the poet’s inferior versecraft and in the Lives of Saints to composition in prose. Whether or not the three texts evidence the rise of a new base element order thus cannot be demonstrated on the basis of data derived from these texts, since the differences could be due to their relative conformity to principles of poetic form. Another sort of problem stemming from insufficient attention to philological matters may be illustrated by reference to Maeda (2008), arguing that because the subjunctive mood is properly a feature of subordinate clauses, its use in some primary clauses in Old English requires historical explanation. An example of such use is fæder alwalda / mid arstafum eowic gehealde / siða gesunde ‘may the all-powerful father with generosity keep you safe in your travels’ (Beowulf 316b–318a), where gehealde ‘keep’ is subjunctive. Maeda’s argument is that such primary clauses are in origin noun clauses beginning with the complementizer þæt ‘that’ preceded by some such primary clause as ic bidde ‘I pray’, though the main clause and the complementizer have been deleted. Before addressing the main philological problems attending this argument, it may be noted that in actuality, the reason that the subjunctive may appear in primary clauses in Old English is that the Germanic subjunctive is formally derived from the Proto-Indo-European optative – some grammars in fact refer to the Germanic subjunctive instead as the optative – and the Germanic subjunctive thus combines the functions of the Proto-Indo-European optative and the subjunctive. The Indo-European optative mood is chiefly to be found in primary clauses. The uses of the optative in Proto-Indo-European are generally assumed to have been like those in Sanskrit, where the optative sense (as in the Old English example just given) is primary, but from it have developed jussive, hortative, and directive senses such as are to be found in the Old English subjunctive. As Whitney (1889: §573) observes of the Sanskrit optative, [T]he expression of desire, on the one hand, passes naturally over into that of request or entreaty, so that the optative becomes a softened imperative; and, on the other hand, it comes to signify what is generally desirable or proper, what should or ought to be, and so becomes the mode of prescription; or, yet again, it is weakened into signifying what may or can be, what is likely or usual, and so becomes at last a softened statement of what is. 8 Quotations from Beowulf derive from Fulk, Bjork and Niles (2013), though the diacritics have been removed. All translations are my own, unless otherwise indicated.

A philological tour of HEL

19

The chief philological problem with Maeda’s argument, however, is that what is termed “the most compelling evidence” (2008: 105) for the historical explanation offered is the witness of Old English primary clauses containing the subjunctive from which the complementizer þætte ‘that’ has not been deleted. An example cited from the Old English Pastoral Care is þætte unlærede ne dyrren underfon lareowdom (Sweet 1871: 8), for which Maeda provides the rendering ‘Don’t let the unlearned take up teaching’.9 But this, like the one other example of such a construction that Maeda cites, is actually a chapter heading, which the editor, Sweet (1871: 9), correctly renders ‘That the unlearned are not to presume to undertake the office of teacher’. That is to say, the chapter title is not an injunction but a description of the argument of the chapter. That Sweet’s is the correct interpretation is shown by the use of similar chapter headings beginning with þætte or þæt, often collected to provide a table of contents before the first chapter of the work, in other Old English texts, including texts in which the aim is not, as in the Pastoral Care, to prescribe Church policy, but simply to provide information about what the chapter concerns, so that the title cannot be interpreted as an injunction. An example from the Old English Bede is Ðæt se ærra Romwara casere Gagius Iulius Breotene gesohte, which the editor (Miller 1890–1898: 6, 7) renders ‘That the first emperor of the Romans, Gaius Julius, visited Britain’. There is thus no evidence in Old English for Maeda’s position, and the use of faulty evidence such as this is due to inattention to the extralinguistic contexts in which the linguistic examples appear. A third example pertains to the idea originating with Kurath (1956), and subsequently adopted in numerous studies and handbooks, that the idiosyncratic orthography of the early Middle English Ormulum is best explained on the assumption that in the dialect represented by this text, Middle English Open Syllable Lengthening (MEOSL) has already taken place. This assumption is designed chiefly to account for the apparent orthographic irregularity in the text whereby a consonant is consistently doubled after a short vowel in a closed syllable but not in an open. For instance, the author, Orm, who is also the scribe, writes wunndrenn ‘to wonder’ and þatt ‘that’, but sune ‘son’ (all with short root vowels; compare uppe ‘up’, with an etymological geminate); compare suþ ‘south’, mæless ‘meals’, and tacnenn ‘signify’, all with long root vowels. If MEOSL has taken place, the orthographic rule is simple: a short vowel is always followed by a geminate. A number of reasons have been offered why this view of the meaning of Orm’s orthography is difficult to credit (see Fulk 1996). By Kurath’s reasoning, if Orm’s practice is regular, geminate consonants can only be a diacritic indicating vowel shortness, which means that original geminates 9 The rendering of the verb as an imperative is Maeda’s, based on the mistaken assertion of Visser (1963–73: 2.806) that this is not a subordinate clause.

20

R. D. Fulk

must be assumed to have been degeminated, and this in turn demands the assumption that the Old English allophonic status of voiced and voiceless fricatives has in this text been phonemicized, as voiceless word-internal geminate fricatives, which were voiceless in Old English, must now contrast phonemically only in terms of voicing with non-geminate fricatives, which were voiced between voiced sounds in Old English. Yet despite the considerable care Orm shows in encoding other phonemic contrasts in his orthography (and even some non-phonemic ones), he draws no distinction between voiced and voiceless fricatives: to heofennlike ‘heavenly’ compare nominative heoffne ‘heaven’, both with [v], and to genitive wifes (with [v]) compare nominative wif ‘woman’ (with [f]). Several purely philological considerations also tell against the supposition that MEOSL has already taken place in Orm’s dialect. Orm not infrequently places a mark like a curl over an etymologically short vowel to distinguish homographs, for example, tăkenn ‘taken’, with an etymologically short vowel, beside takenn ‘token’, with an etymologically long one, and this practice would hardly be comprehensible if the root vowel in the former had been lengthened. Likewise, in the septenary meter of the Ormulum (also called “common meter” or “ballad meter”, as in the English carol “Good King Wenceslas”) the penultimate syllable of even-numbered verses must be heavy, in order for the trochaic foot to fill four instead of two beats, and yet at the end of even-numbered verses Orm studiously avoids using words like wĭtenn ‘know’ and ifell ‘evil’, with an etymologically short root vowel, whereas in this position he does use words like wanenn ‘bewail’ and dedess ‘deeds’, with an etymologically long root vowel. A consideration of scribal practice in regard to orthographic geminates in late Old English and early Middle English suggests further that Orm’s use of the device would be historically anomalous if he intended it to indicate only vowel shortness, since it is at least as closely, if not more closely, to be associated with indicating syllable boundaries. If Orm also wished the device to indicate syllabification, then there is nothing very irregular about it: a geminate after a vowel indicates that the consonant is tautosyllabic with that vowel, whereas a non-geminate indicates heterosyllabicity. From these observations about philological matters in connection with the Ormulum it may be concluded that “the study of historical linguistics may be founded on delusive presuppositions when it is conducted outside of any actual historical contextualization of the texts in which the data are preserved” (Fulk 1996: 508). Examples like these could be multiplied endlessly in the literature.10 Let these few instances suffice to establish the point that conducting linguistic 10 For criticisms of some other linguistic studies that are insufficiently informed in regard to philological concerns, see, e.g., Bately (2009) and Fulk (1999 and 2016: 101–3). A number of other essays contained in the same volume as Maeda (2008) exhibit similar problems.

A philological tour of HEL

21

research on non-contemporary data in a philological vacuum very commonly leads to seemingly scientific but actually unreliable results.

5 Philology as both overt and covert component in English historical linguistics It was pointed out above (§2) that historical language study is associated in the minds of many with nineteenth-century methods. That philological methods are old-fashioned, or even entirely obsolete, is what Winters and Nathan (1992: 358) found to be the prevailing attitude among their younger or American-trained respondents. To be fair, this survey was conducted before the digital revolution, and the respondents could not have envisaged the extent to which corpus linguistics has come to dominate in the field of historical linguistics, and thus the extent to which its methods are anything but old-fashioned. Yet even if a definite distinction is drawn between historical linguistics and philology, the latter, too, has emerged in the past quarter century as a salient component of corpus-based studies in historical pragmatics. Elsewhere the present writer (Fulk 2016: 105–7) has discussed what has been termed “pragmaphilology,” which “describes 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 text” (Jacobs and Jucker 1995: 11). A particular concern discussed there is the encoding of manuscript features in historical corpora, such as capitalization, marginal and interlinear insertions, and various forms of deletion, since these hold clues to the transmission and reception of texts, clues of some relevance to pragmatic analysis. Outside of historical linguistics, too, philological enterprises have been at the forefront in the creation and deployment of digital resources, especially in connection with textual editing.11 A particularly impressive, international project of this sort, involving six general editors and more than fifty contributing editors, is The Skaldic Project (Clunies Ross et al. 2012). Despite philologists’ wholehearted embrace of newer technologies, there is a degree of justice in associating philology with older methods. If texts and the problems associated with them lie at the center of philological concerns, it should be recognized that although the rise of digital resources has greatly 11 Already, when the construction of digitized historical corpora was still young, Rissanen (1990) made the case that such resources point the way to a rapprochement between linguistics and philology.

22

R. D. Fulk

facilitated the editing of texts, editorial methods have changed little since the nineteenth century, since the uses to which editions are put has changed little, and the imperative of taking into account all independent witnesses to the establishment of the text remains unchanged.12 Likewise, in regard to another core philological project, the compilation of grammars of older languages, the needs of a modern readership have changed little since the nineteenth century, and although grammars oriented toward a particular linguistic theory, such as Structuralist or Generative Grammar, have appeared now and then, it is grammars constructed on Neogrammarian principles that remain the standard handbooks, since these provide the basic information on language development that linguistic theory must account for. One would not, for instance, use the Generative Grammar of Gothic contained in Voyles (1981) to develop an Optimality Theory account of Gothic phonological and morphological phenomena, but such a reference grammar as Braune (2004). This is not because reference grammars are in any genuine sense “pre-theoretical” – that is, uninformed by any particular theoretical model – but because they are more genuinely descriptive in orientation than any account aligned with linguistic schools of thought of the twentieth and twenty-first centuries, which inevitably must attempt to account for language phenomena on the basis of a specific linguistic theory to the exclusion of others, and they thus show particular biases. This last point suggests a more informed way to understand the conservatism of philological methods than to regard them simply as old-fashioned. Since the approach of linguists to non-contemporary languages must proceed from the grammatical facts established by the philological methods used to produce reference grammars and dictionaries, it may be seen that such linguistic approaches have a significant philological component built into them, even if they do not themselves engage in philological methods. In other words, however out of date philological methods may appear to linguists, they should be understood not as a thing of the past but as a form of scholarship that perpetually informs language scholarship, because no linguistic study of a historical language can dispense with the grammatical foundation that philology provides. Thus, in a manner analogous to the notion that, in biogenesis, ontogeny recapitulates phylogeny, or to the way, in generative grammar, the oldest sound changes remain fossilized within the grammar as relatively high-level morphologically

12 This is not to deny that editorial procedures are not monolithic, since texts edited for different purposes are edited differently, and editorial theory is hardly uniform even about texts edited for the same purpose (on which see Greetham 1999). But the needs of the text’s modern readership ensure that the range of methods employed in textual editing will continue to vary little from those employed more than a century ago.

A philological tour of HEL

23

conditioned rules, so studies in English historical linguistics inevitably recapitulate the history of the discipline, since they incorporate and are constructed upon a foundation of philological spadework built up since the nineteenth century. Elsewhere, to illustrate what is involved in philological work, the present writer has offered brief sketches of notable philological accomplishments in connection with the history of the English language, including the reasoning involved in the discovery that the Old English Genesis B is a translation from Old Saxon; in the explication of the metrical principles governing the construction of Germanic alliterative verse, revealing much about stress and vowel quantities; in the identification and localization of Old English dialects, along with recognition of the historical continuity between the language of the Old English gloss on the Vespasian Psalter and that of the Katherine Group in Middle English; in the development of the so-called fit-technique for the localization of Middle English texts of no known provenance, along with reliable methods for analyzing texts of mixed dialectal characteristics; and in the development of methods for the recovery of the phonology of Early Modern English (Fulk 2016: 99–101). Mention of the construction of reference grammars suggests another way to illustrate the eclecticism of philological endeavors. A reference grammar of Old English is naturally easier to compose today than it was a century and a half ago, when such a grammar on modern principles had not yet been compiled; now there are several excellent models, as well as other resources that were unavailable then, such as reliable dictionaries and, of course, electronic databases. It is therefore a tribute to Eduard Sievers that his grammar (first edition 1882), the first composed on sound linguistic principles, is still a valuable resource – all the more remarkable because very few grammars of individual texts had been published at the time of its appearance, and not even all the dialects of Old English had yet been correctly identified. Although the grammarian’s work is linguistic in nature, it depends vitally upon the work of textual editors, since they provide the data upon which grammars and dictionaries are based. For example, if an editor misreads a manuscript or fails to recognize scribal errors, this may have significant consequences for the construction of a grammar or a dictionary.13 Before Sievers’ grammar appeared, the editor’s chore was somewhat paradoxical in nature, since he needed a good grasp of Old English grammar in order to recognize errors in the text, and yet it was only after he had done his work that a sound grammar of the language could be formulated. In addition to grammar, lexicon, and idiom, an editor 13 Tinkler (1971) identifies many instances in which inferior editing has produced ghost words in dictionaries.

24

R. D. Fulk

must be skilled in manuscript studies: s/he must be able to read handwriting not only in Old English but also in the more difficult later hands that are not infrequent in the margins and interlinear spaces of Anglo-Saxon manuscripts; s/he must be able to detect alterations in the manuscript, and to offer plausible reconstructions of damaged passages; s/he must know codicology; for many Old English texts s/he must be able to identify and interpret their Latin source or sources; for poetic texts s/he must know Old English metrics and alliterative patterns well enough to detect textual corruption (and Old English meter was not well understood until Sievers published his first study of it, in 1885); and if the text is found in more than one manuscript, s/he must have a command of stemmatics, because how manuscripts are related to one another affects an editor’s choices when manuscripts disagree. The grammarian who relies upon the textual editor’s work must also have a command of all these aspects of manuscript studies, in order to assess the quality of the editor’s work, and thus it is not surprising that most of the great grammarians of historical stages of English have also been editors (see Fulk 2016: 98). There is thus a chain of dependency in historical linguistics, inasmuch as a linguistic study depends upon data provided in grammars and dictionaries, and grammars and dictionaries depend upon the prior work of textual editors, who in turn rely upon a variety of philological resources that are the product of the most elementary spadework, such as manuscript catalogues (in order to locate all instantiations of a text) and the grammars of individual texts compiled by prior editors. This chain of dependency can be seen to emulate and recapitulate the history of the discipline, presented here in reverse order.

6 Summary and conclusions Although no universally agreed-upon definition of philology is possible, for linguistic purposes it seems most relevant to regard it as “the extralinguistic contexts of linguistic data, or the relation between contexts and data” (Fulk 2016: 95). Although such a definition would appear to characterize philology as unscientific in nature, the competing claims of linguistics and philology to methodological validity are examined in section three above, and the basis for regarding one or the other as a more tenable and purposive approach to the analysis of human behavior is found to be more equivocal than might at first appear to be the case. Just as philology depends heavily upon historical linguistics for its successful prosecution, so the reverse dependency must also be recognized. One reason for this is that linguistic studies of a historical nature

A philological tour of HEL

25

very often draw wholly untenable conclusions when philological concerns are not accorded the consideration they demand, as illustrated by the three case studies presented in section four. Another reason is that the data upon which historical linguistics operates can be established only by philological methods. The result of this is that there is a philological component built into every diachronic linguistic study, so that even the most theoretical work on the history of the English language exhibits a temporal layering that represents a recapitulation of the development of the field since the early nineteenth century. Given the ways that historical studies tend to go seriously awry when they do not deal adequately with philological matters, it would seem indispensable to maintaining the scientific status of historical language study that philological methods be studied in linguistic curricula. Yet today most linguists receive little or no philological training. Among other indicators of the differential status of the two disciplines are the absence of any mention of philology from basic linguistic textbooks and the large number of linguistics departments at universities in the anglophone world, by comparison to the vanishingly small number of departments of philology (Winters and Nathan 1992: 353–354). The present situation would thus appear to pose grave dangers to the future of the discipline. The obvious remedy would begin with the rejection of the point of view modeled in Figure 1. Although this aim may appear to be unachievable in regard to non-historical linguists, the examples of the mistakes of historical linguistic studies suffering from inattention to extralinguistic contexts offered in section four argue strongly that the stake of historical linguistics in ensuring that philological methods and the backgrounds of historical data figure prominently in the linguistic curriculum should be a matter of the first concern to historians of the English language.

References Bately, Janet. 2009. Did King Alfred actually translate anything? The integrity of the Alfredian canon revisited. Medium Ævum 78. 189–215. Blockley, Mary. 1999. Philology, linguistics: Should you leave? In Joseph B. Trahern, Jr. (ed.), Thirty Years More of the Year’s Work in Old English Studies, 3–14. Kalamazoo: Medieval Institute, Western Michigan University. Branchaw, Sherrylyn. 2014. Principles of methodology: A case study from the history of English. English Language and Linguistics. 18.549–18.566. Braune, Wilhelm. 2004. Gothische Grammatik, mit Lesestücken und Wörterverzeichnis. 19th edn., rev. by Frank Heidermanns. Tübingen: Max Niemeyer. Clunies Ross, Margaret, Kari Ellen Gade, Guðrún Nordal, Edith Marold, Diana Whaley & Tarrin Wills (eds.). 2012. The Skaldic Project. httpsː//www.abdn.ac.uk/skaldic/db.php (accessed 13 May 2015).

26

R. D. Fulk

de Man, Paul. 1982. The return to philology. Times Literary Supplement 4158 (10 December 1982). 1355–1356. de Man, Paul. 1986. The Resistance to theory. Minneapolis: University of Minnesota Press. Fulk, R. D. 1992. A history of Old English meter. Philadelphia: University of Pennsylvania Press. Fulk, R. D. 1996. Consonant doubling and open syllable lengthening in the Ormulum. Anglia 114. 483–513. Fulk, R. D. 1999. Evaluating the evidence for lengthening before homorganic consonant clusters in the Ormulum. In Gerald F. Carr, Wayne Harbert & Lihua Zhang (eds.), Interdigitations: Essays for Irmengard Rauch, 201–209. New York: Peter Lang. Fulk, R. D. 2016. Philological methods. In Merja Kytö & Päivi Pahta (eds.), The Cambridge Handbook of English Historical Linguistics, 95–110. Cambridge: Cambridge University Press. Fulk, R. D., Robert E. Bjork & John D. Niles (eds.). 2013. Klaeber’s Beowulf and the Fight at Finnsburg. Reprinted from the 4th edn. (2008), with corrections. Toronto: University of Toronto Press. Greetham, D. C. 1999. Theories of the text. Oxford: Oxford University Press. Gumbrecht, Hans Ulrich. 2003. The powers of philology: Dynamics of textual scholarship. Urbana: University of Illinois Press. Jacobs, Andreas & Andreas H. Jucker. 1995. The historical perspective in pragmatics. In Andreas H. Jucker (ed.), Historical pragmatics: Pragmatic developments in the history of English, 3–33. Amsterdam: John Benjamins. Koerner, Konrad. 1982. On the historical roots of the philology/linguistics controversy. In Anders Ahlqvist (ed.), Papers from the 5th International Conference on Historical Linguistics = Referate von der fünften Internationalen Konferenz für Historischen Sprachwissenschaft = Communications de la cinquieme Conference internationale de linguistique historique, 404–413. Amsterdam: John Benjamins. Krygier, Marcin. 2000. Old English (non)-palatalised */k/: Competing forces of change at work in the ‘seek’-verbs. In Irma Taavitsainen, Terttu Nevalainen, Päivi Pahta, & Matti Rissanen (eds.), Placing Middle English in context, 461–473. Berlin: Mouton de Gruyter. Kuhn, Hans. 1933. Zur Wortstellung und -betonung im Altgermanischen. Beiträge zur Geschichte der deutschen Sprache und Literatur 57. 1–109. Kurath, Hans. 1956. The loss of long consonants and the rise of voiced fricatives in Middle English. Language 32. 435–445. Labov, William. 1994. Principles of linguistic change. Volume 1: Internal factors. (Language in Society 20). Oxford & Cambridge, MA: Blackwell. Maeda, Mitsuru. 2008. Insubordination in Old English. In Masachiyo Amano, Michiko Ogura & Masayuki Ohkado (eds.), Historical Englishes in varieties of texts and contexts: The Global COE Programme, International Conference 2007, 93–107. Studies in English Medieval Language and Literature 22. Frankfurt am Main: Peter Lang. Mańczak, Witold. 1990. The object of philology and the object of linguistics. In Jacek Fisiak (ed.), Historical linguistics and philology, 261–272. Berlin: Mouton de Gruyter. McIntosh, Angus, M. L. Samuels & Michael Benskin. 1986. A linguistic atlas of late mediaeval English. 4 vols. Aberdeen: Aberdeen University Press. Mellinkoff, David. 1963. The language of the law. Boston: Little, Brown. Miller, Thomas (ed.). 1890–1898. The Old English version of Bede’s Ecclesiastical History of the English People. 4 vols. EETS o.s. 95, 96, 110, 111. London: Trübner. Milton, John. 1667. Paradise lost: A poem written in ten books. London: Printed, and are to be sold by Peter Parker, etc. [Unpaginated.]

A philological tour of HEL

27

Mitchell, Bruce. 1985. Old English syntax. Oxford: Clarendon. Momma, Haruko. 2013. From philology to English studies: Language and culture in the nineteenth century. Cambridge: Cambridge University Press. Pinker, Stephen. 1994. The language instinct: How the mind creates language. New York: William Morrow. Rissanen, Matti. 1990. On the happy reunion of English philology and historical linguistics. In Jacek Fisiak (ed.), Historical Linguistics and philology, 353–370. Berlin: Mouton de Gruyter. Sievers, Eduard. 1882. Angelsächsische Grammatik. Halle: Max Niemeyer. Sievers, Eduard. 1885. Zur Rhythmik des germanischen Alliterationsverses. Beiträge zur Geschichte der deutschen Sprache und Literatur 10.209–10.314, 451–545. Sisam, Kenneth. 1953. Dialect origins of the earlier Old English verse. Studies in the history of Old English literature, 119–139. Oxford: Clarendon. Suzuki, Hironori. 2010. Ordering main and modal verbs in the production of Old English poetry. In Merja Kytö, John Scahill & Harumi Tanabe (eds.), Language change and variation from Old English to Late Modern English: A festschrift for Minoji Akimoto, Studies in Language and Communication 114, 139–160. Bern: Peter Lang. Sweet, Henry (ed.). 1871. King Alfred’s West-Saxon version of Gregory’s Pastoral Care. 2 vols. EETS o.s. 45, 50. London: Trübner. Taylor, Ann. 2005. Prosodic evidence for incipient VO order in Old English. English Language and Linguistics 9. 139–156. Timofeeva, Olga. 2010. Non-finite constructions in Old English, with special reference to syntactic borrowing from Latin. Mémoires de la Société Néophilologique de Helsinki 80. Helsinki: Société Néophilologique. Tinkler, John D. 1971. Vocabulary and syntax of the Old English version in the Paris Psalter. The Hague: Mouton. Tomasello, Michael. 1995. Language is not an instinct. [Review of Pinker 1994.] Cognitive Development 10. 131–156. Turner, James. 2014. Philology: The forgotten origins of the modern humanities. Princeton: Princeton University Press. Vennemann gen. Nierfeld, Theo. 1997. The development of reduplicating verbs in Germanic. In Irmengard Rauch & Gerald F. Carr (eds.), Insights in Germanic linguistics II: Classic and contemporary, 297–336. Berlin: Mouton de Gruyter. Visser, Frederikus Th. 1963–1973. An historical syntax of the English language. 3 vols. in 4 parts. Leiden: E. J. Brill. Voyles, Joseph B. 1981. Gothic, Germanic and Northwest Germanic. Wiesbaden: Franz Steiner. Warren, Michelle R. (ed.) 2013. Philology and the mirage of time. Postmedieval: a Journal of Medieval Cultural Studies 5 (4). 389–516. Whitney, William Dwight. 1889. Sanskrit grammar, including both the classical language and the older dialects, of Veda and Brahmana. 2nd edn. Cambridge, MA: Harvard University Press. Winters, Margaret E. & Geoffrey S. Nathan. 1992. First he called her a philologist and then she insulted him. In Diane Brentari, Gary N. Larson & Lynn A. MacLeod (eds.), The joy of grammar: A festschrift in honor of James D. McCawley, 351–367. Amsterdam: John Benjamins. Ziolkowski, Jan. 2005. Metaphilology. Journal of English and Germanic Philology 104.239– 104.272.

Donka Minkova

From stop-fricative clusters to contour segments in Old English 1 What is at issue? The evolution of the English consonantal system is hardly the most captivating topic in historical phonology; it is the vowels that usually take the center stage.1 Yet the consonants are also an excellent testing ground for reconstructing sound-spelling relationships and for comparing different sources of diachronic evidence. Consonantal histories that have attracted the most attention are fricative voicing, h-dropping, the vocalization of the rhotics, and initial cluster simplification. This study turns to one equally deserving topic that gets mentioned rarely, if at all: the development of the Present-Day English (PDE) affricates /tʃ/͡ ͡ and /dʒ/. It is quite surprising that Old English and Old Frisian are the only older Germanic languages for which these affricates have been reconstructed (Robinson 1992: 159). Where did they come from, and when can we posit the first contrastive affricates in English? Do affricates fit the pattern of singletons vs. ͡ develop in tandem, or is ͡ and /dʒ/ geminates in the Old English system? Do /tʃ/ there a lag time depending on voicing? The paper starts with a brief introduction to the structural and functional differences between simple, complex, and contour segments. The Old English consonantal inventory is presented in section 3, which also looks at the possible sources and mechanisms of affrication in Old English. The next section addresses the question of singleton vs. geminate (pre-)affricates in Old English. Section 5 surveys the metrical treatment of Old English , , and in alliterative verse. Section 6 discusses the different trajectories of the voiceless and voiced palatalized velars. The relation of the metrical evidence to orthography is covered in section 7. Section 8 offers concluding remarks and proposes a revision of the inventory presented in section 3.

2 Contour, complex, and simple segments The two endpoints in the history of the affricates – from pre-Old English to Present-Day English – are uncontroversial: in Proto-Germanic there may have 1 I am very grateful for the careful editorial reading of this study and especially for a much appreciated, peerlessly erudite and eagle-eyed peer review, belying the term “blind” review.

30

Donka Minkova

been velar geminates due to assimilation, but there were no affricates.2 PresentDay English has affricates, though their realization and the way they are syllabified do raise some questions. For some items there is rivalry between [ʒ] and ͡ I say [gəˈrɑʒ], you say [ˈɡærɪdʒ]. ͡ Is the syllabification of medial affricates [dʒ]: (e.g., Thatcher vs. catcher, pigeon vs. bridges) equally susceptible to morphological boundaries as other sounds resulting from cluster simplification (e.g., finger with medial cluster [ŋg] vs. singer with a medial singleton [ŋ])? Such issues highlight some of the uncertainties in the treatment of affricates, which will be bypassed here in favor of a brief introduction to the characterization of the affricates in the overall consonantal system with focus on the structural and functional differences between simple, complex, and contour segments. First, consider the similarities and differences between (1a) and (1b): (1)

a.

Why choose [waɪ. tʃ͡ uːz]

b.

white shoes? [waɪt.ʃuːz] Examples from Cruttenden (2008: 307)

A slow and careful pronunciation of the question in (1) will distinguish between (a) and (b) by inserting a boundary before [tʃ͡ ] in (1a) and by separating the stop and the fricative in (1b). In fast speech this difference can be neutralized, yet the phonemic content is distinct: choose has the voiceless palato-alveolar affricate ͡ while the similar-sounding white shoes is a sequence of a voiceless alveolar /tʃ/, stop [t] plus a voiceless palato-alveolar fricative [ ʃ ]. Phonetically, affricates are stops in which the release of the constriction produces a prolonged friction, creating a contour segment unrecognized in the IPA consonantal chart. Ladefoged and Maddieson (1996: 90) describe the phonetic nature of affricates as “an intermediate category between simple stops and a sequence of a stop and a fricative”. Crucially, affricates are phonological single units whose complexity is captured with reference to their autosegmental features (see Clements and Hume 1995: 251–257). The basic tenets of that theory are that autosegmental features have a 2 The possibility of voiceless velar geminates based on assimilation of velar stop + [n] in ProtoGermanic is noted in Prokosch (1939: §22) and Krahe-Meid (1969: §99). Both sources remark on uncertainties of the reconstruction, but neither one identifies the most serious problem in positing assimilation of [kn] to [kk], namely that both voicing assimilation and place assimilation are typically regressive, targeting the first consonant (i.e., in intervocalic C1C2 clusters, C2 is the expected trigger; see Jun [2011]).

From stop-fricative clusters to contour segments in Old English

31

degree of functional independence and that these features may be hierarchically structured. In terms of place of articulation, contour segments involve sequences of articulations within a single contrastive unit: choose contrasts with twos, shoes, lose, and booze. Contour segments are “single slot” occupants represented on a separate level as “roots”. If the articulation involves more than one place, segments are “complex”, as in wheel, whale, where the initial [ʍ-] is both labial and dorsal (Hayes 2009: 97); Ladefoged and Maddieson (1996: 328–329) call them “doubly articulated”.3 In terms of segmental representation, features are linked to the segmental slots by association lines. They occupy their own tier and can be independently “active”. (2a) represents a contour segment, while (2b) and (2c) are combinations of simple segments that can have different functional properties: (2b) is a geminate, and (2c) is just an accidental concatenation of two autonomous speech sounds.4 (2)

a.

[-cont] ǀ [t]

[+cont] ǀ [ʃ] \/ /tʃ/͡ choose

b.

[-cont] ǀ [t] /\ [t] [t] OE bitte ‘bucket’

c.

[-cont] ǀ [t] ǀ /t/ white

[+cont] ǀ [ʃ] ǀ /ʃ/ shoes

A relevant phonetic property of English contour segments, which distinguishes (2a) from (2c), is their reduced duration.5 In principle, all consonants are shortened in clusters. However, the affricates are far shorter than the effect of shortening in clusters can account for: if [dʒ] and [tʃ ] were bisegmental clusters, their durations would be around 207 milliseconds for [tʃ ] (Sagey 1986b) and 192 milliseconds for [dʒ]; the drop in duration is comparable to Lavoie’s (2009) measurements:

3 Strictly speaking, “doubly articulated” does not distinguish between contour and complex segments. Since the affricates discussed here share place of articulation, the difference between contour and complex segments will not be pursued further. 4 The most appropriate phonological representation of affricates is still under discussion. See Hall (2012) and references therein. 5 Closure duration is the most salient acoustic indicator of consonantal length/gemination. “In languages with a phonological contrast between long and short consonants, long stops have between one and a half to three times the acoustic duration of short stops in careful speech” (Ladefoged and Maddieson 1996: 92).

32 (3)

Donka Minkova

Duration of PDE contour segments vs. bisegmental sequences in milliseconds.6 Sagey Lavoie Sagey Lavoie Bisegmental: tʃ 207 n/a dʒ 192 157 ͡ ͡ Contour: tʃ 159 133 dʒ 133 92 Simple t 91 36 ([ɾ]) d 88 71 ʃ 121 ʒ 86 Compare s 113 z 75

Sagey (1986a: 82–83) reports an experiment comparing the length of affricates in English to English stops, fricatives, and stop-fricative bisegmental clusters. She ͡ in English are significantly shorter than found that the affricates /tʃ/͡ and /dʒ/ the stop-fricative clusters [ts, ps, gz, ks]. While all consonants are shortened to some extent in clusters – in her data, consonants in stop-fricative clusters are shortened to between 90–98% of their durations in VCV context – the affricates are far shorter than the effects of shortening in clusters alone could explain. On the other hand, while the duration of contour segments is shorter than the com͡ bined cluster duration, English /tʃ/͡ and /dʒ/are still significantly longer than the simple segments in their respective subgroups of voiceless (longer) and voiced (shorter) durations. The phonetic duration of affricates raises the question of the criteria on which they are assigned a single-unit status. Cruttenden (2008: 182–184) identifies four such criteria: distribution, possibilities of commutation of the elements, native speakers’ reaction and speech errors, and glottalization. First, in terms of distribution, the affricates fare best in comparison to the distribution of other bisegmental clusters that could be potentially treated as units, such as [tr, dr, ͡ ts, tz]. English /tʃ/͡ and /dʒ/can be word-initial and word-final (e.g., cheap, jeep, clutch, fudge), and they can be preceded by /l/ and /n/ (e.g., inch, binge, mulch, and bulge). They contrast between two types of word-medial realization: “closeknit” (e.g., butcher, aged) vs. “disjunct” (e.g., lightship, ground-joint). The possibilities of commutation (= substitution) of the components are limited: the voiceless affricate word-initial [t] commutates only with zero (e.g., choose, shoes) while the [ ʃ ] can be substituted by /r, j, w/ and zero (e.g., cheese, trees, Tuesday, tweet, and tease). For the voiced affricate the substitution options are even more limited, not least due to the marginality of initial [ʒ] in English, while 6 Measurements are cited from Sagey (1986a: 82–83) and Lavoie (2009: 36). Lavoie’s measurements are averaged across all speakers and all positions. Lavoie does not include [t] in her chart; it is included only as a flap [ɾ] at 36 ms. The timing of perception (after another segment) is most localized for affricates – it is almost entirely in the release, see also Warner et al. (2014).

From stop-fricative clusters to contour segments in Old English

33

the components of the clusters [tr] and [dr], especially the stops, are much more freely commutable. On the third criterion, native speakers’ reaction, Cruttenden (2008: 183) writes that “it seems that the native speaker does not regard /tʃ/ and /dʒ/ as composite sounds”, though no further specifics are offered. Another test of unitary treatment which he cites, first explored in Fromkin (1971), is the coherent behavior of the affricates in sound transposition: play the game may become pay the glame, but chop the wood will not become *shop the twood. On all three of these counts the Present-Day English affricates align best with singleton phonemes.7 On the other hand, affricates typically display “edge” effects, which are the basis for determining the featural composition and the allophonic patterning of the components of the affricates (see Sagey 1986a; Hall 2012). By definition, such effects are diagnostics for the independent function of the components. (4)

͡ ͡ and /dʒ/: Edge effects of the components of /tʃ/ a. Feature [-continuant] for [t] independently active in aspiration: [tʰɪp] – [tʃ͡ ʰɪp] b.

͡ Glottal reinforcement and replacement of [t]: [[ti:ʔtʃɪŋ], [kaʊʔʃ ]

c.

Feature [+strident] for [ ʃ ] and [ʒ] independently active, matching [s] and [z]

d.

Omission of [t, d], but not of [ ʃ, ʒ] in adjacent affricates

e.

͡ or [tʃ,͡ dʒ] ͡ +C No initial C + [tʃ,͡ dʒ]

One parameter on which the compositionality of the affricates is identifiable is that they pattern together with voiceless stops; both are aspirated word-initially and in the onset of stressed syllables, thus [tʃ͡ ʰɪp] ‘chip’, [tʃ͡ ʰɪ ˈ pɑtli] ‘chipotle’, and [əˈ tʃ͡ ʰiːv] ‘achieve’ (see Hammond 1999: 221–224). This justifies the independent inclusion of the feature [-continuant] in representations such as (2a). (4b) addresses two other allophonic variables (Cruttenden 2008: 180, 183). In British ͡ English [tʃ]͡ is subject to glottal reinforcement (e.g., [ti:ʔ tʃɪŋ] ‘teaching’). This type of allophonic realization occurs in the environment of [p, t, k] + another 7 Although Cruttenden does not include this in the criteria for unitary analysis of the affricates, his comments on the acquisition of affricates (2008: 187) are also relevant in this context: “It might be expected that, being composed of a homorganic sequence of plosive plus fricative, their [the affricates’] acquisition would depend on the prior acquisition of the plosives and fricatives of which they are composed. However, this does not always seem to be the case; in particular, the fricative /ʒ/ may be of later occurrence than the affricate /dʒ/, perhaps due to its comparative low frequency of occurrence in the adult language.”

34

Donka Minkova

͡ consonant; thus [ti:ʔtʃɪŋ] ‘teaching’ identifies the second component of the affricate as an autonomous consonant. Also, the [t] before [ ʃ ] can be fully replaced by the glottal stop, as in [kaʊʔʃ ] ‘couch’. (4c) addresses a very familiar pattern in English morpho-phonology: affricate-final stems pattern with [ ʃ ]- and [ʒ]-final stems in inserting schwa before plural and third person singular present tense forms: bats vs. bashes, batches; pads vs. barrages, badges. The feature responsible for this alignment is [+strident]: an acoustic feature bundling together sibilants, but not stops. A further indication of the autonomy of the components is the behavior of affricates when they appear in sequence as shown in (4d): ͡ ͡ much choice can be realized as [mʌʃ tʃɔɪs] and large jar as [lɑːʒ dʒɑː], but omis8 sion of the fricative element is unacceptable. Finally, there is the consideration that while the inventory of word-initial clusters in English is quite rich, it does not include affricates – there are no affricate + C or C + affricate initial clusters in English (Hammond 1999: 101; Cruttenden 2008: 254–257). In stem-final posi͡ cannot be followed by another consonant except for inflec͡ and /dʒ/ tion, /tʃ/ tional /t/ and /d/.9 Edge effects notwithstanding, the unitary treatment is the analysis adopted in the standard accounts of the Present-Day English phonological system. It fits the contour analysis of single-slot fillers with independently acting components, and provides a relatively clear endpoint in the evolution of these segments, to which their earlier history can be compared. One should be reminded, however, that the analysis is language specific: compare the treatment of PDE /tʃ/͡ and ͡ to Modern German, where the phonemic status of /tʃ/ and especially of /dʒ/ /dʒ/ is controversial, and the voiced segment is found only in loanwords.10 Among all the other Germanic languages, only the newer contact languages, Yiddish and Africaans, have palato-alveolar affricates. The complexities associated with the affricates in the modern language extend all the way back to Old English.

8 This comment appears under “Advice to foreign learners” in Cruttenden (2008: 188), from where the second example is cited. ͡ for [dr-] as in dry, dress, is a ͡ for [tr-], as in tree, true, and [dʒr] 9 It should be added that [tʃr-] common allophonic possibility. Note, however, that the “Priorities and Tolerances” section in Cruttenden (2008: 331) includes the advice that “/tʃ, dʒ/should be kept distinct from /tr, dr/”, though he also recognizes the acceptability of [tj, dj or [ç, ʝ]. 10 In some analyses of German, only [ts] and [pf] are fully phonemically contrastive affricates. Kohler (1990) treats them as allophones, while Prinz and Wiese (1991) argue that all stop-fricative combinations in German are potential phonological affricates. For a commentary on Prinz and Wiese, see Rákosi (2014).

35

From stop-fricative clusters to contour segments in Old English

3 The consonantal inventory of Old English By way of a reminder, the standard consonantal inventory of Old English is shown in Table 1: Table 1: The late Old English consonant system (Lass 1992 in CHEL 2, 41) Labial STOPS Obstruent

Lab-Den

Dental

p(:) b(:)

Palatal

t(:) d(:)

Velar k(:) g(:)

tʃ(:) dʒ(:)

AFFRICATES FRICATIVES

f(:)

NASALS Sonorant

Alveolar

θ(:)

m(:) Lateral

s(:)

ʃ

x(:)

n(:) l(:)

Approximants Central

w

r(:)

j

The shaded cells enclose tʃ and dʒ, explicitly identified as “the new phonemes” in Old English (Hogg 1992b: 107).11 Parentheses indicate the existence of phonemic singleton-geminate contrasts. In that reconstructed system, only /ʃ/, /j/, and /w/ have no geminate counterparts. The direct inference from the inventory is that except for length/gemination, the affricates have been the same for eleven centuries. In modern editions of Old English texts, in textbooks and many essential sources (e.g., Campbell 1959; Hogg 1992a; CHEL I; and CHEL II), but not in the Dictionary of Old English, the reconstructed voiceless affricate is written with an overdotted , the geminate with , and the reconstructed voiced affricate is written .12

3.1 Sources of affrication in Old English As noted in section 1, neither the western Indo-European languages nor ProtoGermanic languages had affricates in their consonantal systems, and of the older Germanic languages, only Old English and Frisian developed palato-

11 The phonemic status of the affricates in Old English has mostly been considered a “closed” matter since at least Kuhn (1970), who posits contrastive affricates /tʃ/ and /dʒ/as far back as ca. 700 Mercian Old English. His phonemic slashes refer to “the allophone which occurs in initial position” (1970: 18). 12 This transcription will be preserved here only when it is relevant to the discussion.

36

Donka Minkova

alveolar affricates.13 The first orthographic indication of a difference between a stop and an assibilated allophone for the alveolar stops [t] and [d] are spellings appearing in the second half of the ninth century.14 (5) Early orthographic indication of assibilation: (ge)fetian ~ (ge)feccan ~ (ge)fecgan ‘fetch’15 ort-geard (×1)

~ orce(a)rd (×17) ‘orchard’

cræftiga

~ cræftca (ÆGl) , cræftigena ~ cræftcena ‘workman’16 (ÆGram MSS DFHhR)

micgern < *mid-gern ‘fat’ (Campbell 1959: 176) (×6, glosses only) bryttian

~ brycian (Bede MS CaO) ‘to distribute’ (a single attestation)

The first three examples are cited in Luick (1914–1940: §667). The last form, bryttian, recorded in the Dictionary of Old English, is a single attestation. The Dictionary of Old English is only up to (January 2016), and “wild card” searches don’t show alternative consonant spellings, so maybe there are some examples that have remained undiscovered. However, nineteenth and twentieth century researchers were very thorough, and it seems likely that fetch and orchard are indeed isolated cases of surviving Old English words in the PresentDay English vocabulary that show affrication of dental stops + /j/.17 13 Old High German developed the voiceless affricates [pf], [ts], and [kx]; see Robinson (1992: 233, 240). The areal restriction of the Old English affricates within older Germanic might suggest that language contact may be at play (i.e., Celtic), but the only affricate tentatively reconstructed for Celtic is /ts/ (Watkins 1955). 14 Assibilation is a term widely used in the literature for the creation of affricates from nonsibilant stops. It therefore includes all contour segments, not just the ones posited for PresentDay English. On the potentially misleading use of assibilation as a separate stage between palatalization and affrication, see CoNE (under VP = Velar palatalization): “Cross-linguistically, palatal stops do commonly have affricated release, which would seem to make an extra change unnecessary; but such releases are palatal, not palato-alveolar or alveolar, which would be required for the output to be called ‘sibilant’”. This is, unfortunately, not testable in Old English. I am keeping the term because it is common in the literature (e.g., Hogg 1992a), and because the sibilant association of the Present-Day English affricates seems beyond dispute. 15 The past tense forms ( fette, fetod, fetedon) are attested only on the basis of earlier fetian. 16 The evidential value of this item may be disputed; see the comments in Hogg (1992a: 271, n.2). 17 It should be noted that the variant form fetian is found in late Old English; for example, ChronD (C): 1017.8: & þa toforan kalendas Augusti het se cyng feccean him [. . .] (EF feccan, C fetian). Commenting on the history of the Present-Day English fetch, the Oxford English Dictionary editors note that “no other instance is known in which the change of ti into cc (=/tʃ/) has occurred”.

From stop-fricative clusters to contour segments in Old English

37

A more sustained source of Old English forms that developed affricates is the assibilation of velar stops subject to West Germanic Gemination, triggered by /j/, as in (6): (6)

West Germanic Gemination of velars and assibilation: PrG *þakjan OE þecc(e)an ‘cover’ PrG *klukjan OE clycc(e)an ‘clutch’ PrG *hakja OE hæc(c) ‘hatch, gate’ Goth. bugjan OE bycgan ‘buy’ PrG *xrugjaz OE hrycg ‘ridge’ PrG *agja OE ecg(g) ‘edge’

The propensity of Germanic velar stops towards gemination is well-known; gemination is shared by all West Germanic languages, but in North Germanic, only the velars were affected by gemination (Robinson 1992: 250).18 In Old English, West Germanic Gemination affects consonants (other than /r/) positioned after a short vowel and before an original /j/ – before it is dropped, the palatal glide triggers anticipatory palatal assimilation and the palatal umlaut of the short vowel. In terms of syllabification, the geminates straddle the syllable boundary, making the stop to the right an onset; at an early stage of the process, the stop and the palatal are tautosyllabic. Another environment in which stem-final velars behave in a parallel fashion is after a nasal, as in drench and singe in (7): (7)

Post-nasal velars and assibilation: PrG *drankjan OE Comp. PrG *drinkan OE Comp.

PrG Goth.

*sangjan singwan

OE OE

drenċan ‘cause to drink, drench’ drincan ‘to drink’19 senġan ‘singe’ singan ‘sing’

18 Denton (1998: 221) writes: “The velars show a strange affinity for gemination,. [. . .] [V]oiceless velars were the only segments geminated by all of the glides in West Germanic. [. . .] [T]here is good reason to believe that if there was a clear progression of gemination in its early stages, voiceless velars were the first segments to be affected”. 19 The Dictionary of Old English Corpus has instances of drinccan ‘to drink’, also drunccon, drunccende, but they are ambiguous. Тhe noun drenc ‘drink’ (Old Germanic *draŋki-z) appears as drencg (BenRGl, Mem), drengc (PsGlF), drenhc (CollGl). There are no spellings for ‘drench’. The earliest spelling for bench < Pr.G *banki-z is thirteenth century.

38

Donka Minkova

The phonetic substance and the phonological status of such items is untestable in the verse; /n/ + any consonant sequence is not a possible word-initial sequence and will therefore always be separated by a syllable boundary within the word, making the resolution test inapplicable. It is only consistent Middle English (ME) spellings that provide more reliable evidence that affrication has occurred. A third source involving input velar stops is affrication in specific palatal/ front environments, as in (8): (8)

Palatalization of velar singletons: (Campbell 1959: 174–175):20 a.

Input V [+front] C V [+front]

*/k/ dīċes ‘ditch, gen.sg.’

*/ɣ/ dæġes ‘day, gen. sg.’

b.

V (I-Umlaut) C (V)

lǣċe ‘leech’

drȳġe ‘dry’

In this case the results for the voiced and the voiceless inputs diverge; in the voiceless set, the input palatal stop resulted in an affricate regularly after /i/ and umlauted vowels, as in (8a) and (8b). The voiced velar, which had remained a fricative in this position, was lenited to a palatal glide.21 The */k/ results are context-specific; an adjacent back vowel or a consonant blocks the assibilation. One therefore finds paradigmatic alternations that can affect the PresentDay English outcomes, as in (be)seech < OEngl. sēċan vs. seek < sēcþ, third pers. sg.; stitch, n. ‘a stab’ OEngl. stiċe < *stiki-z; stick, v. OEngl. stician, p.t. sticode, < *stik-, the root of ‘to pierce’. Also, the results are often influenced by competing Old Norse forms, where no palatalization occurred, so that we get the well-known pairs such as kirk-church, brig-bridge, birk-birch, and so on.22

20 Representing the input PrG voiced velar singleton orthographically as in the sets in (6) and (7) is a shortcut; in both instances the input is a voiced velar fricative that remains a fricative in all positions other than in gemination, as in (6), and after [n] as in (7); see Prokosch (1939: §31). The different proposals on the paths of affrication in (9) reflect the uncertainties regarding the exact manner of articulation of the velar inputs. 21 Additionally Pr.G */ɣ/ was palatalized in the environment V [+front] C + syllabic sonorant, as in næġl ‘nail’, reġn ‘rain’, fæġr ‘fair’. The most extensive coverage of Old English palatalization is found in Hogg (1992a: 258–266). On the phonemic split of */ɣ/ after the mid-tenth century, see Minkova (2014: 82–88). 22 See Luick (1914–1940: § 690) for more instances of paradigmatic leveling of forms; see also Campbell (1959: 177), who is more emphatic about the importance of the Scandinavian element in the /k/ forms.

From stop-fricative clusters to contour segments in Old English

39

3.2 The mechanism(s) of affrication The unifying initial step in the process of affrication is clearly palatalization, which has to be dated after Anglo-Frisian fronting of /a/ to /æ/ but prior to IMutation at the end of fourth century and the beginning of the fifth century (Luick 1940: § 637; Hogg 1979: 120–122; Hogg 1992a: 267–269). In the account proposed by Campbell (1959: 176), the development of the palatal stops [c] and (possibly – he does not spell it out) the palatal fricative [ʝ] went through [tj] and [dj] as an intermediate stage, merging with the palatalized dentals in (5), as in fetian ~ feccan ‘fetch’, ortgeard ~ orceard ‘orchard’. Lass and Anderson, who provide a detailed rule-based account of the palatalizations of the velars in Old English, forgo a gradual formulation because of lack of synchronic justification for it (1975: 132), and instead assume a synchronic one-step shift from a palatal stop to an affricate “whatever the actual historical process was”. They do, however, allow for gradualness in their representation of the development of the historical geminates (Lass and Anderson 1975: 147; see also Lass and Laing 2013: 91); the latter point out that “[n]one of the major handbooks treat this as a single change”.23 Hogg (1992a: 267–270) assumes a gradual change involving palatalization of the velar stops. Those three paths overlap fully only in the endpoints. The proposed pathways are summarized in (9): (9) Paths of affrication:24 a. [*k(k)] ? [ɣ]

> [c(c)] > [tj] > [tʃ ] (Campbell 1959: 176) > [ ʝ( ʝ)] > [dj] > [dʒ]

b. [c(c)] [ɟ(ɟ)]

> [tʃ ] (Lass and Anderson 1975: 132) > [dʒ]

c. [c(c)] > [cʲ] ? [*g(g)] > [ɟʲ]

> [tʃ ] (Hogg 1992b: 270) > [dʒ]

23 Note also the Corpus of Narrative Etymologies discussion: “[W]e take the palatalisation + assibilation cluster as ‘unitary’ for the sake of etymologies, but with the background assumption that there was at least one ‘intermediate stage’ that we need to reconstruct”; see http:// archive.ling.ed.ac.uk/ihd/cone_scripts/view_CCchangeC2.php?chabbr=%28%28VP%29% 29&prntopt=no 24 [c] is a voiceless palatal stop; [ɟ] “barred dotless j” is a voiced palatal stop; [ ʝ] “curly-tail j” is a voiced palatal fricative.

40

Donka Minkova

In terms of articulation, the process starts with fronting: “The more front the vowel, the more front the velar” (Keating and Lahiri 1993: 89). The articulatory similarity between velars and the palato-alveolars before front vowels is also the basis of acoustic similarity.25 Since there is no orthography-based evidence for the intermediate stages, and in light of the synchronic phonetic facts, any path in (9) is plausible; they all start and end at the “right” place. The least secure reconstruction is of the nature of the voiced velar input: stop or fricative, though voiced velar fricatives figure in all proposals. All three accounts assume, explicitly or not, a parallel shift to affricates in voiced and voiceless inputs, at least post-nasally and in geminates. The triggering mechanism, front vowels, is no mystery. But there are other loose ends in the account. The first one is dating in relation to structural position. Affrication in stressed syllable onsets was a relatively late phenomenon. The evidence for this is the continuing identification of palatalized and non-palatalized voiceless velars in alliteration in the entire poetic corpus, irrespective of date of composition, as illustrated in (10):26 (10) Identification of palatalized and non-palatalized voiceless velars in alliteration: cynedom ‘kingdom’ : ciosan ‘choose’ acennedne ‘born’

Beowulf 2376 (ca. 725)

: cildes had ‘childhood’ Guthlac B 1361

cohhetan ‘to cough’ : cirman ‘chirm’

Judith 27027

: Ċeolan ‘Ceola’

Maldon 7628

: cildhade ‘childhood’

Durham 16 (ca. 1100, emended Holthausen)

cafne . . . ‘nimble’

cynne ‘kin’

clene Cudberte ‘pure Cuthbert’

25 Guion (1998) offers a good discussion of the issue; see also Wilson (2006) for a survey of the literature on acoustic similarity. Of significance is that Guion (1998) also found that the voiceless velars and palato-alveolars were more similar than the voiced consonants, a result which is in line with the affrications in Old English: “[T]here were more [k]/[tʃ ] than [g]/[dʒ] confusions. About 15% of the [g] tokens were heard as [dʒ], whereas around 26% of the [k] tokens were heard as [tʃ ]” (Guion 1998: 35). 26 For an extensive discussion of the voiceless velar alliteration in Old English, see Minkova (2003: 3.2, 3.5). 27 Detailed arguments for the tenth century origin of the poem (ca. 930–937) are presented in Timmer (1978: 6–11). 28 cafne mid his cynne, þæt wæs Ċeolan sunu ‘vigorous as his kin, he was son of Ċeola’ (Mald 76). The notation on Ċeola is used in all editions that mark affricates in that way (e.g., Pope-Fulk [2001]).

From stop-fricative clusters to contour segments in Old English

41

Another consideration is that if the consonant was the affricate [tʃ-], one should expect some indication of at least initial bisegmental treatment, so that cild would alliterate with *[til-], for example. But there are no such instances in the Old English poetic corpus. Negative evidence is not very useful, but for comparison, note in (11) that the practice of the fourteenth-century Middle English alliterative poets allows for occasional [t-] : [tʃ-] pairing: (11) Compositional treatment of [tʃ-] in Middle English alliteration:29 time . . . he tok telle and teche chese chiftanis

: : : :

child charite turnen twin

William of Palerne 4674 Piers Plowman C XIX. 2 Chevelere Assigne 357 The Scottish Prophecy 122

The voiced affricate did not appear stem-initially until after the Conquest. The simplification of the Old French affricate /dʒ/ to /ʒ/ was in progress in the thirteenth century in Continental Old French, but the parallel change appears to have been delayed in Anglo-Norman (Pope 1961: 93–94, 450). That lack of symmetry is recognized in the standard accounts, yet it does not affect the phonemic reconstructions discussed above; phonemic status is compatible with distributional restrictions. However, if onset affricates are not part of the phonological system, how solid is the assumption regarding medial affricates? Is the medial voiced affricate, always derived from a CC sequence, treated as a unit or as a sequence? There are no reliable tests for the affrication word finally, but the stages of intervocalic affrication are testable with reference to Old English meter.

4 Singleton and/or geminate (pre-)affricates in Old English? Before we turn to the metrical evidence for affrication, we need to address one more issue: the existence and stability of phonological geminates in Old English. As shown in Table 1, geminate affricates are posited in the descriptions of the Old English consonantal system. The minimal pairs in (12) show that Old English singletons and geminates could be contrastive word-medially:

29 The examples are cited in Schumacher (1914: 155). As rightly pointed out by an observant reviewer, the Piers Plowman example is ambiguous because charite could be stressed on the final syllable. The Scottish Prophecy is dated to the first quarter of the fifteenth century. For the phonetic nature of the identification, see Hardcastle et al. (1995): “[T]he place of articulation of /t∫/ can be predicted from that of independent /t/”.

42

Donka Minkova

(12) Singletons and geminates in Old English: bitela ‘beetle’ bite ‘bit, morsel, cut’ cyle ‘chill, fever’ hopian ‘to hope’ reċe ‘narrate!’ ͡ ? /dʒ/

bitter ‘bitter’ bitte ‘bucket’ cyll(e), cyllan ‘wineskin’ hoppian ‘to hop’ reċċe ‘narrate, present subj. sg.’30 ͡ ??/ddʒ/

The assumption in the textbooks is that with the exception of the approximants /w/ and /j/, all Old English non-initial consonants could appear as either singletons or geminates.31 Specifically for the affricates, Campbell (1959: 175) assumes a (non-dated) merger of ċ and ċċ into [tʃ͡ ] and of ġ (after a nasal consonant) and ͡ Weɫna’s (1986) extensive survey of the various positions on the ċġ into [dʒ]. dating and the question of singleton vs. geminate affricates concludes that both short and long affricates were present in Old English word-medially, but not word-finally (1986: 761). Hogg (1992a: 36–37) posits a stable singleton-geminate ͡ using the example reċe ‘narrate!’ – reċċe ‘narrate, present subj. contrast for /tʃ/, ͡ though he recognizes the doubtful status of sg.’ cited in (12), and also for /dʒ/, ͡ singleton vs. geminate status. He retains the voiced geminate affricate the /dʒ/ “for the sake of clarity [. . .] without distributional justification” (Hogg 1992a: 37). Since these positions are contradictory – and, as far as I know, no study of affrication in English draws on arguments from the behavior of the relevant segments with respect to syllabification, I checked their treatment in verse.

5 The metrical treatment of assibilated velars in Old English verse Spelling and general theoretical considerations make it plausible that the palatalized velars in coda positions (e.g., dic ‘ditch’, pic ‘pitch’) had fully palatalized

30 Cited in Kuhn (1970: 49); Hogg (1992a: 37). 31 However, see the discussion of the geminates in section 3, Table 1, where the palatal sibilant /ʃ/ is also listed as appearing only as a singleton; similarly Lass (1992: 60), but not Hogg (1992a: § 7.37). Positing a geminate [ ʃʃ ] (or something like [ssç]) even for late Old English is questionable at best; there is no way of showing that it existed, since there is no possible evidence from resolution – the digraph is treated as a cluster; see Hutcheson (1991: 52); see also the examples in (21) in section 7. I have not been able to find an example of a potential geminate [ ʃʃ ]/[ssç] contrasting with a singleton.

From stop-fricative clusters to contour segments in Old English

43

surface realizations; the voiceless velars in that position would have been phonetically very similar to the realization of the dental stops +/j/ (e.g., ‘fetch’) by the beginning of the ninth century, but there is no independent justification for that reconstruction other than hindsight based on the Present-Day English pronunciation. The status of the onset pre-affricates was already covered, so now we turn to , and . Establishing the status of medial “future” affricates is not a straightforward exercise. One evidential source that has not been brought into the discussion of the early history of the affricates is their treatment in verse. Hutcheson (1991: 51– 52) identified the problem, comparing Old English with classical Greek, where “the affricates are treated metrically as double consonants”. He assumed that word-medial /tʃ/ in Old English functioned as a single phoneme, but he cited no evidence supporting the assumption and he did not comment on the status of the voiced affricate. Sections 5.1–5.3 try to fill the gap of our empirical knowledge by presenting data on the behavior of inherited intervocalic velar stops with respect to resolution.

5.1 Singleton and the evidence of resolution32 Most of the potential Old English inputs to the Present-Day English affricates go back to geminates, but in the case of the voiceless velar /k/, there are some cases of palatalized and later affricated singletons (e.g., Present-Day English ditch, leech; see also (8)). The majority of the potential examples have a long vowel in the stressed syllable, which renders them ineligible for metrical testing. However, there are some lexical items which can be used as test cases for the treatment of singleton /k/ in palatal environments. In such items the stressed syllable and the following syllable fill a single metrical position (i.e., there is metrical resolution). Without the application of resolution, the verses become unmetrical (i.e., they violate the norm of four positions per verse).33 32 Old English resolution can affect only stressed open syllables with a short vowel in the peak, that is, (C)V-syllables. The metrical treatment of such syllables allows the equivalence of (C)V + any syllable to a single heavy syllable – (C)VV(C), (C)VC(C) syllables, thus metrically the whole word my.cel ‘big, much’ could fill a single strong/ictic verse position in the same way as first syllable in mē.tan ‘to meet’, or the first syllable in win.tra ‘of winters’ function in the meter. 33 Adjacent weak syllables unaffected by resolution (i.e., unattached to a stressed light syllable) count as a single weak (non-ictic) metrical position. Non-ictic positions can, therefore, accommodate either a single syllable or a string of weak syllables. Strong, or ictic positions, can be filled only by a stressed heavy syllable or a resolved sequence of a light syllable + any syllable.

44

Donka Minkova

The treatment of some common lexical items in the verse is shown in (13):34 (13)

The status of in reced ‘hall’, mycel ‘big, much’, bryce (1) ‘breaking’,35 (2) ‘use’: æfter recede wlāt wið þæs recedes weal on þam micelan bēc and þin micele miht for micelnysse þæt he micel āge wudafæstern micel ne sy him bānes bryce ond him bryce heoldon

(w w S-w w s) (w w S-w w s) (w w S-w w s) (w w S-w w s) (w S-w s w) (w w S-w s w) (S-w s w s-w) (w w w S w S-w) (w w S-w s w)

Beo 1572b Beo 326b36 SnS 6a Lord’s Pr 2:33a37 Judgment Day II 186b38 Exhortation 38b Durham 6b39 GuthA 698a ‘breaking’ GuthA 729b ‘did service’40

A close inspection of the metrical use of these items shows that the intervocalic velar represented by orthographic is consistently treated as a single consonant. The is aligned with any other singleton in the system, though its exact phonetic nature is irretrievable; typologically in this position it was most likely a voiceless palatal stop [c]. While its phonetic value is a matter of conjecture based on our current knowledge of phonetics, its systemic-phonemic value is unambiguous: it is functionally a singleton. Moreover, the singleton in

34 All items discussed in this section are marked with an overdotted in the Word Indexes in Campbell (1959), Hogg (1992a), or both. Only the scansion of the relevant verses is shown here. The notation is as follows: w = weak syllable, S = stressed and alliterating syllable, S-w = resolved light syllable, s = stressed unalliterating syllable. 35 The Oxford English Dictionary identifies the etymology of Present-Day English breach as “Middle English breche, partly perhaps repr. Old English bryce, brice . . . partly < French brèche”. The long vowel is due to open-syllable lengthening in the base brek-. 36 Other instances of reced ‘hall’ resolved: receda under roderum (S-w w w w s-w w) Beo 310a; singan on ræcede (s w w S-w w) Rid 31:3b; and his recedes hleow (w w S-w w s) Gen 2443b; Loth on recede (S w s-w w) Gen 2463a; rum recedes muð (S S-w w s) Max II 37a, see also Gen 1584a, Beo 720a, Beo 728a. 37 Same line in Gloria 1:32a. 38 Judgment Day II, as well as Exhortation to Christian Living are “demonstrably late” (Fulk 1992: 264). 39 Ca. 1100. 40 The headword bryce has three separate entries in the Dictionary of Old English, all with a short stressed vowel. The line GuthA 729b has also been interpreted as ‘they kept injury from him’ (DOE, under bryce2). The other attestation of bryce allows ambiguity: bryce on feorweg ‘breakable, fragile’ PPs 119.5b. Another item showing resolution is þecen ‘roof’ in anre þecene Riddle 84: 40b.

From stop-fricative clusters to contour segments in Old English

45

question is not a contour affricate yet. Unless we posit a saltatory change from the voiceless palatal velar [c] directly to [tʃ͡ ], which is unlikely because of the continuing bisegmental perception of in Middle English (as illustrated in the alliterative matching in (11)), the orthographic in palatalizing environments is still [c].

5.2 The medial voiceless geminate (pre)-affricate: ͡ ]? [t ʃ ], [tʃ tʃ ] or [tʃ͡ ] [ttʃ The geminate pre-affricate in intervocalic position was the result of West Germanic Gemination of the voiceless velar stop /k/, the first column of examples shown in (6). The first item I tested, (be)þeccan ‘cover, thatch’, qualified in the Dictionary of Old English as “disproportionately frequent in the verse”, is not recorded in any positions where resolution has to apply to maintain the four positions in the verse. This applies both to the verb and to the derived noun þeccend ‘protector’. (14) Unresolved in þeccan ‘cover’ and þeccend ‘protector’ in Old English verse:41 a. ǣled þeccean ‘flame to cover’

Beo 3015a

comp.: earne secgan ‘to eagle tell’

Beo 3026a (S w s w) Type A

also:

Beo 3048a

discas lāgon ‘dishes lay’

b. þu eart þeccend mīn Psalm70: 16b comp.: under gēapne hrōf Beo 836b ‘you are my protector’ ‘under gaping roof’ (w w S w s) Type B also:

æt his selfes hām ‘at his own home’

or:

under heofones Beo 2015a hwealf ‘under heaven’s arch’

Beo 1147b

Verse-finally, as in (14a), the placement of the verb is uninformative because of Kaluza’s Law, whereby resolution is suspended if the post-stressed syllable is heavy. However, in non-final positions one would expect the syllable in the first strong position in (14b), filled by the word protector, to be either heavy, as in 41 The notation for this word in sources that mark affrication with over-dotting is consistently .

46

Donka Minkova

the metrical analogs gēapne ‘gaping’ in Beo 836b, selfes ‘of self’ in 1147b, or resolved (or syncopated), as in heofones in Beo 2015a. The syllable containing a short vowel followed by is treated as heavy/closed (i.e., the realization is bisegmental). Therefore, (14b) is an indication of the continuing treatment of the historical geminate as such, and not as a contour segment. This precludes [tʃ͡ ], leaving [cj] or [tj] as most likely, but other bisegmental realizations such as [cc], [t ʃ ], or even [tʃ tʃ ] cannot be ruled out. The results from another potential candidate, wæccende ‘watching’, are shown in (15): (15) Unresolved in wæccan – wæccende ‘watching’ a. þæt ge wæccende ‘that he watching’

b. wæccendne wer ‘watching warrior’

Jul 662b

comp.: þæt he Hrōþgāres ‘that he Hrothgar’s’

Beo 717a42 (w w S s w) Type C

also:

Beo 1135a

þā ðe syngāles ‘those that always’ æt ðam æðelinge ‘in that nobleman’

Beo 1268a comp.: sārigne sang ‘sorrowful song’ also:

c. weras wæccende Jud 142a ‘warriors watching’

murnende mōd ‘mourning mind’ lifigende lāð ‘living loathed’

comp.: sunu Ecglāfes ‘son of Ecglaf’ also:

Beo 2374a with resolution Beo 2447a (S s w s) Type E Beo 50a Beo 815a with resolution Beo 590b (S-w S s w) Type D

gifen gēotende Beo 1690a ‘ocean gushing’ hwatum Heorowearde Beo 2161a ‘to bold Heoroweard’ with two resolutions

In (15a), a Type C, the stressed syllable of watching is unresolved, treated in the same way as Hrōþ- in Beo 717a or syn- in Beo 1135a. In (15b), a Type E, the parallel between watching in the first column and sorrowful, mourning, and living in the comparison column suggests that the first syllable is treated as heavy, blocking resolution. In (15c), a Type D, the same parallels apply. In the last items in (a), (b), and (c), which show resolution in the comparable metrical positions, the resolvable syllable is light.

42 Similarly, ‘ac he wæccende’ Beo 708a; ‘gif he wæccende’ Beo 2841a.

From stop-fricative clusters to contour segments in Old English

47

For comparison, the metrical behavior of fetch, to the limited extent that it is testable in the verse, also suggests non-resolution (i.e., heavy stressed syllable), as shown in (16). (16) Treatment of (ge) feccan ‘fetch’:43 līfe gefecce Dr R 138a44 comp.: gūþe gebēodan Beo 603a (S w w s w) bēagas gefeccan Maldon 160a also: ganges getwǣman Beo 968a

It is of interest to compare the metrical treatment of the variant form fetian. In Beo 2190b (Hēt ðā eorla hlēo) / in gefetian ‘in to fetch’ (S w s-w w), resolution must apply (vowel alliteration). The use of the fetigan variant in a late poem, Judith 35b: ofstum fetigan ‘in haste to fetch’ (S w s-w w), also requires resolution. There is nothing surprising about the variable pronunciation of one and the same lexical item synchronically; the only reason to bring up these parallel examples is to point out that the coexistence of the fetian variant might have been an inhibiting factor in the reanalysis of as a sequence of two fullfledged affricates.45 The alignment of the first syllables of words such as þeccan, wæccan, feccan with heavy syllables or resolved stressed syllables in a metrical system that requires rigid onset-maximal syllabification with respect to resolution confirms the continuing bisegmental treatment of the medial voiceless palatals derived from West Germanic Gemination. The exact phonic content of the components of the sequence is beyond recovery, but the parallels between (13) and (14–15) suggest that the very first stage proposed in all three accounts in (9) – [c(c)] – is correct. What the contrast between (13) and (14–15) highlights is that an assumption of a merger of palatalized etymological voiceless velar singletons

43 There may be some ambiguity in the scansion of gefeccan under foldan, Solomon and Saturn 75a: (x) S w w w s w, Type A with anacrusis marked (x), no resolution, or (x) S-w w w s w – Type A with resolution (possible because of the competing form fetian, but not obligatory). Both types are rare; Hutcheson (1995: 194) has thirty-three cases of Type A a-verses with anacrusis and double alliteration in the corpus vs. twenty-one cases in Type A a-verse double alliteration with resolution. Similarly, gefetian on fultum Elene 1052a could be construed ambiguous: (x) S w w w s w, or (x) S-w w s w though the second scansion seems the more likely one. 44 Cited from Eight Old English Poems, Pope-Fulk (2001); ms. ge-fetige, which would require resolution. Admittedly, therefore, the evidence from this example is inconclusive. As a reviewer points out, “[S]ince this poem is attested also on the Ruthwell Cross, it is far from unlikely that gefecce in the Exeter Book is a late scribal Saxonization of the earlier form”. 45 The two variants exist side by side in early Middle English too, though the affricated form appears to be gaining ground by the beginning of the fourteenth century. LAEME lists eightyone tokens of fetch where the consonant is intervocalic – the spelling is with in thirtyfour instances and forty-seven spellings.

48

Donka Minkova

and etymological voiceless velar geminates, as suggested by Campbell (1959: 175), cannot be sustained.

͡ 5.3 The medial voiced affricate: [ɟʝ], [dj], [dʒ], [dʒ], ͡ [dʒ dʒ] or [ddʒ]? ͡ This section looks at the treatment of the precursor of Present-Day English /dʒ/, Old English orthographic in intervocalic position in the meter. Recall from section 3 that the only systematically attested Old English source of the PresentDay English intervocalic voiced affricate was West Germanic Gemination. The singleton voiced velar */ɣ/ is palatalized to [j] when adjacent to palatal vowels, as in drȳġe ‘dry’ as in (8), so the parallelism to [k] > [c] is lost. The development of the voiced velar after a nasal, as in PrG *sangjan > OE sengan ‘singe’ in (7), is not testable in the verse because the stressed syllable is always heavy. That leaves us with the intervocalic geminates due to West Germanic Gemination, the type shown in (6) as in Goth. bugjan > OEngl. bycgan ‘buy’. The treatment of such words in the verse is unsurprisingly consistent. For the sake of the philological record, in (17) I list all items and all instances with a medial in Beowulf. Where possible and relevant, the words are shown in a metrical context. (17) words in Beowulf: a. Verbs and verbal forms bicgan bicgan scoldon 1305b ‘to buy’ licg(e)an lāðne licgean 3040a ‘lie’ also 966a, 1427b, 1586b, 2886a, 3082a, 3129a secgan secgan hȳrdon 273b ‘to say’ also 51a, 344a, 391a, 411a, 473a, 582b, 590a, 875b, 880b, 942b, 1049a, 1346b, 1700a, 1724b, 1818b, 1997b, 2795b, 2864a, 3026a, 3028b forhycgan 435a ‘reject’ also hicgend-: heardhicgende ‘hard-minded’ 799a, also 394a, 919a, 1016a, 2235a also bealohycgendra ‘intending evil’ 2565a, 2716b (ge)fricgan fægre fricgcean 1985а ‘ask, inquire’ also 1826a, 2889a, 3002a also felafricgende ‘well-informed” 2106a ðicgean ofer þā niht 736a ðicgean ‘devour’ also 1010b

From stop-fricative clusters to contour segments in Old English

49

b. Inflected forms of nouns ecg ecge cūðe 1145b ‘edge’ also 483b, 805a, 1168a, 1287a, 1549a, 1558b, 1772a, 1812a, 2140, 2485a, 2564a, 2614a, 2683b, 2828b, 2876a, 2939b, 2961b mæcg Ōretmecgas 332a, ‘warrior’ also 363b, 481b, 491a, 799b, 829a, 2379b secg Forðam secgum wearð 149b ‘man’ also 213b, 490a, 633b, 684a, 842b, 996a, 1672b, 2019a, 2530a, 3128a wicge wicge rīdan 234b ‘horse’ also 286b, 1045a All of these items scan as two-positional, no resolution. The date of the poetic compositions does not have an effect on the treatment of . In the later verse, intervocalic continued to block resolution; it continued to be perceived as bisegmental, as is evident from the examples in (18): (18) Unresolved words in the later verse: sweordes ecgum, þæs þe us secgað bec meca ecgum Meters 9:29b secga swate on þa briċge stop Maldon 78b Ic eow secgan mæg stiðhicgende Maldon 122b niðhycgende wið þas secgas feaht Maldon 298b licgan þence

Brun 6846 Brun 13a Judith 152b Judith 233a Maldon 319b

In concluding this section: the consonant sequence derived from West Germanic Gemination of the voiced velar is structurally the same as any other consonant sequence which does not appear word initially. As for the phonetic nature of ͡ in intervocalic position, positorthographic , since there is no singleton [dʒ] ͡ ing a geminate /ddʒ/ in that position would make for an unusual phonemic inventory.47 Any other sequence which will straddle the syllable boundary and make the first syllable heavy, such as [ɟʝ], [dj], [dʒ], [dʒ dʒ], cannot be ruled out. 46 This text has survived in four copies, the earliest of which is 937. A very detailed comparison of the copies shows no variation in the metrical treatment of ; see Orton (2000: 57–59). 47 A reviewer points out that intervocalic survives in Old English, though the singleton had been lost in that position (though note Northern eher ‘ear’, tæher ‘tear’), and that blocks resolution. This is a question of implicational universals, and I am doubtful if it can be tested reliably on the material we have available. Blevins cites one case (Finnish) as having a geminate engma, but no singleton, but she highlights it as unexpected; see httpsː//www.eva. mpg.de/lingua/conference/08_springschool/pdf/course_materials/blevins_evening_lecture.pdf (online only, as far as I know).

50

Donka Minkova

A rather flat and unsurprising conclusion at this point is that the difference between the examples in (13), (14–15), and (17–18) rests on a genuine quantitative difference between singletons and geminates in intervocalic position. As adumbrated in section 4, with the exception of Campbell (1959: 175), this is the accepted view in the scholarship on this question. However, the metrical attestations discussed in 5.1–5.3 still call for a reassessment of the phonetic and phonological nature of the sounds represented by , , and in Old English.

6 Trajectories of palatalization and assibilation of the Old English velar stops The alliterative practice illustrated in (10) (i.e., the sustained identification of [k] and the palatalized [c] in Old English in onset position) indicates that the two sounds were still most likely allophones of the voiceless velar stop. The degree of palatalization of [c] is unknowable; all we can say is that it was still classified as a voiceless velar stop, since Old English alliteration is consistently based on phonemic identity. What is more, the identification of the palatal pre-affricate with the voiceless velar stop continued into early Middle English: (19) Non-contrastive onset [k-] and [c-] in Lagamon’s Brut:48 Heo bigunnen to chiden; cnihtes come riden 4064 þa þet child wes iboren. wel wes Claudiene þer-foren; 4794 & ladde þes childes moder. for quene nauede he oðer; 4807 Such, arguably intuitive, reaction to the similarity of the relevant onsets suggests that a reconstruction of a singleton voiceless velar stop in initial position, possibly followed by [j], is preferable to reconstructing a singleton phonemic affricate. This contradicts Kuhn’s 1970 analysis, where it is precisely the initial-position allophone that is taken as the basis for positing a phonemic voiceless affricate in early Middle English.49

48 The examples are from the earlier manuscript Cotton Caligula A IX (C). The composition of the text is dated between 1189 and the first half of the thirteenth century; see Minkova (2003: 74–75). 49 It is likely that the late Old English alliterative tradition preserved the alliteration of all stressed syllables with a onset as an archaism, as one reviewer points out, yet it is also clear that the Middle English poets’ and scribes’ choices of alliterating pairs continued to be guided by phonetic similarity. See Schumacher’s (1914: 3) inventory of incomplete and

From stop-fricative clusters to contour segments in Old English

51

In medial position the reconstruction of a singleton voiceless velar stop continues to be the most plausible analysis, as evidenced by the metrical practice in late Old English verse; for example, wudafæstern micel (S-w s w s-w) in the ca. 1100 poem Durham (see section 5.1). Further, as the attestations in (11) show, the cluster that developed into a voiceless affricate in Present-Day English could still be perceived as compositional as late as the middle of the fourteenth century, with the first element occasionally identified as a voiceless dental stop. This leaves the coda position as the only position of potential realization of the palatalized voiceless velar as an affricate. The likelihood of such a realization cannot be rejected, nor can it be tested, especially in view of paradigm-internal variability with respect to palatalization and the different dialectal pathways of word-final Proto-Germanic /-k/, as very revealingly covered in Lass and Laing (2013).50 Since minimal pairs of the type kin-chin would contradict the alliterative evidence, and since affrication involves a bisegmental stage precluded by the metrical evidence (and suggested by the fourteenth-century examples in [11]), the new proposal here is that there was no phonemic singleton voiceless palatoalveolar affricate in the Old English consonantal system. This is a significant departure from the traditional descriptions of the Old English consonantal system.51 Turning to the historical voiceless velar geminates, the metrical evidence ͡ in section 5.2 tells us only that medial cannot stand for a singleton /tʃ/. Moreover, if the voiceless singleton in, for example, mycel ‘big, much’, bryce ͡ then positing a voiceless contour geminate /ttʃ/, ͡ ‘breaking’ is not a contour /tʃ/, as implied in Table 1, is not the right reconstruction. Though not theoretically impossible, syllabification of [ttʃ͡ ] as [t.tʃ͡ ] is questionable in the absence of questionable alliterations of singletons that differ in any other feature except voicing (p : b, d : t, g : k, etc.). The only pair that would differ in manner of articulation is [t] : [tʃ ], if the latter was a genuine affricate. There is also a remarkably long list of what Schumacher labels “eye alliteration” involving and (1914:163–170). That the matching was more than simply “scribal” is suggested by the much shorter list of Germanic [g] matched to Old French [dʒ] (1914:162). 50 An observation that is relevant but cannot be pursued here is that [t ʃ ] may be a factor in triggering shortening (e.g., Old English dīc ‘ditch’, līc ‘body’, compare dike, like). 51 An alternative, which I will not pursue, is that even if [tʃ͡ ] was positionally restricted to the coda, it was phonemic, and the non-affricated realizations elsewhere were allophonic. This is non-viable unless one reconstructs a bisegmental sequence [-t + C], but that is no longer a ͡ Finding relevant minimal pairs in Old English is also probphonemic contour segment /tʃ/. lematic; do we know that Old English pic ‘black resin, pitch’ and Old English pic, piic ‘pick’ were not homophones?

52

Donka Minkova

͡ and syllabification as [tt. ʃ ] is unlikely because of the instability word-initial /tʃ/, of coda geminates (Ladefoged and Maddieson 1996: 92–93). This leaves us with the option of reconstructing a bisegmental sequence: [cj]/[tj], [cc], [tʃ ] (see 5.2). Crucially, even if we posit [tʃ ] in that position, it remains a sequence and not a contrastive singleton.52 The status of the voiced counterpart is less clear because the only possible ͡ testable in the verse Old English source of the Present-Day English affricate /dʒ/ is gemination. As (17) and (18) show, the metrical treatment of the intervocalic geminates precludes unitary treatment. No further phonetic details can be reconstructed with certainty. The most likely reconstruction would be a voiced palatal stop [ɟ] + [+continuant], either [j] or [ʒ]. Note that the deliberately vague ͡ in word-final position, medial sequence does not automatically rule out a /dʒ/ but typological considerations would make it unlikely. In addition to the lack of contrast in intervocalic position, it is also unlikely that the system will have a voiced phonemic affricate unless there is a voiceless phonemic affricate. The use of spelling in Old English can be ambiguous – for example, ecgan ‘harrow’ < ecg ‘edge’ is spelled egede; egide, or for ‘flag(stone)’ (x1), Old Icelandic flaga, MEngl. flag(ge) – which also gives one pause. As for the geminate vs. singleton, the position taken here supports Kuhn’s (1970: 48–49) ͡ need not be postulated, contra the analysis, namely that a geminate /ddʒ/ “established” view. As in the case of the voiceless counterpart, this entails that a singleton phonemic geminate is also not viable.

7 Spelling and the evidence based on resolution The burden of proof, or at least the core new evidence for suggesting a revision of the Old English consonantal inventory, lies with the data on metrical resolution in sections 5.1–5.3. The geminates going back to the velar stops regularly block resolution, and they are also digraphs. One might ask therefore whether the metrical treatment could be a response to the orthographic shape of a word, unrelated to the phonological status of the medial consonant. The only solid argument against considering the number of letters as the guiding factor for resolution in the verse is that letter is treated consistently as bisegmental:

52 I leave the question of the later history of the word-medial geminate in Middle English open.

From stop-fricative clusters to contour segments in Old English

53

(20) Single letter blocking resolution in Old English verse: unweaxenne / wordum lærde Elene 529 ‘not grown-up / with words taught’ and he tofylleð / feaxes scadan Paris Psalter 67. 68 ‘and he breaks to pieces / the top of the head’ comp.: nu ic fitte gen / ymb fisca cynn ‘Now I sing again / of the fishes’ kin’

Whale 1

The presence of two consonant letters intervocalically appears to prevent resolution systematically, but a single consonant can have the same blocking effect. Disappointingly, a search for further possible clues proved unproductive. The Old English Dictionary Corpus database (http://tapor.library.utoronto.ca/ doecorpus/) shows no medial spellings for the resolvable items covered in section 5.1.53 The treatment of word-medial is untestable because the only eligible items I found, Stēphanus, Iōseph, have stressed long vowels. It may be worth noting in this context that the statistics of word division in Old English (Wetzel 1981: 466) indicate that and words in the manuscripts (8212 cases) are divided and in 99% of the cases.54 The palato-alveolar fricative [ ʃ ], commonly posited as part of the Old English contrastive singleton inventory, blocks resolution, as in (21). (21)

Non-resolution of in Old English: discas lāgon on þære ascan bið bisceop bremran

(S w s w) (w w w S w s) (S w s w)

Beo 3048a Phx 231a Men 104a

Once again, the most plausible reconstruction for medial is bisegmental. Hogg (1992a: 272–273) proposes a pathway of palatalization of /*sk/ via gradual assimilation of [s] and [c] > *[sç] > [ ʃ ]; “[I]t is necessary also to note that in the first instance the result of this shift was the geminate consonant [ ʃʃ ], which naturally always simplified in initial position and would also simplify finally. . . . But medially a geminate remained.” Hogg considers singletons and geminate [ ʃ ] to be in complementary distribution. Metathesis of ~ (e.g., axian ~ ascian ‘ask’) and spelling variants for (e.g., axe, acxe, acse, ahse, asce ‘ash’; 53 There are relatively few spellings for much, by far the most common item, in the DOE; out of the forty or spellings, none are in the verse corpus. 54 A line of research that is still open is the psycholinguistic status of contour segments (i.e., whether orthography has an effect on a speaker’s decision to treat phonetic signals as contour segments or as sequences of phonemes); see Hayes (2009: 68) and the references there.

54

Donka Minkova

waxan, wascan, awahxe, awhse ‘wash’), however, make it doubtful that the medial sequence was the geminate counterpart of a singleton [ ʃ ]. All one can assert is that in medial position the spelling did not represent a singleton. Thus the question posed in this section remains, not least because of the conservative nature of the Old English poetic texts. In the absence of more testable material, it is reasonable to stay with the traditional view of the linguistic rather than orthographic motivation for the consistency of Old English resolution.

8 Concluding remarks Table 1 showed late Old English reconstructed as having a fully developed set of affricates. Luick (1921–1940: § 687) dates affrication of [cʲ] > [tʃ ] to the second half of the ninth century on the basis of the earliest spellings of as ‘fetch’. Hogg (1992a: 272) offers a comprehensive survey of the earlier opinions and writes: “Of course a precise chronology can never be obtained, but general phonological principles would suggest an earlier rather than a later date for the development of affricates.” The data presented in section 5.1 challenge the reconstruction of an initial and intervocalic voiceless affricate based on a palatalized velar stop, and support a reconstruction of a continuing allophonic relation between the singleton [k] and a singleton palatalized stop [c] into the eleventh century or even later. The metrical data in sections 5.2 and 5.3 do not confirm or disconfirm Luick’s early dating for phonetic intervocalic assibilation or incipient affrication of the sequence represented by , yet what Hogg called “general phonological principles” suggest that a bisegmental realization in that position is uninformative ͡ The with respect to the existence of a singleton phonemic contour segment /tʃ/. ͡ is therefore a process that was still under way in early reanalysis of [c] to /tʃ/ Middle English. Phonetically it is plausible to assume a chain of input [k] > palatal [c] and bisegmental [cj] ~ [tj] ~ [tʃ ], and the possibility of a bisegmental [t] + [ ʃ ] analysis into the fourteenth century suggests that contour /tʃ/͡ is even later. Dating the affrication of the voiced velar is equally complex. Here again “general phonological principles” lead to a rejection of a phonologically con͡ in Old English.55 The dating of the full phonemicization of trastive contour /dʒ/

55 A singleton voiced velar affricate in final position would be in complementary distribution with its counterpart geminate in medial position; such distribution runs against their full phonemic status.

From stop-fricative clusters to contour segments in Old English

55

the voiced affricate cannot be posited independently of the phonemicization of ͡ first, because in coda position one would expect voicing neutralization, and /tʃ/: because in initial position, [dʒ] is the result of Old French loan phonology after the thirteenth century in words such as jay, jangle, jargon, juice, jealousy. The problematic nature of positing finite and invariable phonemic inventories based on written sources has been recognized for a long time. In 1930 van Lagenhove surveyed the literature on the assibilation of Old English palatal stops, pitched two venerable philological traditions against each other, and sided with Sweet, Wyld, Kaluza, and others in concluding that “[e]ven with regard to the XIth c. it seems to me that the indications at our present disposal are momentarily not sufficient to admit as an indisputable fact the beginnings of dentalization + assibilation of OE palatal stops” (van Langenhove 1930: 75). Then came the wave of strict phonemic interpretation, most explicitly articulated in Penzl (1947) and now firmly ensconced in the textbooks, within which the unpredictable velar /k/ before a palatal vowel due to I-Umlaut, as in cynn ‘kin’ alongside inherited palatalized /c/ in cinn ‘chin’ < PrG Germanic *kinnjo, compels a phonologization analysis. Bringing metrical arguments to bear on the history of the affricates in early English prompts a revision of the traditional strict phonemic categorization of the precursors of the Old English affricates. As Kiparsky (2014: 83) writes: “The classical phoneme has turned out to be something of a straitjacket and has not been helpful for understanding the rise and merger of phonological contrasts.” The patchy evidence, the apparent contradiction between the loss of conditioning environment for palatalization of the velars, and the stubborn continuing perception of the voiceless palatalized velars as sufficiently identical with the non-palatalized /k/ has been a conundrum in English historical phonology, and it seems a perfect case study for the long trajectory in the rise of phonological contrasts. Kiparsky (2014, 2015) proposes a more discriminating approach to phonemicization, separating the structural notion of contrastiveness from the perceptual notion of distinctiveness. In that schema the Old English palatalized voiceless velar singletons start out as non-contrastive and non-distinctive in initial and medial positions (i.e., they are allophones), but in final position and in gemination in late Old English they are closer to being perceptually distinctive, though not yet contrastive (i.e., quasi-phonemes). The reanalysis from quasi-phonemic to a full-fledged phonemic status is a Middle English process, possibly related to the loss of singleton-geminate contrasts and, for the voiced velar fricative, definitely related to the borrowing of new lexical items with initial /dʒ/. Departing from the established inventory of contrastive consonants in Old English, I propose an inventory of the late Old English consonants as shown in Table 2.

56

Donka Minkova

Table 2: The late Old English consonant system revised Labial STOPS Obstruent

Lab-Den

Dental

p(:) b(:)

Alveolar

Palatal

t(:) d(:)

Velar k(:) g(:)

AFFRICATES FRICATIVES

f(:)

NASALS Sonorant

m(:) Lateral

θ(:)

s(:)

?(ʃ)

x(:)

n(:) l(:)

Approximants Central

w

r(:)

j

Old English did have intervocalic geminate palatal stops, or pre-affricate bisegmental clusters, but it did not have contour affricate phonemes with the properties that these phonemes show in Present-Day English. For the conscientious or finicky teacher and reader of Old English, the [cj] ~ [tj] and [ɟj] ~ [dj] we articulate in reading Old English aloud are both phonetically and phonologically distinct from their Present-Day English counterparts. In a September 19, 2015 e-mail exchange on assibilation and affrication, Roger Lass wrote to me: “I think there are more clines and fuzzy spectra than there are Aristotelian boxes”. More boxes relating to the later history of the affricates remain locked. Contrary to the appealing idea that the voiced and the voiceless velar affricates developed in a parallel fashion, the two changes have to be treated separately – at least until after the entry of initial voiced affricates in Middle English, that is, after ca. 1250.56 The contribution of intervocalic affricates to syllable weight in ME is another open question. Not least, there is the puzzling problem of velars vs. dentals with respect to affrication: Why did PrG *sitjan, OEngl. sittan ‘sit’ not become *sitch (compare future, bet you)? Why did PrG *bidjan, OEngl. biddan ‘bid’ not become *bidge (compare soldier, bid you)? The lack of velar affrication and the blossoming of dental affrication in later English, the problem of procure [prəˈkju(ə)r] vs. mature [məˈtʃʊ(ə)r], is also on the long list of questions deserving further attention.57 On a broader scale, a more detailed picture of the phonemicization of affricates in English may contribute to the debate on what type of phonological change affrication represents, lenition or fortition.58

56 See Hogg’s skeptical comments (1992a: 260, n.3) on Luick’s and Campbell’s views on the parallelism between the changes of the voiced and the voiceless velar. 57 This is a problem identified by Liberman (2007, 2012), but his suggestion of merger avoidance does not entirely solve it, in my opinion. 58 Buizza and Plug (2012) discuss two mutually exclusive theoretical interpretations: plosive affrication is lenition vs. plosive affrication is fortition.

From stop-fricative clusters to contour segments in Old English

57

References Buizza, Emanuela & Leendert Plug. 2012. Lenition, fortition and the status of plosive affrication: The case of spontaneous RP English /t/. Phonology 29. 1–38. Campbell, Alistair. 1959. Old English Grammar. Oxford: Clarendon. CoNE: A Corpus of Narrative Etymologies from Proto-Old English to Early Middle English and accompanying Corpus of Changes. Roger Lass, Margaret Laing, Rhona Alcorn & Keith Williamson (compilers). Edinburgh, 2013–. http://www.lel.ed.ac.uk/ihd/CoNE/CoNE.html. Clements G. N. & E. V. Hume. 1995. The internal organization of speech sounds. In J. A. Goldsmith (ed.), The handbook of phonological theory, 245–306. Cambridge, MA: Blackwell. Cruttenden, Alan. 2008. Gimson’s pronunciation of English, 7th edn. London: Routledge. Denton, Jeannette Marshall. 1998. Phonetic perspectives on West Germanic consonant gemination. American Journal of Germanic Linguistics & Literatures 10 (2). 201–235. DOE: Dictionary of Old English: A to G online. 2007. Angus Cameron, Ashley Crandell Amos, Antonette diPaolo Healey et al. (eds.) Toronto: Dictionary of Old English Project. http:// www.doe.utoronto.ca. DOEC: Dictionary of Old English Web Corpus. 2009. Antonette diPaolo Healey with John Price Wilkin and Xin Xiang (compilers). Toronto: Dictionary of Old English Project. http://tapor. library.utoronto.ca/doecorpus/. Fulk, Robert D. 1992. A history of Old English meter. Philadelphia: University of Pennsylvania Press. Fromkin, Victoria A. 1971. The non-anomalous nature of anomalous utterances. In V. A. Fromkin (ed.), Speech errors as linguistic evidence, 215–269. The Hague: Mouton. Guion, Guignard Susan. 1998. The role of perception in the sound change of velar palatalization. Phonetica 55. 18–52. Hall, Tracy Alan. 2012. The representation of affricates in Cimbrian German. Journal of Germanic Linguistics 24 (1). 1–22. Hammond, Michael. 1999. The phonology of English: A prosodic optimality-theoretic approach. Oxford: Oxford University Press. Hardcastle W. J., F. Gibbon & J. M. Scobbie. 1995. Phonetic and phonological aspects of English affricate production in children with speech disorders. Phonetica 52. 242–250. Hayes, Bruce. 2009. Introductory phonology. Oxford: Wiley-Blackwell. Hogg, Richard. 1979. Old English palatalization. Transactions of the Philological Society 77 (1). 89–113. Hogg, Richard. 1992a. A grammar of Old English, Vol. I: Phonology. Oxford: Blackwell. Hogg, Richard. 1992b. Phonology and morphology. In Hogg, R. (ed.), The Cambridge history of the English Language I, the beginnings to 1066, 67–164. Cambridge: Cambridge University Press. Hutcheson, Bellenden Rand. 1991. Quantity in Old English poetry: A classical comparison. In Geardagum 12. 44–53. Jun, Jongho. 2011. Positional effects in consonant clusters. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Karen Rice (eds.), The Blackwell companion to phonology, 1103– 1123. Malden, MA & Oxford: Wiley-Blackwell. Keating, Patricia & A. Lahiri. 1993. Fronted velars, palatalized velars, and palatals. Phonetica 50. 73–101.

58

Donka Minkova

Kiparsky, Paul. 2014. New perspectives in historical linguistics. In Claire Bowern & Bethwyn Evans (eds.), The Routledge handbook of historical linguistics, 64–102. London and New York: Routledge. Kiparsky, Paul. 2015. Phonologization. In Patrick Honeybone & Joseph Salmons (eds.), The Oxford handbook of historical phonology, 563–582. Oxford: Oxford University Press. Kohler, Klaus 1990. German. Journal of the International Phonetic Association 20. 48–50. Krahe, Hans & Wolfgang Meid. 1969. Germanische Sprachwissenschaft Vol. 1. Berlin: Walter de Gruyter. Kuhn, Sherman M. 1970. On the consonantal phonemes of Old English. In James L. Rosier (ed.), Philological essays: Studies in Old and Middle English language and literature in honour of Herbert Dean Meritt, 16–49. The Hague: Mouton. Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s languages. Oxford: Blackwell. Lass, Roger. 1992. Phonology and morphology. In Norman Blake (ed.), The Cambridge history of the English language, vol. II: 1066–1476, 23–155. Cambridge: Cambridge University Press. Lass, Roger & John Anderson 1975. Old English phonology. Cambridge: Cambridge University Press. Lass, Roger & Laing Margaret. 2013. The early Middle English reflexes of Germanic *ik ‘I’: Unpacking the changes. Folia Linguistica Historica 34 (1). 93–114. Lavoie, Lisa. 2009. Testing consonant weakness phonetically. In D. Minkova (ed.), Phonological weakness in English: From Old to Present-Day English, 29–46. Houndmills: Palgrave Macmillan. LAEME: A linguistic atlas of Early Middle English, 1150–1325. 2007. Margaret Laing and Roger Lassv (compilers). University of Edinburgh. http://www.lel.ed.ac.uk/ihd/laeme1/laeme1. html. Liberman, Anatoly. 2007. Palatalized and velarized consonants in English against their Germanic background with special reference to i-umlaut. In C. M. Cain & G. Russom (eds.), Studies in the history of the English language III. Managing chaos: Strategies for identifying change in English, 5–36. Berlin: Mouton de Gruyter. Liberman, Anatoly. 2012. Sound change and distinctive features in light of dynamic synchrony. In Stephen J. Harris, Michael Moynihan & Sherrill Harbison (eds.), Vox Germanica: Essays in Germanic languages and literature in honor of James E. Cathey, 1–13. Medieval and Renaissance Texts and Studies 429. Tempe, AZ: ACMRS. Luick, Karl. 1914–1940 [1964]. Historische Grammatik der englischen Sprache, Vol 1, 2. Leipzig: Tauchnitz. Minkova, Donka. 2003. Alliteration and sound change in early English. Cambridge: Cambridge University Press. Minkova, Donka. 2014. A historical phonology of English. Edinburgh and New York: Edinburgh University Press and Oxford University Press. Orton, Peter. 2000. The transmission of Old English poetry. Turnhout, Belgium: Brepols. Penzl, Herbert. 1947. The phonemic split of Germanic k in Old English. Language 23. 34–42. Pope, John C., ed. 2001. Eight Old English poems, 3rd edn. Revised by R. D. Fulk. New York: Norton. Pope, Mildred. 1934/1961. From Latin to modern French with especial consideration of AngloNorman. Manchester: Manchester University Press. Prinz, M. & Richard Wiese. 1991. Die Affrikaten des Deutschen und ihre Verschriftung. Linguistische Berichte 133. 165–189.

From stop-fricative clusters to contour segments in Old English

59

Prokosch, E. 1939. A comparative Germanic grammar. Philadelphia, PA: Publications of the Linguistic Society of America, University of Pennsylvania, Philadelphia. Rákosi, Csilla. 2014. Inconsistency in two approaches to German affricates. Part 2: The basic inconsistency of German affricates in Prinz & Wiese’s approach. Sprachtheorie und germanistische Linguistik 24 (2). 151–182. Robinson, Orrin W. 1992. Old English and its closest relatives: A survey of the earliest Germanic languages. London: Routledge. Sagey, Elizabeth. 1986a. The representation of features and relations in non-linear phonology. Cambridge, MA: Massachusetts Institute of Technology dissertation. Sagey, Elizabeth. 1986b. The timing of contour segments. MIT Working Papers in Linguistics 8. 208–220. Schumacher, Karl. 1914. Studien über den Stabreim in der mittelenglischen Alliterationsdichtung. Bonn: Peter Hanstein Verlagsbuchhandlung. Timmer, B. J. (ed.). 1951 [1978]. Judith. Methuen, revised edn. Devon: University of Exeter. van Langenhove, George. 1930. The assibilation of palatal stops in Old English. In N. Bøgholm, A. Brusendorff, and C. Bodelson (eds.), A grammatical miscellany offered to Otto Jespersen on his 70th birthday, 69–75. Copenhagen: Levin & Munksgaard. Warner, Natasha, James M. McQueen & Anne Cutler. 2014. Tracking perception of the sounds of English. The Journal of the Acoustical Society of America 135 (5). 2295–3006. Watkins, Calvert. 1955. The phonemics of Gaulish: The dialect of Narbonensis. Language 31 (1). 9–19. Weɫna, Jerzy. 1986. The Old English digraph again. In D. Kastovsky & A. Szwedek (eds.), Linguistics across historical and geographical boundaries, Vol. 1, 753–762. Berlin: Mouton de Gruyter. Wetzel, Claus-Dieter. 1981. Die Worttrennung am Zeilenende in altenglischen Handschriften. Frankfurt am Main: Peter Lang Verlag. Wilson, Colin. 2006. Learning phonology with substantive bias: An experimental and computational study of velar palatalization. Cognitive Science 30. 945–982.

Stefan Dollinger

On the regrettable dichotomy between philology and linguistics: Historical lexicography and historical linguistics as test cases Eine einzelwissenschaftliche Untersuchung, etwa eine physikalische Untersuchung kann ohne weitere Umschweife mit der Bearbeitung ihres Problems beginnen. Sie kann, sozusagen, mit der Tür ins Haus fallen; es ist ja ein “Hausˮ da: ein wissenschaftliches Lehrgebäude, eine allgemein anerkannte Problemsituation. (Karl Popper, Vorwort to the first edition of Logik der Forschung, Wien, 1934).1 What, then, is the “Lehrgebäudeˮ (the sum of the accepted presuppositions) in historicial linguistics or historical lexicography?

1 Introduction2 The improvement of existing theories and corresponding descriptions in the pursuit of knowledge is the most central theme in academic inquiry. Historical linguistics has seen its fair share of paradigm shifts in this process of knowledge creation, from the prioritization of linguistic pre-history in the nineteenth century (e.g., Schleicher 1866; Müller 1862) to current trends of complex adaptive system simulation (e.g., Ritt 2004; Kretzschmar 2015; Ritt 2016). One dichotomy of approaches in historical linguistics, however, seems to be more fundamental than others, which is the “clash” between philological and linguistic approaches. By philological, I refer to an approach to the study of language that requires intimate familiarity with historical material and knowledge of the culture and literary texts of the language in question. This more “humanistic” approach is 1 A field-specific academic inquiry, for instance a study in physics, can address, without much ado and fanfare, the specific problem. It can, so to speak, come straight to the point, because the basic points have already been made: via a teaching and research paradigm and a generally accepted problem situation. (Translated to preserve the meaning, not the words, by this paper’s author). 2 A debt of gratitude is owed to my Viennese teachers Prof. Dr. Herbert Schendl and Prof. Dr. Mag. Nikolaus Ritt, for whom this paper’s concepts are nothing new, while its examples will hopefully be pleasing. The reviewer and book editors are thanked for their careful input, especially Don Chapman, whose incisive, detailed and humanistic comments made this paper a better one. The usual disclaimers apply.

62

Stefan Dollinger

often comparative and multilingual and centres in textual scholarship; it is “bridging” the study of literature and language to a greater or lesser degree and is often, if not usually, carried out in a qualitative manner. It tends to interpret linguistic instantiations in their precise temporal, cultural, social, linguistic, and – to the degree possible – situational contexts in order to arrive at meaningful conclusions. In other words, it particularizes to arrive at interpretations and prefers to explain processes post hoc. By linguistic, I mean an approach to language that presents itself in express opposition to literary scholarship, an approach that generally rests on a quantitative framework that is used to test hypotheses in an experimental, usually quantitative, manner. It does not necessarily require competence in the language of study, as long as the variables of modelling are clearly circumscribed. It aims to maximize the predictive power of its models and rests on statistical decisionmaking processes and hypothesis testing. In other words, it generalizes from the existing data to reach beyond particular instantiations. This approach strives to make predictions based on existing data and is, not by design but in practice, de facto a monolingual activity. The dichotomy between the two approaches of philology and linguistics has not been left undocumented. On the contrary, it represents one of the most profound debates in the wider field. In Roger Lass’s important anthology on English historical linguistics from the late 1960s, the issue was given centre stage. Lass characterized the dichotomy between philologists and linguists as a “breach” between two disciplines (1969: v), between a time-tested “older” philological approach and a “newer” linguistic approach. In that same volume, Robert Stockwell, historical linguist extraordinaire, attested a lack of vibrancy in philological approaches, while he saw a rather dynamic field in linguistics at the time. As a historical linguist, Stockwell (1969: iv) found himself “little satisfied by the quality of explanation offered for historical change by the familiar tradition of [philological] scholarship”. Much has changed since the 1960s, and today historical linguists have learned to integrate philological knowledge with linguistic theory building. In an inspiring volume, Adamson and Ayers-Bennett not only demonstrate the merits and relevance of philology, they go so far as to attest “various forms of ‘rephilologisation’” (2011: 204). In this context, Anna Morpurgo Davis (2011) identifies three general types of practitioners: literary philologists, linguistic philologists, and linguists. One might dare to venture that most English historical linguists, probably most historical linguists as such, would perceive of themselves as situated somewhere between the extremes of linguistic philology and linguistics.

Historical lexicography and historical linguistics as test cases

63

What is true of historical linguistics as a whole, however, cannot be said of one of its sub-disciplines, historical lexicography. The reason is not found in a lack of philology in that discipline, on the contrary. In a way, historical lexicography has had no need to “rephilologize”, simply because it has not left the philological corner. Since historical lexicography offers generalizations by methods different from those used in linguistic theory, there has not been much interchange between the two fields. Corresponding with this disciplinary division, a not insignificant number of historical lexicographers would probably describe themselves as situated somewhere between literary philologists and linguistic philologists, that is, on the farther end of philology and quite on the opposite of linguistics. Such differences are remarkable, considering that both approaches deal with the same historical source texts. The present paper exemplifies Lass’s “breach” between philology and linguistics as it pertains to historical lexicography in the first instance and explores, where conveniently possible, the issues as they pertain to historical linguistics and linguistics more generally. The case is made that an exchange between philological methods in historical lexicography and the methods of sociolinguistic theory, which is one of the most science-oriented approaches to language, would help both disciplines. It is argued that such exchange would improve linguistic approaches through philology, while philology would be enriched with more powerful generalizing methods that open up historical dictionaries to novel uses. Perhaps the most challenging point in this paper is the argument that it is not only the newer linguistic-computational methods that are needed to improve historical dictionaries, as the counterpoint is made that philological competence and methods have something vital to contribute to linguistic questions, though the latter point is more difficult to demonstrate. Historical lexicography is not singled out to let off historical linguistics scot-free, however, as the latter is confronted with academic/scientific theory problems of its own, that is erkenntnistheoretische or wissenschaftstheoretische Probleme, ‘problems pertainting to the theory of science’, for lack of an equivalent English word. These problems persist in historical linguistics, albeit at a much more advanced level than in historical lexicography, and shall be addressed in the latter part of this paper. This paper begins with a brief assessment of the state of lexicography in academia and its two obvious modes: a literary-philological approach of doing lexicography, which is central to historical lexicography, and a computationallinguistic approach, which has been in use for about three decades and has become a force in the last fifteen years that could no longer be ignored (those projects and publishing houses that did not adapt quickly were shut down or downsized considerably). The core of the paper consists of three case studies from three historical dictionaries of English, which are preceded by a brief

64

Stefan Dollinger

historical contextualization of the Oxford English Dictionary and Grimms’ historical Deutsches Wörterbuch to highlight biases that affect linguistic work with the digital versions of both pioneering dictionaries. The three case studies aim to show individually (a) that historical lexicography is precise enough to inform linguistic approaches to semantic-pragmatic phenomena, (b) that lexicographical evidence would enrich sociolinguistic theory building, and by extension other linguistic theories, by offering hard and dateable evidence of linguistic change and, (c) that the reader of historical dictionaries would have to have philological skills to contextualize information appropriately. All three cases combined cast into sharp relief the need for “dual competencies” in linguistic methods and in philological methods in historical lexicography and historical linguistics and, by extension, linguistics more generally. A discussion section on theoretical problems relevant to the discipline follows the case studies. Such wissenschaftstheoretische reflections (for which I use the term academic/science theoretical) are necessarily introduced only after the case studies. While they carry a considerable part of the main argument, they need to be read in the context of the preliminary remarks and the three cases: readers are therefore asked to bear with the author. A brief conclusion succinctly summarizes the main point and invites, finally, lexicographers to carry out their projects from an underlying theoretical question of interest beyond the general purpose of their dictionaries.

2 Literary-philological and linguisticcomputational tensions Lexicography as a discipline is deeply rooted in literary traditions, and it is fair to say that English lexicography would not exist in the present form without literary and philological approaches. A literary bias is evident in English historical lexicography. It is commonplace that English philological scholarship “leaned heavily on German writings” throughout the nineteenth century (Turner 2014: xvi), though a more profound dependence on the literary approach developed in English. For instance, Richard Chenevix Trench, the later (Anglican) Archbishop of Dublin, is usually remembered as the first to propagate a descriptive approach to English. His dictum “It is no task of the maker of [a dictionary] to select the good words of a language” (Trench, 1857: 4) is generally read as the beginning of a descriptive tradition in English lexicography, yet there were clear limits to Trench’s linguistic universe, not only as a man of faith but in connection with

Historical lexicography and historical linguistics as test cases

65

views of acceptable literature at the time. A Victorian lexicographical preference for accepted and canonized literary sources at the expense of more “vulgar” ones is evident. The selection of source materials made the Oxford English Dictionary (OED) into the primary record of literary language, and, one needs to add, a truly astounding one. The German lexicographical tradition, for instance, has developed rather differently. Grimms’ dictionary, quite in contrast to the OED, includes fewer genteel sources, which can be explained in this dictionary’s indirect assistance with the unification of the German “Volk” in the mid-1800s. On the other hand, and consistent with that implicit goal, it applies a much more stringent criterion, ruling out “obviously” (“wie sich von selbst versteht”), “the bigger part of loan words in circulation” (“des gröszten theils der unter uns umlaufenden fremdwörter”, Grimm 1854: xxxiv). Jacob Grimm’s conviction that all languages share a natural propensity “to keep at bay foreign material” (“das fremde von sich abzuhalten”) and, where such material finds entry into the language, “to reject it, or at least neutralize it with native items” (“und wo sein eindrang erfolgte, es wieder auszustoszen, wenigsten mit den heimischen elementen auszugleichen”, Grimm: xxvi) gives expression to a political agenda that has nothing to do with linguistic descriptivism. The sociolinguistic advantage of the inclusion of more folk and vulgar terms in Grimms’ Deutsches Wörterbuch is counterbalanced by the arbitrary and Volk-ideology-inspired exclusion of (integrated) loan words. To summarize, while a focus on genteel or acceptable literature is found in the OED, Grimms’ dictionary was meant to help the common German people in their linguistic emancipation, with the German language “instilling a more vivid sentiment” when compared with a foreign language (“eine lebhaftere empfindung für den werth [. . .] einflöszt”, Grimm 1854: xiv), but showed obvious signs of “language policing”, namely the exclusion of subjectively disliked loan terms. As the lead dictionary for English, the OED’s practice has influenced many lexicographers even beyond English, much more so than the Grimms’ pioneering historical dictionary whose first volume antedates the first OED fascicle by almost half a century. Among the OED’s successors is dialectologist-lexicographer Walter S. Avis, Michigan-trained chief editor of the first edition of the Dictionary of Canadianisms on Historical Principles (DCHP-1, see Dollinger et al. 2013). Avis, in apparent contradiction of his dialectological persuasions, shied away from the inclusion of manuscript or speech sources because of the OED’s status as a “lead dictionary”: To follow this tradition [as established by the OED], however, is to accept certain restrictions, for it precludes the entering of terms for which there is only oral evidence and of others for which the printed evidence is fragmentary or otherwise inadequate. As a consequence, many interesting items remain in the files, awaiting fuller substantiation in printed sources. (Avis 1967: n.p.)

66

Stefan Dollinger

In the English tradition, spoken data has been included from the start in the Dictionary of American Regional English (DARE; Cassidy and Hall 1985–2013), which, begun in the 1960s in earnest, was the first major historical dictionary of English that expressly considered spoken data.

2.1 The editor and the philological method as a “black box” There are other aspects of the philological enterprise that may reduce the quality of the works in often subtle, yet clear ways. In dictionaries, the editors’ knowledge is necessarily socially and regionally restricted, which creates the inevitable risk of reproducing the attitudes of the editors’ social group and not necessarily “the language” as such. This methodological problem is compounded in historical lexicography by the diachronic distance to the target period. Jucker et al. (2013: 1) describe the well-known problem of interpreting words of past periods in the following way: How, for instance, can we establish the meaning of a particular word in an older text? We may deduce a possible reading on the basis of what would make sense in the given context, but the notion of what makes sense is very likely to be influenced by our modern reading habits, which may differ significantly from the reading habits of past readers.

The authors make the case for the consideration of the “multiple interactions of linguistic organization including orthography, morphology, syntax and, ultimately, pragmatics” Jucker et al. (2013: 1). This is certainly the best route of action, especially when combined with knowledge about a period. It is also equivalent in intent yet different in methodological execution to what literary scholars call “close reading”, that is the inclusion of all textual and social information at hand. While linguists and linguistic philologists usually bring a more thoroughly defined descriptive apparatus to the task, the basic idea is the same. The method, however, is subjective to a considerable degree and runs the risk of over-interpreting signs, every “dot on paper”, as meaningful and relevant to the analysis. The task of generalizing from linguistic examples in a dictionary context is entirely left to the lexicographer, while the precise process remains underspecified and buried in the abstraction capabilities of often just one person. This caveat applies to lexicographic greats like Grimm or Murray and to beginners alike, as both deal with a methodological problem that is situated in the concrete case, the concrete attestations available, a given editor’s backgrounds and preferences, and, ultimately, the editor’s current state of mind, for lack of a better word: an editor is forced to make choices, but there is

Historical lexicography and historical linguistics as test cases

67

no principle, beyond the most general ones, no documentation, and often no clear and precise rationale behind the selection of, well, almost anything in the lexicographical editing task. I am perhaps painting a picture bleaker than necessary or worse than daily practice proves, but given that users generally assume a principled approach behind dictionaries, it seems worthwhile to stress that the literary-philological method is an inherently subjective one. The linguistic-computational method, which is poised to go beyond the editor’s or the editorial team’s linguistic, social, and philological experience, promised to alleviate the subjective bias. It is fair to say that over the past thirty years, the linguistic-computational approach has influenced lexicography to a considerable degree. Historical linguistics has not been far behind these developments. Pioneered by the Helsinki Corpus in the mid-1980s (Kytö 1996), a large family of historical corpora has developed since. Corpus-linguistic approaches to language change have become standard in the field (see, e.g., Renouf and Kehoe 2006), as corpora present a convenient means for quantitative methods that have, traditionally, not been at the forefront of philology. It seems noteworthy, however, that quantification has been used by philologists since the nineteenth century, as William Whitney’s quantitative-descriptive approach from the early 1880s attests to (1971 [1884]). While the advent of corpus linguistics has altered some aspects of the literaryphilological method in synchronic lexicography, the basic underpinnings of the literary-philological method, I would like to argue, were left largely unaffected, including its exposure to subjectivity in lexicography. Prior to the 1980s, dictionaries were exclusively edited from quotation files, but since then quotations files have been replaced by searches in electronic corpora to a considerable degree. At the level of primary data, the use of corpora put the lexicographical enterprise on a more balanced, more representative level. The interpretation of corpus data, however, is still fraught with potential for human error. The basic problems of interpretation of meaning persist, even if all relevant analytical categories are identified in the data. Corpora, however, were only the first step in a quantitative turn that has been sweeping through the linguistic landscape more generally, and with it sophisticated statistics procedures and mathematical routines have found uptake (see, e.g., Gries’s 2009 textbook). A quantitative turn brings not only chances but also dangers. The removal of historical materials from their sociohistorical and sociocultural contexts is an obvious one for statistical methods. This danger was not left unremarked by the first generation of historical corpus linguists. Matti Rissanen, one of the creators of the Helsinki Corpus, spoke in the late 1980s of the “risk that corpus work and computer-supported quantitative research methods will discourage the student from getting acquainted with

68

Stefan Dollinger

original texts, from being on intimate terms with his material and thus acquiring a profound knowledge of the language form he is studying” (1989: 16). This “philologist’s dilemma”, as Rissanen called it, would in the extreme “mean the wane of philologically oriented language studies” (1989: 16). It goes without saying that such development would not be an improvement over the existing situation: one might manipulate databases and run statistical tests yet with no or only a limited idea about the language that is being processed. Better quantitative skills must not be bought at the cost of worse or no philological expertise.

3 Case studies The three case studies that follow demonstrate – beyond the obvious truism that linguistic-computational approaches have something to offer philology, which is indirectly expressed in Stockwell’s dictum from 1969 – that philological approaches in historical lexicography, and more generally in historical linguistics, do have something to offer to the linguist. The first case shows that historical and philologically sensitive readings of corpus material help improve historical dictionary entries. The second case exemplifies that carefully researched lexical items may inform sociolinguistic theory in important ways and allows the dating of changes in progress. It highlights, in a select but crucial aspect, the relevance of – in the ignorant words of some linguists – “old-fashioned and outdated” ways of doing linguistics. The third case is a call for caution when consulting historical dictionary entries. This case suggests that philological knowledge is essential not only for the writing but also for the reading of historical dictionary entries. This case offers perhaps the most forceful support to fight against Rissanen’s “philologist’s dilemma”: even when not reading old texts as such, but when merely using a supposedly general historical dictionary, one cannot dispense with philological knowledge. All three cases combined are offered as proof that philology, while an older, more venerable approach, is not “passé” but very much needed. I make the point that not just historical scholars but linguists of a number of persuasions would benefit from training and expertise in philology – not instead of but in addition to the linguistic-computational training they generally receive today. This paper understands itself as a constructive piece of critique of linguistic approaches that fundamentally work in abstraction: it is argued that they would especially benefit from philological expertise, if in some cases only for a fail-safe that what is thought to be studied is actually studied.

Historical lexicography and historical linguistics as test cases

69

3.1 Historical pragmatics and lexicography: Cull and cully This first case demonstrates that historical lexicography is precise enough to inform historical-pragmatic work, which in turn helps improve dictionaries. Historical pragmatics focuses on precise contextual analyses, often enhanced with quantitative elements, to a level of detail that is usually not carried out in (historical) lexicography in the literary-philological tradition, if only by virtue of dealing with thousands of terms rather than just one. The difference is best shown with an example, which I take from Roxanne But (2013). The cant term cull (“gullible person”) offers an interesting yet simple test case for the OED. But’s study (2013: 140–142) uses historical corpora and close readings of the texts to inform theories about eighteenth-century semantics. The approach is generally qualitative, although some basic counting is used in table format. For instance, between 1725 and 1800, fourteen collocations of cull with modifiers are identified, such as country cull, dossy cull, sleeping cull, or sucking cull, all used to specify the referent more precisely, that is, the persons who were robbed by thieves. Figure 1 shows the relevant entry for cull from the first edition of the OED dated from 1893:

Figure 1: OED, s.v. cull, n. 2

It appears that the OED definition is largely in line with the interpretation of But’s fourteen collocations for cull, which is good news. But’s analysis, however, also reveals some detail that is not found in the entry cull, n. 2. Figure 2 shows both the original digitized file from the Old Bailey court proceedings and the transcription, which But used in her corpus search yet did not include in her account because of her focus. These can, however, influence certain readings of

70

Stefan Dollinger

the term. Figure 2 shows the minimum degree of abstraction from an original in the loss of the orthographic, typographical, and aesthetic information (old print type, ligature s, different spacing, and the like), such as the (verbatim) transcription from the Old Bailey Corpus below shows: Mr. Corson the surgeon deposed that the fracture in the deceased’s leg was the cause of his death. “The prisoner called. Henry Parsons, who deposed, that when the prisoner came into the house, the deceased said that he did not believe that the prisoner had been knocked down; that the prisoner said it was no business of the deceased; that then the deceased abused him and called him a crying cull, and said he was no man, and challenged him to wrestle for a pound; that the prisoner said he could not wrestle, and he would not; that then the deceased challenged him to fight, if he would bet twenty-six shillings to his guinea, on account of the difference of their age. That the prisoner asked his uncle to lend him a guinea in order to fight the deceased; that no blows passed; that when the prisoner’s uncle went away,

Figure 2: Crying cull in context (Old Bailey Corpus). Context: JOHN DEWELL, Killing > murder, 19 February 1772.

The parallel text makes obvious what a verbatim transcript without obvious errors can look like. The apparent issue would be to compare the LModE print text with the original transcript, as some issues have been reported for LModE that were probably normalized by the printer, capitalization being one of them. But this is a point that cannot be pursued here for space constraints (see Osselton 1985 and Dollinger 2004 for two such cases). When reading the passage as is, however, some confusion arises as to the denotation of the term crying cull: who is the crying cull, the victim or the perpetrator? In Figure 2, the person killed, the “deceased”, insults the accused as being a crying cull, so there is some doubt about the meaning of cull. In her corpus evidence, But finds unambiguous evidence, and this is the keystone in the argument, that cull does not only refer to victims but also, at the example of scamp cull in (1) shows, to perpetrators:

Historical lexicography and historical linguistics as test cases

(1)

71

I believe you are a Scamp Cull, or else you would not go to stop old Collectors

As Scamp Cull is glossed as “highwayman” in the mid-eighteenth-century source (But 2013: 141), no doubt remains: cull is used not just for victims but also for perpetrators. A quick, simple check in the OED would have led us along the garden path. It seems that the OED, to be more precise James Murray and his team for OED-1, fascicle C, missed the “perpetrator meaning” of cull and its compound nouns. The damage, however, is not as great as it seems at first, if one performs more than just a quick check on one entry and exercises philological rigour. This is because of the connection with cully, which is offered as a possible source for cull (see Figure 1, “Etymology”). The entry for cully shows not one but two meanings: the “victim meaning” (1) and also a more neutral meaning (2), that explains the use of the term for the perpetrator: 1. One who is cheated or imposed upon (e.g., by a sharper, strumpet, etc.); a dupe, gull; one easily deceived or taken in; a silly fellow, simpleton. (Much in use in the 17th c.) 2. A man, fellow; a companion, mate. Figure 3: OED meanings for cully

Meaning (2) appears to be the source for the perpetrator meaning of cull. The example shows that historical-pragmatic readings of corpus material – with the focus on reading, not on mere counting – further increase the precision of the lexicographical entry and will help in improving the semantic structure of cull (n.) by identifying a second meaning for cull (n. 2). Historical pragmatics has therefore a sort of “applied” function in the improvement of historical dictionaries, which depends on the philological competence of the analyst. Rissanen’s philologist’s dilemma comes to mind, as only someone with background training in the period texts would be likely to make sense of the meaning distinctions in the corpus materials. With relatively simple means, and without going too much into quantification, certainly not beyond a level of basic tallying, this is probably as precise as descriptive methods can get. The example also shows that information might be somewhere in a standard reference work, but not always in the most obvious places. Here it is under cully, not cull, while the link between the two is presented as speculative. A linguist without philological training, I would argue, might just miss that connection via a possible but speculative etymological relation.

72

Stefan Dollinger

3.2 Philological data as real-time evidence The second case is different from the first, as it offers the crucial philological piece of evidence that proves a development that was originally proposed with sociolinguistic reasoning. As is well known, the description of social and linguistic constraints has been developed to an impressive degree since William Labov’s (1963) inaugural study in variationist sociolinguistics. One of the strong suits of the discipline is the detection of systematic linguistic patterns in what is chaotic behaviour to the casual observer (e.g., Labov 1972). Often difficult to pinpoint, however, are the social motivations for linguistic change, and it is in this area where historical lexicography can help. The present case of “telling time” rests on questions that were first explored in the Linguistic Atlas of the United States and Canada projects (e.g., Kurath et al. 1972 [1939–1943]). The questions found their way into the Dialect Topography of Canada (Chambers 1994), from which the sociolinguistic data for the present case study derives (Pi 2000). Respondents were shown two clock dials with the corresponding questions:

Q. 18 What time is it? (Please write in words what you would say.)

Q. 49 What time is it? (Please write in words what you would say.)

There are two basic variants for answering the above questions: on the one hand twenty to twelve and half past nine, which are called analog variants, and on the other hand, eleven-forty (11:40) and nine-thirty (9:30), which are the digital variants. Tony Pi assumed that analog formats are older than digital formats but made litte effort to date his claim in any verifiable way. Historical, philological work on the history of time keeping gives an infinitely more substantial picture: Samuel Pepys, for instance, the mid-seventeenth-century London diary writer who was obsessed with time, never referred to any partial hours and, if at all, he approximated them, as in “About 4 o’clock comes Mrs. Pierce to see my wife” on 25 May 1667 (1971 [1893]: 1741). About one and a half centuries later in her journal entries, Canadian Charlotte Harris routinely refers to half hours but already occasionally to quarter hours, such as in “Feb 13 We left again at a quarter to nine” (CONTE; Harris and Harris 1994: 18). Digital time telling is not yet found in these sources.

Historical lexicography and historical linguistics as test cases

73

Table 1 shows data from the Eastern United States and Canada from the 1990s. It is characteristic of analog time telling to have a number of variants: for 11:40, people use twenty to, twenty before, twenty till/’til, or twenty of. For 9:30, one finds for instance half past nine, half nine, half after nine. As with other open questions in written questionnaires (Dollinger 2015a: 243–244), a noticeable number of multiple responses can be seen in the data. In the present case, dual responses, when respondents answer half past nine and 9:30 for instance, are of particular importance, as shown in box B: Table 1: Cross-tabulation of responses, 2,440 responses (Pi 2000: 92)

The data is consistent: most responses, to be precise 1,693 (or more than 69 percent of all responses), are in box C. These are the (majority of) people who answered in a split way, using the digital 9:30 while the analog twenty-to. The next largest numbers are then found in box A (249), which stands for both variables as analog, and box I (371), where both variables are digital. Based on these data, Pi (2000) argued for a change in progress from A to B, on to C then to F, and finally to I. This trajectory is particularly convincing, since boxes B (50) and F (51) show sizeable figures and could be seen as transition stages. The white-shaded boxes, however, feature only very low numbers, from one to fourteen at most, and therefore indicate outlier behaviour. As box B suggests, the half-hour point is the first to change, with variation in half past and 9:30, and the 11:40 time-spot following only with a considerable delay. This is good social-science reasoning, as it builds up a case from the raw data. It is good sociolinguistic reasoning as it involves the speaker-ages in an apparent-time construct to arrive at a solidly supported hypothesis. Apparenttime dating allowed the conclusion that around 70 percent of respondents would have responded with the digital variant 9:30 as early as the 1930s. With informant comments from the Linguistic Atlas of the Upper Midwest (LAUM; Allen 1973–1976), Pi (2000: 83–84) was able to suggest that the change for the

74

Stefan Dollinger

half-point was likely older and a result of the railroads and their necessity for precise time keeping. His evidence is anecdotal, though: watches, Pi (2000: 82) surmises, “were commonly known as ‘railroad watches’ in the late nineteenth century”, but without any evidence. We therefore have an interpretation that hinges on apparent-time evidence from the early 1990s (Pi’s data) and some real-time evidence from the late 1940s (LAUM’s data) for the dating of the half-hour conversion to digital. Access to the latter, it needs to be pointed out, involves some basic philological skill of checking back with older secondary literature, which is not a given in a science tradition that considers papers older than ten years as obsolete in some disciplines. What is still missing, however, is the real-time “anchor” for this change. At this point the Dictionary of American Regional English (DARE; Cassidy and Hall 1985–2013) comes to the rescue of the present interpretation as it provides the essential clue, not under “railroad watch”, as Pi might surmise, but under railroad time, as captured in Figure 4:

Figure 4: DARE’s entry for railroad time, meaning (2), with quotations to mid-century

The 1865 quotation is the earliest attestation for railroad time. What is striking is that it is already used in a jocular, generalizing context – railroad time is not explained in this quotation from the Atlantic Monthly but rather is the punch

Historical lexicography and historical linguistics as test cases

75

line of a joke: “She is . . . so punctual that railroad time might be kept by her instead of a chronometer”. The evidence would allow a dating to the early 1860s if not a little earlier, by which time readers of the Atlantic Monthly would have been familiar enough with a digital time format. The matter-of-fact tone of the 1880 and 1883 quotations do not contradict this dating process. Please note that this evidence is ambiguous as to whether it means digital time (e.g., 11:43) versus exact time (e.g. two minutes before quarter to twelve). The two, however, eventually are identical as precision requires the introduction of digital time reading at some point. Table 2 is a summary of the dated changes from analog to digital time telling. The majority pattern in the data, 9:30 combined with twenty to, is dated as lasting from about 1930 to 1980. The youngest-age cohort still uses the combination 9:30 and twenty to in most cases. These speakers grew up in the 1970s and 1980s, when digital LCD displays became common. The transition of the 11:40 variant coincides with this time line and likely contributes to an acceleration of the change. Digital clocks, however, are not the driving force behind the change to digital, which, as we have seen, is a much older development: Table 2: Trajectory of change for time telling half past twenty to

mid-19th century

→

nine-thirty or half past twenty to Mid-19thcentury to 1920s

→

nine-thirty twenty to

1930s to 1970s

→

nine-thirty twenty to or eleven-forty

→

nine-thirty eleven-forty

1980s to present

The present case demonstrates the importance of real-time evidence in sociolinguistic theory, for which DARE provides philological evidence from its historical files that comes close to a “smoking gun” in linguistic change, in this case of usage. Sometime between Charlotte Harris’s journal entries from 1848 and 1849 and the 1865 DARE quotation, “railroad time” became a known concept and term in eastern North America. From then, it has trickled down from official language use to the general language, affecting more and more contexts, in a change that is only now, after a century and a half, starting to complete its turn to digital time telling. Without the philological backdrop of DARE, which is otherwise a project based on social-science principles, Pi’s theory would remain nothing but an unsubstantiated hypothesis that cannot be anchored in real time.

76

Stefan Dollinger

3.3 Historical lexicography and philological readers The last case study is taken from the first edition of A Dictionary of Canadianisms on Historical Principles (DCHP-1, Avis, Crate, Drysdale, Leechman, Scargill, and Lovel 1967; see Dollinger, Brinton, and Fee 2013 for DCHP-1 Online). DCHP-1 records Canadianisms, which are defined as words, meanings, or expressions that are “native to Canada” or that are “distinctively characteristic of Canadian usage though not necessarily exclusive to Canada” (Avis 1967: n.p.). This scope makes DCHP-1, and more so the new edition, DCHP-2 (Dollinger and Fee forthcoming), a contrastive dictionary: what is in DCHP-1 and is not marked as nonCanadian (with a dagger symbol) meets the definition. Figure 5 shows the entry for lemma ice (n.) and lists three main meanings:

Figure 5: Entry for ice, n. in DCHP-1 Online

Meaning (3) is a curling-specific sense, as indicated by the label, and it does not seem very surprising to anyone familiar with Canadian winter culture that it is listed separately. Curling plays a big role in Canada to the degree that it is part of national identity for some, but in contrast to (ice) hockey, it is an amateur sport of national importance, much like downhill skiing in Austria or ski jumping in Finland. In rural regions and smaller towns, curling is a major part of social life. The 1956 quotation for meaning (3) is from a sports commentary in a rural Northern Saskatchewan context and is thus unspectacular. It is also contemporary with work on DCHP-1, which facilitated a balanced editorial judgement.

Historical lexicography and historical linguistics as test cases

77

The 1911 quotation, however, is more important not just because it is older but also because it originates from a rather different source. It reads: (2)

The ice was capital – for the floor was a wooden one and twenty-four hours’ frost had been quite enough . . .

This quotation is short, which is characteristic for many DCHP-1 quotations (in contrast to DCHP-2), is relatively early in the Canadian context, and is, as shown in the bibliographic reference on the right of the screenshot, from the 1911 novel The Singer of the Kootenay – A Tale of To-Day by Robert Edward Knowles, which is listed as published in Toronto; the quotation appears on page 281. This information seems intuitively right and representative, and this is where the danger lies, as is argued below: curling, a Canadian sport, shows a quotation for ice in the special meaning of an ice curling rink as early as 1911. Together with the 1956 quotation it was chosen to represent meaning (3). They were selected, as Avis put it “[1] to supply the earliest and, usually, the latest evidence on file; [2] to supply fuller descriptive detail than is possible or desirable in a definition; [3] to supply definitions where acceptable and to which the reader may be referred; [4] and to supply interesting, entertaining, and otherwise appropriate contexts in which the term occurs” (1967: xviii). Of these four reasons, (1) is quite logical; (2) represents one of the key features of the quotation paragraph in historical lexicography (see Sheidlower 2011); (3) is a restriction in paper-based lexicography and primarily motivated by space constraints, with (4) being a more fuzzy criterion. 3.3.1 Particularizing the 1911 quotation So far we have seen the interpretation of ice, n., def. 3 in a generalizing manner, as is implicit in dictionary entries. But there are other elements to the interpretation. By situating the 1911 quotation in its sociohistorical context, some details of the quotation will be unearthed. To begin with, the author of the 1911 quotation is Robert Edward Knowles (1868–1946), a native of Ontario and son to Irish parents. Knowles originally became a Presbyterian priest and served as minister in Galt, Ontario. He was a strong supporter of the temperance movement. At the time of publication of The Singer of the Kootenay, Knowles had made a name for himself as a “spellbinding” preacher and speaker (City of Cambridge 1997) and a prolific novelist, with The Singer being his sixth and final novel. The year The Singer was published, Knowles tragically suffered a train accident, the effects of which led to his resignation from the ministry in 1915. From that point on he dedicated himself fully to his writing, switching genres and

78

Stefan Dollinger

soon becaming known for his newspaper coverage, in particular his many interviews with people of high exposure in the Toronto Daily Star and the Toronto Star Weekly (O’Grady 1993). Significantly in the present context, he was chaplain of the Ontario Curling Association for some time, a religious post not uncommon in those days for sports teams, so Knowles knew what he was writing about. In 1997, he was inducted into the “Hall of Fame” of the City of Cambridge in recognition of his service to the development of the community and “humanity in general”, as the statutes read. The biographical background is of relevance to many literary methods, though it is generally not intended to overrule the textual evidence. In the absence of a manuscript source, the printed book will serve for an evaluation of the quotation. Figure 6 shows the page that the 1911 quotation references:

Figure 6: Scan of the page containing the 1911 quotation in The Singer (p. 281)

The bibliographical details of the book are noteworthy: The Singer’s place of publication is listed as Toronto (see Figure 5), yet that city is only one of five listed locations, and the only Canadian one. The full list is: “New York, Chicago, Toronto, London and Edinburgh”, therefore comprising places in three countries, which is not rare for international publishing houses. The sole choice of Toronto in DCHP-1, however, certainly creates a Canadian angle for the book

Historical lexicography and historical linguistics as test cases

79

that it does not have. The novel’s publishing house is Fleming H. Revell Company, which is a company devoted to “help bring the Christian faith to everyday life”, which, in the words of George H. Doran, Revell’s Vice President at the time, “embarked more into the field of general publishing with a religious or highly moral flavour” (Baker Publishing Website). The company’s origins lie in Chicago in the 1870s; by the 1890s Revell was the biggest religious publishing house in North America with offices in other cities, of which Toronto, set in the biggest market in Anglophone Canada, was just one. So Knowles was writing for an American, most likely even international, English-reading audience. Given the Edinburgh connection of the publisher, he was therefore not writing for the Canadian market, which casts doubt on the generalization that ice “curling rink” was merely Canadian: otherwise, why would Knowles use a purely Canadian meaning, as suggested by inclusion in DCHP-1 and its non-marking as non-Canadian, for an international audience? The newly revised OED entry for ice, n. shows that the meaning is originally Scottish and antedates the Canadian attestation by a century (Figure 7, dated from 2012):

Figure 7: OED entry for ice, n., def. 2d

John Ramsay’s An Account of the Game of Curling: By a Member of the Duddingston Curling Society is as much an insider account as is Knowles’s a hundred years later. Printed in Edinburgh, the account is good evidence for labelling the meaning Scottish in the first instance. But this does not preclude, of course, it being Canadian as well.

80

Stefan Dollinger

A closer reading of Knowles’s book reconfirms the Scottish context: Murray McLean, the young protagonist leaving Ontario for British Columbia, is reminded of his Scottish heritage when he hears the sound of the curling stones: Suddenly he [Murray] stood still beneath the frosty skies. For he had heard a sound he was at no loss to recognize – “his heart more truly knew that peal too well.” Every Scot, native born or extracted, would have known it at once. It was a kind of union of roar and shout and shriek. He listened again – and some word fell on his ear that absolutely convinced him. “They’re curling!” he exclaimed excitedly, still standing and gazing across the common that intervened between them and a low shed about a hundred yards away. (The Singer, p. 280)

There is agreement among historians of curling that the sport owes its dissemination to Scottish emigrants. Its origins in Canada date back to 1807 with the Montreal Curling Club for organized sports and is attributed “to the consistent, enthusiastic and ubiquitous presence of the Scots” (Bailey, Redmond, and Freeborn 2010: n.p.). Since the foundation was laid in early Canada, the sport “has thrived to a prodigious and unparalleled extent here” (Bailey, Redmond, and Freeborn 2010: n.p.) so that today curling is seen as a Canadian sport. What the 1911 quotation does not tell us, crucially, is that at that point in time curling was not yet Canadian, but rather Scottish, and that the term would become Canadian only over the course of the first half of the twentieth century. The context of a novel by a curling insider is hardly evidence for widespread dissemination of ice in this special meaning. Even today, there is reason to believe that one would not find this meaning much beyond the immediate sports context. It would be difficult to relay such information in the quotation paragraph, and the logical conclusion would be to offer some sort of account of the development of the meaning, which DCHP-2 aims to do with its “word story” – an essay-style mini account of the etymology and semantic and social developments of the term (Dollinger 2015b). What the example of ice, n., def. 3 illustrates is the generalizing force of dictionary entries, which are interpreted against the purpose of the dictionary, in this case a dictionary of Canadianisms, turning the meaning into a Canadian meaning regardless of the temporal dimension. In many ways, the close reading methods of case 1, crying cull, go some way towards solving the generalizing problem we have just seen in case 3 on ice “curling rink”. The limitation of a usage label in DCHP-1, Curling, is not much help, however: the label is ambiguous as it may be read as the term being limited to that field and not known in the general language or as merely marking the term’s primary usage domain. The philological key procedure of going back to the original texts, as shown

Historical lexicography and historical linguistics as test cases

81

here, offers a fuller interpretation that is otherwise not available in the quotation and shows the virtue of a philological reading. In one way, the brevity of the quotation paragraph almost commands cross-checking against the sources. In other words, a full reading of a dictionary entry is only possible with philological expertise.

4 Discussion The three case studies combine to make one essential point: that philological expertise is not an extra, unnecessary addendum to more social science–based ways of doing linguistics. It is a skill set, an expertise, and a way of looking at language that offers, as seen in all three case studies, a way to improve existing approaches. In the first case study, the improvement was carried out by particularizing evidence that was used to augment a dictionary entry (from the OED), which is necessarily of a general nature. In the second case, a study based and built on a bottom-up data-driven approach that is typical for social science, new insights were arrived at and hypotheses were substantiated with philological evidence from the DARE quotation file. In this case, the DARE generalization was still specific enough to allow the connection between railroad time and digital time telling. In the third case study, philological reasoning that is more reminiscent of literary approaches to texts today was necessary to avoid misinterpretations from a concise and condensed dictionary presentation with bare quotations that would have led to wrongful generalizations. This error is “built into” the design of historical dictionaries by disregarding certain aspects of the historical development of a term and by stripping quotations from their precise contexts. The information is, as shown above, recoverable with sufficient philological expertise. In the third case, more particularization revealed the nature of the presented quotation evidence (from DCHP-1) in a much more reliable light. There is a second point that can be made with these case studies when we put them into the context of the development of qualitative versus quantitative approaches in language study, which we link back to time-tested insights on how to best approach academic/scientific inquiry.

4.1 The linguist’s dilemma: on the qualitative (philology) versus quantitative (social-science) split Historically adequate data is of elementary importance in many disciplines and, most obviously, in the historical-linguistic approaches that rest on real-time

82

Stefan Dollinger

data. It is in this area where historical lexicography, historical linguistics, and philology have most to contribute to social-science approaches to language, which generally extrapolate via apparent-time constructs and their potential shortcomings (see Labov 1994: 43–72), though in recent years sociolinguistic variationists have expressed stronger interest in historical data than before (see, e.g., Labov 2008 for inspiring real-time/apparent-time merged interpretations). The danger, however, arises from the fact that a lack of philological expertise and experience will produce suboptimal, or worse, wrong, results. The integration of synchronic and historical approaches has, as a matter of fact, been a goal in Labov’s work at least since Weinreich, Labov, and Herzog’s (1968) key paper, which aimed to introduce a standard for describing language change in historical linguistics. Sociolinguistics aims to follow “the principle that Jespersen laid out that to understand something, you have to understand how it came to be” (Labov in Gordon 2006: 338). The goal is the same in historical lexicography, yet the real-time data of historical linguistics can turn out to be crucial, as the first and second cases have shown. The case needs to be made that philological approaches are not only useful but necessary. In analogy to Rissanen’s “philologist’s dilemma”, I would like to propose a “linguist’s dilemma”, which can be summarized as follows: In order to understand linguistic phenomena, one must include historical data, in Jespersen and Labov’s senses. In order to do so properly, philological competence is required, competence that is often wanting in the profoundly quantitative linguistic approaches. It is at this point where philological linguists need to make the case for qualitative ways of interpretation, as we have seen in the first and third cases. There is historical precedence for a measured combination of philology with science. When James A. H. Murray (1987 [1888]: 181), long-term editor of OED-1, spoke in the context of the OED’s etymologies of the “methods and results of philological science,” he opposed conjecture (which was rampant in English etymology at the time), conjecture that, as we have seen, can also be found in quantitative approaches, such as in the second case, which speculated beyond the data about the motivations for linguistic changes. Quantification does not necessarily mean “better”; but then it just might. In areas in which quantification is possible, one generally has a more thorough basic methodological awareness and a more easily communicable method. Labov, however, is one of the first to acknowledge that despite advantages of quantification, it is “not as if every aspect of our field is open to quantification. Your concepts have to become clear and solid and countable. There are areas, not only abstract arenas of grammar but areas of discourse analysis, where the attempts at quantification may be quite premature” (Labov in Gordon 2006: 334).

Historical lexicography and historical linguistics as test cases

83

It is clear that Labov, as the innovator in language change par excellence, understands the need for a push towards quantification but realizes that certain areas are very difficult to study quantitatively. These are interesting words from the man that lifted the quantitative approach to language change to an entirely new level. One might add that the qualitative interpretation of linguistic phenomena will continue to have its rightful place in linguistics for a long time hence. Historical lexicography, for instance, is one area where it will be very difficult to replace editorial, qualitative decisions by some form of algorithm, but it is certainly worth any attempt.

4.2 Popper’s “Logik der Forschung”: The myth of “theory-neutral” data collection While historical dictionaries offer much information, they are written for a number of purposes. The separation of data collection and the “application” or use of the data comes with considerable challenges. Science theory has long shown (e.g., Popper 1966 [1934]) that no “theory-neutral” collection or interpretation is possible. Popper became famous for recognizing and re-popularizing in the 1930s the necessity for theory in addition to observation, regardless of how many repetitions of an observation one can observe. Since “theory-neutral” is an oxymoron, Popper argues, we have only two basic states in academic inquiry: either data collections with expressed collection theories or collections with unexpressed ones, possibly even utterly unexplored underlying theories that determine what goes in, what does not, and how it is treated. The former is the better choice that allows for consideration of biases, and the latter is, of course, the much worse option, since one cannot consider any biases that are unknown. That latter Popper refers to as the “naturalistic methodological approach” (my translation), which he “rejects” (lehnen wir ab), because it does not recognize the presuppositions it makes (1966 [1934]: 23–25). The lack of theoretical bias is one aspect for which philological reference projects such as dictionaries or linguistic atlases have been criticized, for example, the critique on the Linguistic Atlas of the United States and Canada (Labov, Ash, and Boberg 2006: 4). As lamented by Robert Stockwell in the introduction to this paper, historical linguistics used to be content with the mere description of changes. Such a theoretical approach exposed the discipline to a mere “branch of cultural history” rather than “linguistics” (Labov, Ash, and Boberg 2006: 4), a charge that is not entirely fair but understandable. While historical lexicographers have enough tasks to perform already, it would, however, be important that they clarify the

84

Stefan Dollinger

underlying theoretical assumptions of their work. Such moves would increase the usability of large-sized data-collection projects, which have been underutilized by theorists who are used to working with very specialized but small data sets. Popper writes in this connection about the “rules of the academic/ scientific game” (1966 [1934]: 26, my translation) and lists two such rules as examples: “First, that academic/scientific inquiry never ends: s/he who one day no longer tests statements leaves the academic/scientific realm. Second, that hypotheses that were once confirmed must not be rejected without very convincing reasons” (Popper 1966 [1934], my translation). Popper’s line of thought, thus, rests on a synergetic and corrective combination of theory (the method), on the one hand, and experimental data, on the other hand, but never on the application of one without the other, which would be equivalent to defaulting back to the “naturalistic method”, which he so heavily criticized as “uncritical” (“unkritisch”; Popper 1966 [1934]: 25). Applying Popper’s critique to the subject at hand leads one to conclude that historical dictionaries should be written with some theoretical goal in mind that, most importantly, must be made explicit. The theoretical angles may, of course, vary from historical dictionary to dictionary and will be, indeed, quite different. In the case of DCHP-2, the dictionary I am most intimately familiar with, for instance, the theoretical focus lies on the question of linguistic autonomy versus heteronomy, which is approached via a typology of Canadianisms (Dollinger 2012: 1867–1868), and on noun compounds, which are especially numerous and therefore important in Canadian English. An overt explicit theoretical focus would alter the character of dictionary projects, perhaps considerably so. Yet this idea is not new, as there is precedence with DARE, which narrowed its scope and based data collection on a well-articulated theory of geographical space, thus breaking with tradition in English lexicography. DARE’s theoretical underpinning is thus a thorough investigation of the concept of region (e.g., Cassidy 1973, 1985; Hall 1997 for theoretical work) rather than an attempt at collecting “everything” or as much as possible from all domains. Because of this focus, I would argue, DARE’s legacy is ensured, much more so than comparable works, such as Craigie and Hulbert’s (1968 [1938– 1944]) Dictionary of American English, based on OED principles and today of relevance mostly for its historical role but not for the insights it offers. In the case of the OED, the proverbial and so-promoted dictionary of “everything”, an underlying theme has been introduced through the backdoor of semantic organization, with the Historical Thesaurus of the Oxford English Dictionary (HTOED; Kay, Roberts, Samuels, and Wotherspoon 2009). The HTOED organizes many of the OED’s isolated items according to semantic criteria that break away from the alphabetical straightjacket that made somewhat sense

Historical lexicography and historical linguistics as test cases

85

only in a pre-digital world. But in general, and more so in the future, the onus of proving that historical dictionaries are useful in more than the traditional ways falls fair and square onto lexicographers, who should take that challenge seriously. The three case studies oscillate on the cline from particularizing to generalizing on very different spots, from, if one will, philological to scientific extremes (in the sense of the “hard sciences”, “Naturwissenschaften”). The key question, therefore, is not so much about what empirical method we apply but about the theoretical construct & underpinnings that we are presented with – whether implicitly or explicitly. That theoretical construct ought to be, however, in every case and without exception, explored and made explicit to the greatest possible degree. Philological reference tools, such as historical dictionaries and linguistic atlases, cannot be exceptions to that rule. There can be no exceptions, if we take Popper’s axiom seriously, unless one leaves the academic/scientific field of inquiry, which would be paramount to resorting to the hobby of collecting folk items or, and this is indeed seen, the propagation of “schools of thought” that operate in isolation from each other. Perhaps this was the reason why Alexander Ellis, phonetician and dialectologist of the early hour, summarized the mood of the late nineteenth century as pertaining to dialectology and lexicography “as an amusement, not as laying a brick in the temple of science” (1871: 1087). This means that historical dictionaries need to express their underlying theories. This is not seen much in English historical lexicography, with the possible exceptions of DARE and DCHP-2, and likewise in other languages’ lexicographies, though there may be – it is hoped that there indeed are – projects on the one or the other language that do apply Popper’s principles.

5 A conclusion with an invitation What I aimed to show in this paper were three aspects underlying work in historical lexicography in particular and historical linguistics in general. The statements are, quite naturally, applicable to linguistics more generally. First, philology as a generally qualitative discipline (but not exclusively so) is a particularizing approach that offers important insights to more generalizing approaches such as historical lexicography or social-scientific approaches to language. Second, the qualitative versus quantitative split can be seen as a straw man, deflecting attention from the erkenntnistheoretische, ‘academic/scientific theoretical’, consequences that linguists and lexicographers need to address (and are overdue to do so). Both approaches, qualitative and quantitative alike,

86

Stefan Dollinger

are empirical and aim at generalization, although with different means. The qualitative-quantitative debate seems to cloud the view of the central issue, which, as I hope to have shown, affects historical lexicography disproportionally more as a linguistic discipline that has not yet undergone the quantitative explorations that historical linguistics more generally underwent a while ago, as Lass’s (1969) anthology attests to. The central issue in the narrower field is the academic/scientific theoretical need for theoretical angles that inform data collection in historical lexicography, which is expressed in my call for dictionary editors to expressly place their work in a theoretical context, despite their general, larger goal of writing for as wide a readership as possible, and to interpret the dictionary data from one theoretical angle. This, I argue with Popper, will create better reference tools through the backdoor, as it were, tools that are less likely to fall prey to being a collection of “everything” in their field, that is, a naïve belief in positivism. It would also go a considerable way towards putting the field’s “Lehrgebäude”, in the sense of the introductory motto, on a sounder footing than current practitioners, (historical) linguists, and more so lexicographers have done.

References Adamson, Sylvia & Wendy Ayres-Bennett. 2011. Linguistics and philology in the twenty-first century: Introduction. Transactions of the Philological Society 109 (3). 201–206. Allen, Harold B. (ed.). 1973–1976. Linguistic atlas of the upper-midwest. Minneapolis: University of Minnesota Press. Avis, Walter S. 1967. Introduction. In Stefan Dollinger, Laurel Brinton & Margery Fee (eds.), DCHP-1 Online. 2013. http://dchp.ca/DCHP-1/pages/frontmatter (accessed 26 October 2015). Avis, Walter S., Charles Crate, Patrick Drysdale, Douglas Leechman, Matthew H. Scargill & Charles J. Lovell (eds.). 1967. A dictionary of Canadianisms on historical principles. Toronto: Gage. Bailey, Patricia G., Gerald Redmond & Jeremy Freeborn. 2010. Curling. In Canadian Encyclopedia. http://www.thecanadianencyclopedia.ca/en/article/curling/ (accessed 26 August 2015). Baker Publishing Website. 2015. http://bakerpublishinggroup.com/revell/about-revell (accessed 26 August 2015). But, Roxanne. 2013. The role of context in the meaning specification of cant and slang words in eighteenth-century English. In Andreas H. Jucker, Daniela Landert, Annina Seiler & Nicole Studer-Joho (eds.), Meaning in the history of English: Words and texts in context, 129–154. Amsterdam: Benjamins. Cassidy, Frederic G. 1973. The Atlas and DARE. In Christoph Schwarz & Dieter Wunderlich (eds.), Handbuch der Lexikologie, 91–95. Königstein: Athenäum. Cassidy, Frederic G. 1985. Introduction. In Frederick G. Cassidy & Joan Houston Hall (eds.), Dictionary of American regional English. Vol. 1. Introduction and A–C., xi–xxii. Cambridge, MA: Belknap Press.

Historical lexicography and historical linguistics as test cases

87

Cassidy, Frederick G. & Joan Houston Hall (eds.). 1985–2013. Dictionary of American regional English, vols. 1–6. Cambridge, MA: Belknap Press. Chambers, J. K. 1994. An introduction to dialect topography. English World-Wide 15. 35–53. City of Cambridge. 1997. Hall of Fame Members: Reverend Robert E. Knowles. http://www. cambridge.ca/cs_pubaccess/hall_of_fame.php?aid=28&cpid=33&scpid=0&did=0& sid=34 (accessed 27 October 2015). CONTE = Corpus of Early Ontario English. 2006. Stefan Dollinger (ed.). Vienna, Austria: University of Vienna, Institut für Anglistik und Amerikanistik. Craigie, William & James R. Hulbert (eds.). 1968 [1938–1944]. A dictionary of American English on historical principles. 4 vols. Chicago: University of Chicago Press. DARE. See Cassidy & Hall 1985–2013. DCHP-1. See Avis et al. 1967 DCHP-1 Online. See Dollinger et al. 2013. DCHP-2. See Dollinger & Fee Forthcoming. Dollinger, Stefan. 2004. Historical corpus compilation and “philological computing” vs. “philological outsourcing”: A LModE test case. VIEWS 13 (2). 3–23. Dollinger, Stefan. 2012. Canadian English in real-time perspective. In Alexander Bergs & Laurel Brinton (eds.), English historical linguistics: An international handbook, vol. 2. [HSK 34.2], 1858–1880. Berlin & New York: Mouton de Gruyter. Dollinger, Stefan. 2014. What the capitalization of nouns in early Canadian English may tell us about ‘colonial lag’ theory: methods and problems. http://www.academia.edu/4001684/ What_the_Capitalization_of_nouns_in_Early_Canadian_English_may_tell_us_about_ colonial_lag_theory_methods_and_problems (13 Jun. 2016). Dollinger, Stefan. 2015a. The written questionnaire in social dialectology: History, theory, practice. Companion website to IMPACT 40. Amsterdam/Philadelphia: Benjamins. Dollinger, Stefan. 2015b. How to write a historical dictionary: A sketch of The Dictionary of Canadianisms on Historical Principles, 2nd edn. Ozwords 24 (2). http://ozwords.org/. Dollinger, Stefan, Laurel Brinton & Margery Fee (eds.). 2013. DCHP-1 online: A Dictionary of Canadianisms on Historical Principles, 1st edn. Based on Walter S. Avis et al. 1967. www. dchp.ca/DCHP-1. Dollinger, Stefan & Margery Fee (eds.). Forthcoming. DCHP-2: The Dictionary of Canadianisms on Historical Principles, 2nd edn. With the assistance of Alexandra Gaylie, Baillie Ford & Gabrielle Lim. Vancouver and Gothenburg: University of British Columbia and University of Gothenburg. www.dchp.ca/dchp2. Ellis, Alexander John. 1871. On early English pronunciation, with especial reference to Shakespeare and Chaucer. London: Asher. Gordon, Matthew J. 2006. Interview with William Labov. Journal of English Linguistics 34 (4). 332–351. Gries, Stefan Thomas. 2009. Statistics for linguistics with R: A practical introduction. Berlin: Mouton de Gruyter. Grimm, Jacob. 1854. [Vorrede]. In Jacob Grimm and Wilhelm Grimm (eds.), Deutsches Wörterbuch, vol. I: A-Biermolke, ii–lxvii. Leipzig: S. Hirzel. Hall, Joan. 1997. Introduction: Forum: Dialect labelling in dictionaries. Dictionaries: Journal of the Dictionary Society of North America 18. 94–96. Harris, Robin S. & Terry G. Harris (eds.). 1994. The Eldon House diaries: Five women’s views of the 19th century. Toronto: The Champlain Society.

88

Stefan Dollinger

Hitchcock, Tim, Robert Shoemaker, Clive Emsley, Sharon Howard & Jamie McLaughlin (eds.). 2012. The Old Bailey proceedings online, 1674–1913, www.oldbaileyonline.org, version 7.0 (accessed 24 March 2015). Jucker, Andreas H., Daniela Landert, Annina Seiler & Nicole Studer-Joho (eds.). 2013. Uncovering layers of meaning in the history of the English language. In Andreas H. Jucker, Daniela Landert, Annina Seiler & Nicole Studer-Joho (eds.), Meaning in the history of English: Words and texts in context, 1–15. Amsterdam: Benjamins. Kay, Christian, Jane Roberts, Michael Samuels & Irené Wotherspoon (eds.). 2009. Historical thesaurus of the Oxford English Dictionary. With additional material from A thesaurus of Old English. Oxford: Oxford University Press. Kretzschmar, William A. 2015. Language and complex systems. Cambridge: Cambridge University Press. Kurath, Hans et al. 1972 [1939–1943]. Linguistic atlas of New England, 3 vols. New York: AMS Press. Kytö, Merja. 1996. Manual to the diachronic part of the Helsinki Corpus of English texts, 3rd edn. Helsinki: University of Helsinki. Labov, William. 1963. The social motivation of a sound change. Word 19. 273–309. Labov, William. 1972. Sociolinguistic patterns. Philadelphia, PA: University of Pennsylvania Press. Labov, William. 1994. Principles of linguistic change. Volume 1: Internal factors. Malden, MA: Blackwell. Labov, William. 2008. Triggering events. In Susan Fitzmaurice & Donka Minkova (eds.), Studies in the history of the English language IV: Empirical and analytical advances in the study of English language change, 11–54. Berlin: Mouton de Gruyter. Labov, William, Sharon Ash & Charles Boberg. 2006. Introduction. In Atlas of North American English, 3–10. Berlin: Mouton de Gruyter. Lass, Roger. 1969. Preface. In Roger Lass (ed.), Approaches to English historical linguistics: An anthology, v–viii. New York: Holt, Rinehard and Winston. Morpurgo Davies, Anna. 2011. Philology and linguistics: When data meet theory: Two case studies. I: The case of hieroglyphic Luwian. Transactions of the Philological Society 109 (3). 207–219. Murray, James A. H. 1987 [1888]. Preface to volume I of the New English Dictionary. In Dictionaries: Journal of the Dictionary Society of North America 9. 179–195. Müller, Max. 1862. Lectures on the science of language. Delivered at the Royal Institution of Great Britain, April, May, June 1861. New York: Scribner. http://www.gutenberg.org/files/ 32856/32856-pdf.pdf (accessed 16 February 2016). O’Grady, Jean. 1993. Special writer: An annotated bibliography of the writings of R. E. Knowles in the Toronto Daily Star and Star Weekly. Toronto: Colombo. OED. See Simpson & Proffitt. 2000–. Old Bailey Corpus = see Hitchcock, Shoemaker, Emsley, Howard, & McLaughlin. 2012. Osselton, Noel E. 1985. Spelling-book rules and the capitalization of nouns in the seventeenth and eighteenth centuries. In Mary-Jo Arn, Hanneke Wirtjes & Hans. Pepys, Samuel. 1971 [1873]. The diary of Samuel Pepys (unabridged). H. B. W. Brampton. Raleigh, NC: Hayes Barton Press. Pi, Chia-Yi Tony. 2000. Canadians telling time: A study in dialect topography. Toronto Working Papers in Linguistics 18. 80–102. Popper, Karl R. 1966 [1934]. Logik der Forschung. Zweite, erw. Auflage. Tübingen: Mohr (Siebeck).

Historical lexicography and historical linguistics as test cases

89

Renouf, Antoinette & Andrew Kehoe (eds.). 2006. The changing face of corpus linguistics. Proceedings of the 24th ICAME Conference, Guernsey, UK, April 2003. Amsterdam: Rodopi. Rissanen, Matti. 1989. Three problems connected with the use of diachronic corpora. ICAME Journal 13. 16–19. Ritt, Nikolaus. 2004. Selfish sounds and linguistic evolution: A Darwinian approach to language change. Cambridge: Cambridge University Press. Ritt, Nikolaus. 2016 [In press]. “Linguistic pragmatics from an evolutionary perspective”. In: Barron, A., Grundy, P. and Yueguo G. (eds.). Routledge Handbook of Pragmatics. London: Routledge. Schleicher, August. 1866. Compendium der vergleichenden Grammatik derindogermanischen Sprachen. Kurzer Abriß einer Laut- und Formenlehre deridg. Ursprache. 2. berichtigte und vermerte [sic!] und um gearbeitete Auflage. Weimar et al: Hermann Böhlau. Sheidlower, Jesse. 2011. How quotation paragraphs in historical dictionaries work: The Oxford English Dictionary. In Michael Adams & Anne Curzan (eds.), Contours of English and English language studies, 191–212. Ann Arbor: University of Michigan Press. Simpson, John & Michael Proffitt. 2000–. Oxford English Dictionary, 3rd edn. www.oed.com. Stockwell, Robert P. 1969. Foreword. In Roger Lass (ed.), Approaches to English historical linguistics: An anthology, iii–iv. New York: Holt, Rinehard and Winston. Trench, Richard Chenevix. 1857. Some deficiencies in our English dictionaries. London: John W. Parker. Turner, James. 2014. Philology: The forgotten origins of the modern humanities. Princeton, NJ: Princeton University Press. Weinreich, Uriel, William Labov & Marvin Herzog. 1968. Empirical foundations for a theory of language change. In Winfried P. Lehmann & Yakov Malkiel (eds.), Directions in historical linguistics, 95–195. Austin, TX: University of Texas Press. Whitney, William D. 1971. [1884]. On the narrative use of imperfect and perfect in the Brahmanas. In Michael Silverstein (ed.), Whitney on language: Selected writings of William Dwight Whitney, 306–335. Cambridge, MA: MIT Press.

II Particulars of authorship

Seth Lerer

The history of the English language and the history of English literature Writing in 1869, in his still-remarkable study, On Early English Pronunciation, with Especial Reference to Shakespeare and Chaucer, Alexander J. Ellis reflected on the changes that had made the sound of Shakespeare’s language alien to his contemporary readers. In an idiom resonant with earlier, literary comments on linguistic change – from Chaucer’s remarks, in Troilus and Criseyde, how words that once had “prys” now seem “nyce and straunge”, to Caxton’s meditations on the shifts of sound and variations of dialect that he could notice in his own lifetime – Ellis avers that the English of to-day do not know the tongue that Shakspere [sic] spake. They may be familiar with the words of his plays according to their own fashion of speech, but they know no more how Shakspere would have uttered them than they know how to write a play in his idiom. The language of Shakspere has departed from us, and has to be acquired as a new tongue, without the aid of a living teacher (1869: 22).

At first glance, Ellis’s reflections might seem but the expressions of a nowoxidized Victorian philology, grounded in an association between language and national identity and a privileging of Shakespeare as the embodiment of that association, all expressed in a heightened diction, part donnish lecture, part pulpit oratory. The history of the English language and the history of English literature – indeed, the history of Englishness itself – had defined the project of Victorian linguistic inquiry, and it is now a critical commonplace to find it in everything from Murray’s New English Dictionary through Sweet’s Anglo-Saxon Reader. Fifty years of modern scholarship, from Hans Aarsleff (1967) through Haruko Momma (2012), have exposed the ideologies of such philological nationalism, and to evoke the question of the history of language to the history of literature may be seen, now, dangerously to tread on such politically charged surfaces. Matthew Giancarlo, building on the work of Robert Stockwell and Donka Minkova on the Great Vowel Shift, invokes Ellis to show, with some nuance, that behind his history of language was always a history of literature. Ellis’s table of historical pronunciations – what Giancarlo calls “the ur-diagram” of the Great Vowel Shift – names linguistic periods as literary periods. Ellis offers not dates but names: Chaucer, Spenser, Shakespeare, Milton, Dryden, Pope, and Goldsmith. Giancarlo notes: “While it obviously provided Ellis’s audience (as it does for us) with what they wanted – a phonetic map for the great authors of the

94

Seth Lerer

past – it also tacitly reduces the English language to a single tradition and a unified dialect that implies a standard language uniting not only the literary tradition but also the entire language itself; and it comes before the technical analysis and summaries, effectively trumping much of their historical detail” (2001: 42). Giancarlo makes clear the ways in which Ellis’s work, and that of a host of other nineteenth-century philologists, effectively created the “Great Vowel Shift” as a unified, systematic, and rule-governed pattern of linguistic change. Rather than offering a fine-grained analysis of the stratifications of regional dialects, the patterns of scribal orthography, or the emerging patterns of “educated” or class-defined pronunciation, the nineteenth-century linguistic historians sought to offer a coherent, linear narrative. It was not that they did not know of the heterogeneity of earlier forms of English; it was, rather, that in response to the emerging Neogrammarian concern with sound law and in response, too, to the Victorian anxieties about dialect itself (articulated from the novels of Dickens to the plays of Shaw), the nineteenth-century came to privilege certain forms of literary evidence over others. Major authors stand as signposts in Ellis’s history of pronunciation, much as they stand in the Oxford English Dictionary (OED) as signposts in the history of lexis. Of course, the writing of a history of English through its literary monuments was not original with the OED; this was, explicitly, Johnson’s project in his Dictionary of 1755, and was, even then, already embedded in post-Augustan ideals of prosody and style. But what the later nineteenth-century philologists did that the eighteenth-century lexicographers, critics, and metaphysical etymologists could not was to develop an empirical methodology of linguistic change tout court – lexis, grammar, orthography, phonology – along the axes of literary authorship. Such an approach differentiates the OED’s use of major authors in particular from Johnson’s. For example, Johnson’s first edition has more quotations from Shakespeare than Milton; by the fourth edition the priorities of the two authors are reversed, largely as a result of Johnson’s own critical reevaluation of Milton (in particular Paradise Lost) and his growing purpose as Allen Reddick has put it “to provide a sacred connotation for a word or to change the timbre of the entry so that it is particularly controlled by the force of the Miltonic additions” (1996: 23). In the first edition of the OED, by contrast, the two leading major authors (after the King James Bible) are Shakespeare and Sir Walter Scott. This is not a decision based on taste or moral criticism or an ideal of the sublime. This is a decision based on Scott’s use of what were taken to be regional, dialectal, and archaic terms. Scott becomes a mine for the historical linguist and dialectologist. Indeed, when the editors of the OED began to incorporate American literary

The history of the English language and the history of English literature

95

authors into their examples, from the first through the third editions, the American equivalent of Scott – for precisely these same reasons – was Mark Twain. Scott and Twain stand, philologically, as self-conscious, literary repositories of language variety and change itself. It is not simply that they used unusual words. It is that their fictions took as their driving themes the nature of linguistic communication, the relationships among dialect, character, and social drama, and the challenges to written narrative to convey the variety and instability of the vernacular (Willinsky 1994; Lerer 2003). It is precisely this variety and instability of the vernacular that nineteenthcentury philology sought to confront. One of the fundamental paradoxes of the philological project lay in recognition that historical documents of the vernacular did not present a homogeneous history of sound and grammar; and yet, the pedagogical project of philology was precisely to enforce a sense of linguistic homogeneity period-by-period. Sweet’s creation of “Early West Saxon” and the editorial ministrations of a regularized “language of Chaucer” are but two examples of the attempt to regularize the forms of early English for largely pedagogical purposes. These projects were about more than spelling or phonology. In the case of Chaucer, as Tim Machan has recently illustrated, the editorial regularization of second-person pronoun usages creates the impression that the social stratification of the formal and informal in late fourteenth-century London was as regular as, say, the vous/tu distinction in nineteenth century Paris. What Machan concludes about Chaucer may be said about any one of a number of major literary authors: “Chaucer’s language has served as the epitome of Middle English, and in this way Chaucer offers a model for imagining a linguistic history of discrete periods with historical continuity among them” (2012: 175). These impulses are largely pedagogical in purpose: that is, they make the history of the English language teachable. In a sense, they answer Ellis’s concern: that earlier forms of the vernacular have to be acquired “as a new tongue, without the aid of a living teacher” (1869: 22). It is precisely with the later, nineteenthcentury rise of those living teachers – professional university philologists – that the history of the English language becomes a discipline in all senses of that word: not just a subject of academic study, but a lesson plan for rule-governed sound change, recoverable etymology, and rigorous textual criticism. My remarks to this point will be perceived as of a piece with a recent trend in philological historiography: exposing myths and ideologies behind historical linguistics, recognizing the political and institutional environments in which supposedly empirical methodologies developed, and exploring the often aesthetic criteria for editorial and lexicographical choice. Giancarlo, Machan, and Momma are but a few of the participants in this demystification of linguistic history. Richard Watts is another, and his recent Language Myths and the History of

96

Seth Lerer

English illustrates (to my mind, at least, a little ham-fistedly) a set of cultural and social presuppositions that have created the pedagogical axes of English as a language with antiquity, greatness, and relative homogeneity. My goal today is not to revisit these critiques, but rather to interrogate another, fundamental set of assumptions about literary and linguistic canonicity. Behind much historical philology and lexicography is a basic association of authorship with linguistic innovation. The canon of vernacular authorities takes many bases for its formation: the capturing of lived experience (e.g., Chaucer, Shakespeare, and Dickens); the elevation of moral principle (Milton); the articulations of style (Joyce); the brilliance of characterization (Twain); the codification of form and idiom (Austen). I could go on. As John Guillory put it, in a signal formulation that has governed a generation of critical thought, no single work or author is canonical in and of itself. Canonicity, instead, is a property of relationships of value, and each work has a place in the system of social and aesthetic valuation. Thus “major” authors are major because they participate in a system of literary authority. Such a system is also one of linguistic authority. Ever since fifteenth-century Chaucerians praised Chaucer for his “purifying” of the English language, linguistic innovation has become a key criterion for literary canonicity (Guillory 1987). It is precisely this association between authorial canonicity and linguistic innovation that has governed post-Johnsonian lexicography. As Charlotte Brewer has shown in great detail, the OED had, from its inception through its late twentieth-century supplements, been motivated to engage with literary sources. Literature was “central” to the history of the language and the making of the OED itself. R. W. Burchfield, the aegis of the OED’s Supplements, “preserved to the last” (in Brewer’s words) “his determination to quote from ‘great writers,’ on occasion expressing a strong sense of embattlement against the destructive practices of modern linguists, whom he characterized as ‘great marauding bands’, ‘scholars with shovels intent on burying the linguistic past and most of the literary past and present’” (Brewer 2007: 184–185). Any dictionary that could have a place for Orm should have a place for Edith Sitwell and W. H. Auden. Brewer has shown that this great “love for literary sources” prolonged the production of the Supplements, as if Burchfield himself could not let go of possible new coinages or usages (2007: 185). In this great love of literature, Burchfield, of course, was neither alone nor new. Nineteenth-century lexicographers grounded their work in writers who had altered English: in the words of Henry Bradley in 1904, those who influenced the language “either because of their boldness in the introduction of new words and new senses of words, and the extent to which their innovations have found acceptance, or because their writings have afforded abundant material for literary allusion” (1904: 235).

The history of the English language and the history of English literature

97

I would add another set of categories and another set of issues for assessing the relationships between authorship and linguistic change. These categories would include what I would call a vernacular self-consciousness: a recognition that any act of deliberately “literary” writing has designs on vernacular usage; that writing that aspires to literature also aspires to linguistic change; and that what has been called “style” represents a departure from discursive norms with the purpose of ultimately influencing or changing those norms. This last point represents no naïve return to the stylistics of the 1970s. It represents, instead, a recognition that stylistic choice expresses difference synchronically and diachronically. In other words, it represents not simply variation from a current norm; it represents engagement with past norms and, potentially, an aspiration to change future norms. One way of exploring this idea is to go back to Chaucer: to the notion, brilliantly articulated by Christopher Cannon, that Chaucer’s linguistic achievement was not, in fact, to add to the vocabulary of English literature or change the idioms of poetry, but rather to create the rhetorical pose of a linguistic innovator. Chaucer’s borrowings, distinctive usages, and juxtapositions of old and new words as Cannon has put it “help Chaucer’s English to gather the quality of novelty to itself and to present that novelty as constitutive of its own making”. Chaucer does not so much invent a new English as he invents the post of someone who invents a new English (Cannon 1998: 134). It is this pose that defines one version of the canon of major vernacular authors. Milton, Wordsworth, Dickens, Twain, Joyce, Norman Mailer, Toni Morrison, Salman Rushdie – each, in their own way, creates the impression of linguistic innovation not so much by genuinely coining new words or new phrases but by making us feel as if they did so. And, to some extent like Chaucer, they create the impression not only of linguistic innovation but of linguistic representation. What Chaucer managed to do, probably for the first time in vernacular English fiction, was to create the impression that people were talking. Now, we know of course that the characters in the Miller’s Tale did not speak in rhymed iambic pentameter couplets; we know that classical heroes did not speak in rhyme royal; we know that birds never spoke. Nonetheless, there is a rhetorical ability in Chaucer’s narrative poetry to create the impression not that we are reading someone writing but that we are listening to people speaking (Howard 1976: 384–385). This ability to evoke speech in fiction participates in a broader narrative of literature’s relationship to language change: a narrative that recognizes idiolect and variation; a narrative that recognizes that the sources of linguistic change lie, often, in the heterogeneity of spoken discourse. What many literary writers recognize, therefore, is what Uriel Weinreich, William Labov, and Marvin Herzog

98

Seth Lerer

recognized in 1968: language always varies from speaker to speaker and from time to time; it is that synchronic variation that, in essence, drives diachronic change (Weinreich, Labov, and Herzog 1968; Watts 2011: 47). A good, modern literary example of this phenomenon is Norman Mailer’s presentation of the speech of soldiers in his novel of World War II, The Naked and the Dead (1948). Early in the novel, Sergeant Brown and Private Stanley are discussing Brown’s wife. Sergeant Brown recognizes that his wife is not simply “gonna sit around and wait for me”. I know what she’s doing, and when I get back I’m going to have a little accounting with her. I’m going to ask her first, ‘Been having any dates?’ and if she says, ‘Yes,’ I’ll get the rest out of her in two minutes. And if she says, ‘no, honey, honest I haven’t, you know me,’ I’m just going to do a little checking with my friends and if I find she’s been lying, well, then I’ll have her, and, man, maybe I won’t give her some lumps before I kick her out. (Mailer 1948: 15)

Mailer evokes the vernacular of the barracks. But he also evokes the vernacular of the bedroom. The genius of this scene lies in its multiple layers of linguistic representation: the voice of the man imagining the voice of the woman. Just as Mailer’s job, as a novelist, is to create the impression of unmediated colloquial expression, it is Sergeant Brown’s job, as a storyteller, to create the impression of anticipated colloquial dialogue. Mailer records phrasings that, by the midtwentieth century, were understood as colloquialisms: the eye-dialect “gonna”; the hyper-correction of “have a little accounting”; the idiom “I’ll have her”. Sergeant Brown records phrasings that had become the defining idioms of intimacy: the use of “honey”, which though it goes back to Middle English as a term of endearment, had been revived in the twentieth century as a distinctive Americanism (OED, s.v. honey, n., def. 5.a); the use of the word “honest” to emphasize the truth of a statement, defined by the OED as a colloquial Americanism and first attested, in of all places, Twain’s Tom Sawyer (1876; OED, s.v. honest, adj. adv., def. 5. b). But the most innovative verbal moment in this speech is the use of the word man. The OED reports its history as a vocative of address or as a parenthetical expletive, and it locates that history as distinctively nonwhite: in the vocative, the Dictionary notes, “In 20th cent., in Caribbean English and among African Americans”; as an interrupting colloquialism, “formerly chiefly among AfricanAmericans and South Africans” (OED, s.v. man, n.1 (and int.), def. IV. 16. b). The OED’s many twentieth-century quotations of this use of man are often in black voices, as for example, this quotation from the 1933 issue of the magazine, Metronome: “Trum’s greeting was in the Negro dialect he usually employed, ‘Man! How is you?’” As far as I can tell, the first white author quoted by the

The history of the English language and the history of English literature

99

OED in these sections is from a 1969 poem of Allen Ginsberg, and though the OED quotes Zora Neale Hurston’s use of the idiom from 1934, Mailer may well be the first white fiction writer to use it (OED, s.v. man, n.1 (and int.), def. IV. 16. c). He renders it accessible for everyone: white, black, soldier, reader. And he reveals what wartime does to language: it brings dialects and idioms together, channels what would have been marginal expressions into mainstream. Soldiers, whether returning from the south, the western front, the Pacific, Southeast Asia, or the Middle East, bring back many things, and what they bring back is language changed irrevocably by contact with voices they would not have heard at home. Mailer evokes the language of a particular kind of talk in order to say something about how language varies and, in turn, how it changes. He illustrates how a term from one register (African-American) can be accepted into another. And he illustrates how it is the job of any good narrator, be it a novelist or a barracks sergeant, to evoke the sound of colloquial speech in the telling of a story. There may well be linguistic innovation here, but this is as much a rhetorical position as Chaucer’s. It illustrates a relationship between literary and linguistic history, lexicography and the social imagination. But that relationship differs from another kind of self-conscious, lexicographical imagination. Since the eighteenth century, literary authors wrote not just for the dictionary but from it. Samuel Johnson thought that one of the responsibilities of the “good poet” was to “unearth words that have been long hidden from the people’s view” – thus, the poet was, in himself, a historical linguist, but in addition, a provider of linguistic resources for the lexicographer. Since Johnson, poets have had complex relationships to dictionaries. Recent scholarship has shown, for example, that Emily Dickinson wrote with a copy of Noah Webster’s American Dictionary close at hand (Hallen and Harvey 1993; Deppman 2002; Lerer 2015). Many of her verbal collocations come directly from Webster’s entries, for example in this stanza: Perhaps you think me stooping I’m not ashamed of that Christ – stooped until He touched the Grave – Do those at Sacrament Commemorate Dishonor Or love annealed of love Until it bend as low as Death Redignified, above? (Dickinson no. 833, [1894] 1979: 631–632)

Webster had defined “stooping” as “Bending the body forward”, generating the association between stoop and bend in Dickinson. Looking up the words commemorate, dishonor, low, dignify, and above in Webster generates a set of

100

Seth Lerer

verbal associations that would have provoked Dickinson to collocate them here (Hallen and Harvey 1993: 133–134). The words commemorate, death, love, and Christ all appear in Webster’s definition of “sacrament” (Hallen and Harvey 1993: 134–135). And Webster’s etymology of anneal as coming from “to anoint with oil” (an etymology we now know to be bogus) may have led to a string of associations leading to Christ (Webster defines Christ as “the anointed”) (Hallen and Harvey 1993: 135). Of course, Dickinson’s poetry is more than the sum of her Websterian readings. But much like Webster, she captures a markedly nineteenth-century American view of language itself: a way of reading both the lexicons and literature of English and transforming them into imagined landscapes. As Webster said, “it is quite impossible to stop the progress of language”; he compared it to the Mississippi River, “the motion of which, at times is scarcely perceptible; yet even then it possesses a moment quite irresistible. Words and expressions will be forced into use in spite of all the exertions of all the writers of the world” (Webster 1828; quoted in Lerer 2015: 188) Another version of this lexicographical poetics lies in Auden, who practically lived with the OED since his undergraduate days. As a student at Christ Church, Oxford, he dazzled his contemporaries with resonant, dictionary words: glabrous, sordes, callipygous, peptonized, all of which showed up in his early poetry (Carpenter 1983: 66). He mined the OED throughout his life for archaic, challenging, or particularly redolent words. In turn, he personally advised R. W. Burchfield on adding words to the OED’s supplement (Brewer 2007: 190–197). Auden’s Oxford experience of Anglo-Saxon similarly filtered into his poetry, to the point where an early exercise like “Something is Bound to Happen” (1931) seeks to evoke the alliterative prosody of late Old English, ventriloquized by Gerard Manley Hopkins. Doom is dark and deeper than any sea-dingle. Upon what man it fall In spring, day-wishing flowers appearing, Avalanche sliding, white snow from rock-face, That he should leave his house, No cloud-soft hand can hold him, restraint by women; But ever that man goes Through place-keepers, through forest trees, A stranger to strangers over undried sea, Houses for fishes, suffocating water, Or lonely on fell as chat, By pot-holed becks A bird stone-haunting, an unquiet bird.

The history of the English language and the history of English literature

101

Look up sea-dingle in the OED and you will find the definition, “(now only arch.) an abyss or deep in the sea”, followed by a single quotation from the thirteenthcentury prose treatise, Sawles Warde, “His runes ant his domes þe derne beoð ant deopre þen eni sea dingle”, and then the quotation from Auden’s poem which, as is absolutely clear, is simply a modernization of those lines from Sawles Warde. It is as if Auden found the quotation in the first edition of the OED, lifted it wholesale, and then gets into the later editions of the OED. It is meaningless for the Dictionary to state, “now only arch.” for there is no “now” of usage. It exists in post-medieval English solely as a self-conscious lexicographical quotation (Bloomfield 1948; Toswell 1997; Jones 2006: 93). Auden’s poem stands as a fulcrum between editions of the dictionary: on the one hand, using the OED for archaic terms (sea-dingle, chat, stone-haunting, pot-holed, and beck); on the other hand, being later used by the OED to illustrate. Thus, the stanza I have quoted is the sole example for stone-haunting; it is one of five quotations for pot-holed. Clearly, the OED was also a source for some of the collocations in this poem. Even though fell and beck are archaisms and regionalisms (the former being Northern English for a moorland ridge; the latter meaning a brook or stream bed), Auden would have found the two words together in the OED’s quotation under fell from 1880, R. Broughton, Second Thoughts: “Fells and becks, whose cool memory has often come back to her”. These dictionary games with Auden, much as with Dickinson, illustrate the many ways in which lexicography generates poetic discourse and, in turn, how poetic usage generates lexicographical research. Auden may be extreme in his relationship to the OED, but it is clear that he had a deep sense not just of the word games of the dictionary but of the ways in which poetic diction may have a self-conscious historical linguistic sensibility. Auden, of course, was not the only twentieth-century writer who developed a vernacular, lexicographical self-consciousness. One could well argue that the project of High Modernism itself – from the classicism of Eliot, through the Celticism of Yeats, to the interlingualism of Pound – grew directly out of a renewed sense of the poetics of the lexicon and the history of the vernacular. Such a linguistic self-consciousness developed not only in response to new developments in lexicography but also in response to new developments in the anthology. The study and the teaching of both literature and language, from the mid-nineteenth century onwards, often proceeded through excerpts assembled into literary or linguistic narratives. Thus culture of the excerpt, as Leah Price (2001) has shown, had a profound impact on the rise of the English novel and the habits of Victorian reading. Indeed, one could well see the OED itself as nothing more, or less, than a collection of excerpts, “beauties” of literary language which, when read in sequence, offer something of a narrative of English.

102

Seth Lerer

Little wonder that Arnold Bennett, in 1928, called the OED “the longest sensational serial ever written”. To read that Dictionary was akin to reading other kinds of late Victorian narrative: journalism, book reviews, and encyclopedia entries. Such forms of discourse, in Price’s words, “encouraged readers and writers alike to think of texts as accumulations of freestanding beauties strung together by longer stretches of narrative padding” (2001: 139). To study the history of the English language was, as well, to see the strings of linguistic beauties. Sweet’s Anglo-Saxon Reader (1876), William Peacock’s English Prose from Mandeville to Ruskin (1903), and George Saintsbury’s History of English Prose Rhythm (1912) presented literary and linguistic history through excerpts chosen for their representative power. The politics of nineteenth-century anthologies shared in the promulgation of a national literary culture, an awareness of the social stratifications of dialect, and (not of least importance) a new Imperial market for university press textbooks (Gibson 2002: 173–174). It has long been recognized that James Joyce directly engaged with these various traditions in his “Oxen of the Sun” chapter in Ulysses – a chapter that narrates the various engagements among visitors to a hospital maternity ward through a series of imitative paragraphs, presented in a sequence miming the history of English prose (Gibson 2002: 173–74). Beginning with alliterative phrasings that evoke the sounds of Anglo-Saxon, Joyce moves through the Middle English of Everyman, Mandeville, and Malory before turning to imitations of Elizabethan historians, seventeenth-century Latinate prose, Bunyan, Addison, Goldsmith, Burke, Gibbon, Carlyle, Dickens, Ruskin, and a host of others, before fracturing into what Joyce himself called in a letter of 1920 “a frightful jumble of pidgin English . . . Cockney, Irish, Bowery slang and broken doggerel” (Ellmann 1976: 252). A great deal has been made of this chapter: as an “Irishization of things English” and as a “gleeful celebration of the ersatz” (Gibson 2002: 179–80; see also Gooch 1999). There is parody, overstatement, wit, and weirdness. Most obvious are the openings of the Anglo-Saxon section, “Before born babe bliss had”; the sinuous sentences of the Miltonic section; the characteristic uses of “so” and “by and by” in the Pepysian section; the rhetorical questions of the Lamb section (“What is the age of the soul of man?”); the repetitions of the Dickensian section (“it had been a weary weary while”) (Joyce [1922] 2002: 367, 378, 392, 399). The chapter offers, in the words of Andrew Gibson, “a tissue of contradictions,” a series of tensions between the historical and the modern, the English and the Irish, the story of a baby’s birth and the decline of a middle aged Leopold Bloom (2002: 182). The most powerful of those contradictions, however, lies in that between the written and the spoken. Joyce juxtaposes the history of English written

The history of the English language and the history of English literature

103

prose against a drama of contemporary Irish talk. The chapter begins and ends in voices: at the opening, with Latin prayer and English incantation (“deshil holles eamus . . . hoopsa, boyaboy, hoopsa!”); at the end, with expletives and nonsense (“Hark! Shut your obstropolos. Pflaap! Pflaap!”) (Joyce [1922] 2002: 366, 406). It is precisely this tension between the written and the spoken that makes this chapter a profound meditation on the study of the history of the English language. Particularly brilliant are the moments early in the chapter when he offers up the archaisms of Mandevillian prose: And the learning knight let pour forth childe Leopld a draught and halp thereto the while all they that were there drank every each. And childe Leopold did up his beaver for to pleasure him and took apertly somewhat in amity for he never drank no manner of mead which he then put by and anon full privily he voided the more part of his neighbor glass and his neighbor nist not of his wile. And he sat down in that castle with them for to rest him there awhile. Thanked be Almighty God. (Joyce [1922] 2002: 370)

The brilliance of this paragraph, in particular, lies not simply in its evocations of late Middle English or the humor of its medievalist exaggerations. The feel of this paragraph is the feel of the student confronting a passage in the classroom and trying to translate it. Reading word for word, the student begins by simply putting the old words in modern dress: keeping an idiom like “learning knight” in its original form; keeping the word order the same, as in “there drank every each”. This is not so much a parody or imitation of Mandeville as it is an imitation of a student trying to read Mandeville. It self-consciously makes us aware of the archaic quality of Mandeville’s own prose: its use of what by the late fifteenth century would have been recognizably old fashioned phrases, the kenning “learning knight” (a Biblicism that translated the Vulgate’s disciplulus) in particular. The challenge of evoking the history of English prose lies in the challenge of not just reading that prose but reading it aloud. Is this a history of writing or a history of talk? My question leads me to one of the most curious of anthologies, the 1953 collection edited by James Sutherland, The Oxford Book of English Talk. Presented as a collection of written documents illuminating “the voices of those men and women who once lived in that present which is now the past”, the book seeks to discover evidence of “authentic conversation” from the past (Sutherland 1953: vi). Sutherland recognizes that much of what passes for “talk” in literary prose and drama is evocation rather than transcription. Many of the texts he offers are more documentary than literary, at times even purporting to be accurate records of speeches and conversations. Reading through, we cannot but be struck by the variety: Elizabethan vagabonds; Charles I facing his accusers; Swift’s table-talk;

104

Seth Lerer

the drunken expostulations of Victorian Oxford undergraduates; the House of Commons debates after Munich. But for all of its variety, this is as much a document of cultural linguistic politics as anything in Saintsbury or Sweet. It offers up a sense of “Englishness” as rooted in the mouth – a definition of a people as a nation of speakers. The idea of the history of the English language garnered from this volume is one of people performing their lives, speaking the speeches as they have been pronounced. This is a book of national recovery, a book about a postwar England that had recently recovered from the Blitz, an England that had, just the year before its publication, seen a Queen new crowned, an England that begins with Margery Kempe called before the archbishop and that ends with a Mrs. Lillian Balch describing her holiday on the seaside for the BBC. Indeed, in this last passage, dated 1949, the book affirms that once again the English may take holidays at the seaside. “What do you think of that?” – the last words of the book (Sutherland 1953: 445). And yet behind these family pastorals lie threats and subversions. The arenas of speech can be the royal trial as much as the radio studio. The very nature of the English legal system lay in talk: in the vernaculars of testimony, in the readings of charges, in the bearing of verbal witness. Charles I faces his accusers in Parliament, smiling “often during the time, especially at the words tyrant, traytor, murtherer and publique enemy of the Commonwealth” (Sutherland 1953: 108). It is as if we are presented with a lexicon of royal failure, a dictionary of the despot to which Charles can only smile. “Remember”, he avers, “I am your King, your lawful King”, who can address the court: “Well, sir, remember that the King is not suffer’d to give his reasons for the liberty and freedome of all his subjects”. The Lord President responds, “Sir, you are not to have liberty to use this language” (Sutherland 1953: 108, 113). Whether these passages record the true speech of the King and Parliamentary judges is, to me, far less important than the content of their dialogue. For this is an exchange about English itself: about the words that cut a character, about the liberty to use a language. Sutherland’s volume is a volume about liberty and language, and it offers, in the end, less a record of how people spoke than a reflection on the powers of that speech to grant the English liberty and nationhood. The speakers in the House of Commons after the Munich agreement of 1938 constantly call attention to their act of speaking: “I am so glad to be able to deliver a speech which I confess I am . . . bursting to deliver” (Sutherland 1953: 405). So, too, the RAF pilot in 1940 cannot wait to open his mouth: “I want to tell you of a fight a few days ago” (Sutherland 1953: 416). Reading The Oxford Book of English Talk against “The Oxen of the Sun” shows us the tensions and temptations of finding a voice behind a written word. It shows us that, among the various relationships of literary and linguistic

The history of the English language and the history of English literature

105

history, one of the central challenges is to use literature as evidence for speech; but also to recognize that speech itself is voiced and literary. I began with a reflection on some recent critiques of the Great Vowel Shift and with the recognition that, for much of traditional philology, it was the history of sound change that defined the history of the language. What I suggest then, in closing, is that the history of English literature be understood, not simply as evidence for linguistic change, but as a participant in that change; that the attempt to evoke human speech, be it in Chaucer or in Norman Mailer, is a self-conscious act of vernacular assertion; that the mining of dictionaries for poetic words goes on not simply in the silent study but in public spaces (how else can we understand the anecdote about Auden using his OED terms among classmates); and that any literary text, real or imagined, takes on a new life when read aloud. Joyce took the liberty upon himself to use the language, and those liberties we may both see and hear the interplay among the history of English and the history of literature.

References Aarsleff, Hans. 1967. The Study of Language in England, 1780–1860. Princeton: Princeton University Press. Bennett, Arnold. 1928. Books and persons. Evening Standard, 5 January. Bloomfield, Morton. 1948. “Doom is dark and deeper than any sea-dingle”: W. H. Auden and Sawles Warde. Modern Language Notes 58. 548–552. Bradley, Henry. 1904. The making of English. New York: Macmillan. Brewer, Charlotte 2007. Treasure-house of language: The living OED. New Haven: Yale University Press. Cannon, Christopher. 1998. The making of Chaucer’s English: A study of words. Cambridge: Cambridge University Press. Carpenter, Humphrey. 1983. W. H. Auden: A biography. London: Unwin. Deppman, Jed. 2002. “I could not have defined the change”: Reading Emily Dickinson’s definition poetry. The Emily Dickinson Journal 11. 49–80. Dickinson, Emily. 1979. The poems of Emily Dickinson. Thomas H. Johnson (ed.). Cambridge, MA: Belknap Press of Harvard University Press. Ellis, Alexander J. 1869. On early English pronunciation, with especial reference to Shakespere [sic] and Chaucer. London: Asher & Co. Ellmann, Richard. 1976. Selected letters of James Joyce. New York: Viking. Giancarlo, Matthew. 2001. The rise and fall of the Great Vowel Shift: The changing intersections of philology, historical linguistics, and literary history. Representations 76. 27–60. Gibson, Andrew. 2002. Joyce’s revenge: History, politics, and aesthetics in Ulysses. Oxford: Oxford University Press. Gooch, Michael. 1999. Saintsbury’s Anglo-Saxon in Joyce’s “Oxen of the Sun”. Journal of Modern Literature 22. 401–404.

106

Seth Lerer

Guillory, John. 1987. Canonical and non-canonical: A critique of the debate. ELH 54. 483–527. Hallen, Cynthia L. & Lavra M. Harvey. 1993. Translation and the Emily Dickinson lexicon. Emily Dickinson Journal 2 (2). 136–146. Howard, Donald R. 1976. The idea of the Canterbury Tales. Berkeley: University of California Press. Jones, Chris. 2006. Strange likeness: The use of Old English in twentieth century poetry. Oxford: Oxford University Press. Joyce, James. 2002 [1922]. Ulysses. Mineola, NY: Dover. Lerer, Seth. 2015. Inventing English: A portable history of the language, revised and expanded edn. New York: Columbia University Press. Edition. Lerer, Seth. 2003. “Hello, dude”: Philology, performance, and technology in Mark Twain’s Connecticut Yankee. American Literary History 15. 471–503. Machan, Tim William. 2012. Chaucer and the history of English. Speculum 87. 147–175. Mailer, Norman. 1948. The naked and the dead. New York: Reinhart. Momma, Haruko. 2012. From philology to English studies: Language and culture in the nineteenth century. Cambridge: Cambridge University Press. Peacock, William. 1903. English prose from Mandeville to Ruskin. London: Grant Richards. Price, Leah. 2001. The anthology and the rise of the novel. Cambridge: Cambridge University Press. Reddick, Allen. 1996. The making of Johnson’s Dictionary 1746–1773, revised edn. Cambridge: Cambridge University Press. Saintsbury, George. 1912. A history of English prose rhythm. London: Macmillan. Sutherland, James. 1953. The Oxford book of English talk. Oxford: Clarendon Press. Sweet, Henry. 1876. Anglo-Saxon reader in prose and verse. Oxford: Oxford University Press. Toswell, M. J. 1997. Auden and Anglo-Saxon. Medieval English Studies Newsletter 37. 22. Watts, Richard. 2011. Language myths and the history of English. Oxford: Oxford University Press. Webster, Noah. 1967 [1828]. An American Dictionary of the English Language. Foundation for American Christian Education: Anaheim. Weinreich, Uriel, William Labov & Marvin Herzog. 1968. Empirical foundations for a theory of language change. In W. Lehmann and Y. Malkiel (eds.), Directions for Historical Linguistics, 95–195. Austin: University of Texas Press. Willinsky, John. 1994. Empire of words: The reign of the OED. Princeton: Princeton University Press.

Xingzhong Li

“Of harmes two, the lesse is for to chese”: An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse 1 Introduction1 In Chaucer’s historical romance Troilus and Criseyde, when Criseyde learns from his sly uncle, Pandarus, of Troilus’s infatuation with her and potential death of lovesickness, she was on the one hand fearful of falling into disgrace exchanging letters with the unseen warrior and on the other hand concerned about his fatal destiny. It was at this juncture that she figures “Of harmes two, the lesse is for to chese” (Well, of two evils one should choose the less) (Tr2.470, Coghill 1971: 61)2 and weighs Troilus’ fatalistic death above her own potential disgrace, expressible as DISGRACE > DEATH . This weighing principle in a way underlines the contemporary linguistic optimality theory (OT): Candidates compete through a set of strictly ranked evaluative, violable weighted constraints, and the winner eventually emerges as the optimal finalist. This study adopts OT and Maxent Grammars (Hayes, Wilson, and Shisco 2012) and integrates them to an OT-Maxent approach in a simplified manner to account for constraints governing Chaucer’s syntactic inversions with a purpose to uncover Chaucer’s underlying metrical principles. Whereas in OT, higherranked constraint violations rule out the generated output, in Maxent Grammars penalty scores derived from violations of mathematically weighted constraints identify the optimal output.3 1 This article grew out of a talk entitled “‘Last Monday’ or ‘Monday Last’? An Optimality Treatment of Syntactic Inversions in Chaucer’s Verse,” written for SHEL-8 in September 2013. I am grateful to Donka Minkova for inspiring questions and comments at SHEL-8. My thanks are also due to Colette Moore, who read the early version of the paper with helpful comments, to Miranda Wilcox, who meticulously inserted the translations on top of her professional editorial work, and in particular to Don Chapman and an anonymous reviewer, whose insightful critiques and invaluable advice helped me improve the organization and quality of the paper. Errors, however, are entirely my responsibility. 2 This line has since become the earliest attested source of the modern proverb “Of two evils, choose the prettier” (Mieder, Kingsbury, and Harder 1992: 186). 3 For more information on OT, see Prince and Smolensky (2004); for metrical studies under OT, see Golston (1995, 1998), Hayes and MacEachern (1998), Hayes (2000), Fitzgerald (2007), and Youmans (2009); for more information on Maxent Grammars, see Hayes and Wilson (2008), Hayes, Wilson, and Shisko (2012).

108

Xingzhong Li

Verse lines with non–syntax-driven syntactic inversions are formed with marked word order, and reversions of the inversions restore the lines’ unmarked word order. Restored lines, along with actual lines, can serve as generated output for evaluation and as evidence that brings to light a poet’s metrical principles. Such is the assumption behind the word-order test (WOT) in metrical studies pioneered by Youmans in 1983. Well recognized in metrics literature,4 WOT has been recently recast in a constraint-based OT approach (Fitzgerald 2007, Youmans 2009) as pioneer work in accounting for syntactic inversions in English poetry, but not yet in an integrated OT-Maxent model. This article is organized as follows. I first explain the establishment of a data corpus of syntactic inversions by Chaucer. Next I review current OT treatments to metrics. I then use the data corpus to identify and rank constraints governing syntactic inversions for selection of optimal lines. To conclude, I briefly compare the findings in this study with two earlier metrical studies and discuss the implications of the ranked constraints for Chaucer’s metrics.

2 The corpus of data My data corpus is a systematic 3,020-line sample (every tenth line) drawn from Chaucer’s entire iambic pentameter canon in Benson 1987. In aggregate, I identified 1,450 lines (48%) that contain syntactic inversions. Of them 56 are either syntax-driven (e.g., question formation), run across multiple lines, as in (1), or form fossilized expressions, as in (2) below. (1)

They moste take ti in pacience at nyght [ti Swiche manere necessaries as been plesynges To folk that han ywedded hem with rynges], ‘They’ve got to take in patience at night Such customary necessary things As please the folk who wedded them with rings’. (MLT 710–713, Wright 1985: 133)

(2)

a.

Quod the turtel, “If it be youre wille ‘Said the turtle dove, “If it’s your will. . .”’ (PF 510; Kline)

b.

“Tehee!” quod she, and clapte the wyndow to, ‘“Ha-ha!” she laughed, and clapped the window to’ (MilT 3740, Wright 1985: 95)

4 As attested in Kiparsky 1989; Hayes 1989; Li 1995, 2004, 2008; Youmans & Li 2002.

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

109

Prose paraphrases in unmarked word order of inversions such as in (1) inevitably affect rhyme, syllable counts, stress patterning, as well as line boundaries and thus typically have gross effects on metricality. Such inversions offer little useful evidence about metrical principles. Similarly, the verb quod, as in (2), which occurs some 850 times in the entirety of Chaucer’s iambic pentameter verse, according to my search of the online Chaucer Concordance (University of Maine at Machias 2007), always appears in the inverted order, thus forming a case of syntactic subregularity and behaving like a wild bull rushing about in a metrical china shop. Consequently, I excluded all the 56 lines, retaining instead the rest of 1,394 lines in which inversions are confined to single lines, such as these in (3) below: (3)

a.

That may [ti thyn herte] setten ti in quiete? ‘That will cause your heart no disquiet’ (Tr4 490, Coghill 1971: 196)

b.

Tyl [tj dounward] [ti went] the sonne ti tj wonder faste. ‘Till downward went the sun wondrously fast’ (PF 490, Kline)

c.

[tk Ful prively] [tj two harneys] [ti hath] he ti dight tj tk, ‘Prepares two sets of armour secretly’ (KnT 1630, Wright 1985: 42)

Lines of this type can include one or more inversions within them: (3a) contains one, (3b) two, and (3c) three. In counting lines with inversions, I treated lines like (3b) and (3c) as two because the five inversions distribute in two lines. Inversions like these provide evidence about subtle metrical principles, such as the constraints regularizing rhyme and governing phrasing and the patterning of stressed and unstressed syllables. My scansion of these lines focused on determination of metrical effects – positive, negative, neutral – as correlated with other factors, including syllable count, phrasal structures, extra and extrametrical syllables, elisions and resolutions, stress matching and foot types, caesuras and alignments of linguistic and metrical boundaries. The results of such analyses constituted bases for establishing constraints that run across phonology, semantics, metrics, and syntax.

3 Current OT treatments to metrical constraints and their interactions The focus of this line of research is on how meter, syntax, and phonology interact in individual poets’ metrical systems. Below, I sketch three representative OT treatments:

110 (4)

Xingzhong Li

Representative OT Treatments to Metrics a. Metrical Constraints > Syntactic Constraints (Golston 1995, 1998; Fitzgerald 1994; Rice, 1997) b. Syntactic Constraints > Metrical Constraints > Syntactic Constraints (Fitzgerald 2007) c. Phonology, Syntax (Youmans 2009)

> Meter

> Syntax

The rank in (4a) states that when metrical and syntactic constraints conflict, the latter yields. However, the rank in (4b) claims that because not all syntactic constructions are subject to inversion, metrical constraints must dominate parts of syntax and be dominated by other parts of syntax, or in Fitzgerald’s term, syntactic and metrical constraints must be “interleaved” (2007: 205). As a result, Fitzgerald proposes the following strictly ranked OT constraints to characterize Shakespeare’s sonnets: (5)

SPEC LFT >> MATCH >> *CLASH >> HEAD LFT, ADJ-N (214–215)

Here, SPEC LFT outranks metrical constraints MATCH (stress) and (stress) *CLASH , both of which in turn outrank the two syntactic constraints HEAD LFT and ADJ-N. It follows that the two metrical constraints can trigger inversions of HEAD LFT and ADJ -N but not SPEC LFT. The rank in (4c) differs from (4b) in that phonology (rhyme) is now part of the rank. Specifically, Youmans presents (6) to account for constraints governing syntactic inversions in rhymed verse by Chaucer, Shakespeare, and Milton: (6)

Art+N!, Rhyme! > Metre > Adj+N > V+Comp > AdvP (2009: 118)

Despite the different names of constraints, (6) is more comprehensive and more accurate in comparison to (5). Briefly, not only syntax and metrics interleave, but phonology is also part of the interleaving process. In addition, only Art+N! outranks meter but not SPEC LFT, which covers too wide a range of phrasal categories of which Art+N! is only one of them. This cautious treatment, as it turns out toward the end of this article, describes Chaucer’s metrics better than (5). Besides, (6) addresses AdvP that (5) leaves out.

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

111

4 Constraints and their interactions in Chaucer’s iambic pentameter verse This section begins with discussing phonological constraints, then semantic constraints, followed by metrical constraints, and finally syntactic constraints.

4.1 Phonological constraints As mentioned earlier, my data corpus, derived from analyzing 3,020 sampled lines drawn from the entirety of Chaucer’s iambic pentameter verse, includes 1,394 lines that contain in-line syntactic inversions. Of them, 1,021 (73%) reinforce rhyme scheme, and none of them shifts a rhyming word away from the rhyming position, leaving 373 lines with syntactic inversions driven by other constraints. This is strong evidence that RHYME! (RM !) is an inviolable constraint in Chaucer’s verse, as the exclamation mark indicates. As a phonological constraint, RM ! enforces inversion, as pointed out by Gascoigne as early as in 1575: You shall do very well to use your verse after the English phrase, and not after the manner of other languages . . . if we should say in English a woman fair, a house high, etc. it would have but small grace: for we say a good man, and not a man good, etc. . . . Therefore even as I have advised you to place all words in their natural or most common and usual pronunciation, so would I wish you to frame all sentences in their mother phrase and proper Idióma, and yet sometimes the contrary may be borne, but that is rather where rhyme enforceth, or per licientiam Poëticam, than it is otherwise lawful or commendable. ([1575] 1868: 37)

The lines in (7) and my rewrites of them (those preceded with left-headed arrows) help to see how rhyme enforces inversion: (7)

a.

. x . . . x . x . x His resons, as I may my rymes holde , ‘What he said, . . . if my rhyme will hold’ (holde: olde, Tr3.90-91, Coghill 1971: 114) ⇐

b.

(my rewrite)

x . . [holde my

x rymes],

. x . x . . . x . x The trompes come, and in his baner larg e ‘The trumpeters come, and on his great banner’ (large: charge, Anel 30) . x . x ⇐ [his large baner]

112

Xingzhong Li

Typically, inversions affecting rhyme occur in the second hemistich of verse lines. In (7a), the inversion overrides syntax so as to regularize the WS-rhythm, as indicated by “.” and “x”, and maps the rhyme to the end of the line. In cases such as this, it is clear that both rhyme and rhythm outrank syntax, but it is not as immediately clear as to which of them triggers the inversion. However, inversions such as in (7b) move the rhyming word without affecting the stress patterning, and it is thus certain that it is R M ! but not rhythm that induces the inversion. Coupled by the fact that Chaucer’s verse is 100% rhymed while not all of his metical feet are iambic, one must conclude that R M ! is a categorically definitive and inviolable constraint. In a nutshell, R M ! overrides rhythm (R THM ), which in turn overrides syntax (S NTX ), schematically expressed in (8a) and mathematically represented in (8b): (8)

a. R M ! > RTHM > SNTX b. 0% > 20% > 50%

The percentages are rates of constraint violations in (7a) and (7b) combined. Specifically, for RTHM two pyrrhic feet of the ten supposedly iambic feet equal to 20% violations, and for SNTX two of the four hemistiches are affected and yield a 50% violation rate. Thus, an OT tableau, like (9), expresses how candidates are evaluated by weighted constraints and statistically compete for the optimal form, and a finger symbol points to the optimal candidate: (9) IntMLT2.95–96 (-bake : make) . ☞I w ‘I’ll . ⇐ I w

x . x in pros e, spek e s w s speak in prose; it’s up to x spek e s

. in w

x pros e, s

RM ! RTHM SNTX

. and w him to

x . x . x lat him rymes make . s w s w s make the poetry’ (Wright 1985: 115)

. and w

x lat s

. him w

x mak e s

x . rymes.”5 w s

*

*

*

The higher a constraint is ranked, the more weighted it becomes, the lower the probability of lines that violate it, and the higher the penalty score of a violation. As a result, violations of higher-ranked constraints induce “a gradient reduction in the overall level of [candidates’] well-formedness” (Hayes, Wilson, and Shisco 2012: 695). 5 The word-final -e is elided with a following vowel but can also be elided with a following sonorant, in this case, the /r/ in rhymes.

113

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

The syllable count (S YL C NT ) makes another strong phonological constraint, manifested in the fact that a line with unmarked word order, when containing an extra in-line syllable, mostly the final -e, is nearly always inverted, shifting that syllable to a position for elision. The line in (10) illustrates this process: (10)

x x . x . . . e til that the So long abid ‘just wait at least till night is over’ (TR2.990, Coghill 1971: 80) ⇐

. x Abid

x so

x . longe

x til

x nyght

. x departe ,

...

The final -e in the rewrite can’t be elided with the following initial /t/ and thus creates an antibacchius with a disruptive extra syllable, but the inversion elides the final -e with the following schwa. Chaucer is highly conscious of syllable count and strongly tends to void three-syllable feet. After all, only 34 (0.2%) of 15,100 feet in the 3,020-line sample take the form of [. . x] (Li 1995: 236). This fact means that SYL CNT is a nearly inviolable constraint, that is, Chaucer follows it in 99.8% of all his metrical feet, with a violation rate as low as 0.2%. As such, it should be immediately ranked after RM !, but as a violable constraint. In the current study, I have found 38 (10.2%) of the 373 lines that prevent actual lines from containing extra syllables, and (11) provides another example. (11)

RM ! SYL CNT RTHM SNTX a. ☞ . x . x . x . x . x Troie town, The Grekys stronge aboute ‘The Greeks were ranged, encamped about Troy town’ (Tr4.30, Coghill 1971: 180) b. ⇐ . The

x . x . . x . stronge Grekys aboute

x . Troie

x town,

*

*

*

Here, the violation of SYL CNT simultaneously disrupts the rhythm, hence RTHM is also violated. According to Minkova (1991), by 1400 final schwas had most likely been lost in England (30), but metrical considerations played a role in retaining or dropping them (Minkova 1987). In my scansion, I counted them as syllables when they are immediately before initial consonants in following words and elide them when they are adjacent to following vowels, as the superscripted -e has indicated. As shown in (11), the inversion prevents the actual line

114

Xingzhong Li

from violating SYL CNT by shifting the final -e to the eliding position and reducing syllable count by one. If it were not for SYL CNT, metricality in the rewrite would get very close to the optimal line because it contains a canonical unmarked word order with stress patterning nearly identical to the actual line.

4.2 Semantic constraints Semantic constraints are often symbioses with other linguistic and metrical constraints and remain structurally obscure. Theoretically, all inversions with marked word order automatically shift semantic foci to varying degrees, but there seems to be no need to set up an independent semantic constraint because other constraints are often so structurally conspicuous that constitute sufficient reason to account for inversions. This happens especially when inversions have positive metrical effect, shown in (12): (12)

a.

. x . x . x . . To grounde ded she falleth as ‘To the ground she falls dead as a stone’ (Anel 170)

. a

x ston;

⇐

. x . She falleth

. a

x ston;

b.

. . . x . x . x . e e Of which I confort caught , and went in ‘From which I took comfort, and entered fast’ (PF 170, Kline)

x faste.

⇐

. Of

x faste .

. which

. to

. I

x . x grounde ded

. as

x . x . . x . caughte confort, and went e in

The rewrite in (12a) disrupts the stress patterning and creates a pyrrhic and two trochees against the iambic norm, and the first trochee violates Kiparsky’s Monosyllabic Word Constraint (1977), a metrical principle that requires all major stresses in W metrical positions to be monosyllables unless these stresses are immediately preceded by phrase boundaries. The rewrite in (12b) results in a three-syllable amphibrach, violating SYL CNT. However, Chaucer’s lines undeniably re-mark the semantic foci to grounde dead and confort. Among the 373 non-rhyme inversions in my data corpus are 340 cases (91%) that have positive metrical effect, but simultaneously they also all re-mark foci. Thus, they all

115

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

constitute strong evidence for metrical principles as well as semantic principles, albeit not direct structural evidence for independent semantic constraints. Fortunately, structurally independent semantic constraints do surface when inversions have negative metrical effect. The line in (13) is such an example. (13)

SYL RM ! FCS CNT RTHM SNTX x . . x . x . x . a. ☞ Blak was his berd, and manly was his ‘Black was his beard, and powerful his face’ (KnT(1)2130; Wright 1985: 54)

x face;

. b. ⇐ His

x . x . x . x . berd was blak, and manly was his

x face;

. c. ⇐ His

x . x . . berd was blak, and his

x . . x . face was manly;

*

*

*!

**

**

*

*

*

Chaucer’s actual line remarks the focus on blak at the sacrifice of one violation of RTHM and two violations of SNTX . That is, focus remarking overrides metrical regularity and syntactic norm. If this is right, we ought to take focus (FCS ) as an independent preferential semantic constraint and rank it above RTHM . Just because metrical and syntactic constraints tend to structurally obscure FCS , as when inversions have positive metrical effect, does not mean FCS is not an effective semantic constraint. I propose to rank FCS equally with RM ! because all inversions more or less re-mark foci, though I would treat it not as an inviolable constraint but as a preferential one just because it is in general not as structurally evident as RM ! is. The actual line in (13a) contains two inversions. First, R M ! drives face to the rhyming position (Trace : face, KnT(1) 2129–2130), extrametricalizes the final -e to avoid violating SYL CNT, and transforms the second hemistich to the adjective + be + subject word order. Then, a domino effect seemingly brings about the inversion in the first hemistich for a parallelism. In this light, parallelism plays just a secondary role and thus functions not as a primary constraint. This reasoning is further supported by two additional observations. First, if FCS were not a semantic constraint, (13b) would emerge as optimal, which surely is not the case. Of course, one could substitute parallelism for FCS for an identical effect, but that does not explain the second observation: similar inversions in the first hemistich would still occur when there appears no call for parallelism with the second hemistich, as shown in (14):

116

Xingzhong Li

(14) a. x . . x Strong was the wyn, and wel to drynke us ‘Strong was wine; we matched it with our thirst’. (GP 750, Wright 1985: 20) b. x . . x Crul was his heer, and as the ‘His hair was curly, and like gold it shone’ (MilT 3314, Wright 1985: 84)

gold

c. x . . x Go we to-nyght to Rome, and we ‘Let us go to Rome tonight, and we shall see’ (LGW 1710)

leste.

it

shoon,

shal se.

These inversions, obviously not triggered by parallelism, conspicuously re-mark the foci at the beginnings of the lines. Eleven (3%) of the 373 lines in our data corpus have negative metrical effect, and most of them exhibit similar patterns as in (13) and (14). The point to draw is that negativity of metrical effect renders FCS prominent, so is neutrality of metrical effect. Of the 373 lines, 22 (6%) have neutral effect, as shown in (15): (15)

SYL RM ! FCS CNT RTHM SNTX . x . x . x . x . x a. ☞ Of this and that they pleid e, and gonnen wad e ‘They joked on this and that, and there were sallies’ (Tr2.150, Coghill 1971: 50) . x . ⇐ They pleid e of

x . x this and that,

. x . x and gonnen wad e

. . . x . x . x thy tast, b. ☞ For thow of lov e hast lost ‘For you for love have lost your taste, I guess’ (PF 160, Kline) . ⇐ For

. . x . thow hast lost thy

x tast

. of

x lov e,

*

*

. x I gess e,

. x I gess e,

*

*

*

Without FCS , both rewrites in (15) would emerge as winners. Later in section 4.3 below, the line in (19) offers an even stronger argument for FCS to be treated as an indispensable thematic constraint.

*

117

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

4.3 Metrical constraints Thus far, we have contented ourselves with RTHM as an umbrella metrical constraint. However, earlier research findings and empirical facts in my data corpus suggest we break RTHM to two constraints: BFES “beginning free and ending strict” and LH “a rising rhythm from low to high.” BFES finds its ground in the concept that metrical positions in iambic pentameter are not created equal; that is, the fifth foot is the most constrained position, the second or third foot the next most, the rest the least, depending on in-line divisions – usually 4//6 or 6//4. Correspondingly, metrical irregularities tend to occur line- or hemistichinitially, and metrical regularities line- or hemistich-finally (Youmans 1989: 346–349; Li 1995: 50–55; Youmans and Li 2002: 153–155). This concept is both phonologically and metrically supported by the strong closure principle.6 It is not a surprise that when in-line divisions segment a line, they also create more opportunities for BFES to apply. Working in tandem with BFES is LH. It regularizes stress patterning, and if necessary BFES shifts irregularities to domain-initial positions. There are two other familiar metrical constraints popular in metrical literature, namely, *CLASH (avoid two adjacent stresses) and MATCH (match stress with strong positions),7 but I choose not to use them because LH alone subsumes both of them and works along with BFES more effectively, as illustrated in the comparison between (16a) with (16b) below: (16)

SYL RM ! FCS CNT MCH *CLSH SNTX x x . x . x . x . x a. ☞ Al peynted was the wal, in length e and bred e ‘Frescoed throughout its entire length and breadth’ (KnT 1970, Wright 1971: 50) . x ⇐ The wal

. x x . . x . x was al peynted in length e and bred e

*

*

*

**

*

*

6 Line-final positions have been understood to reflect metrical patterns more strictly than elsewhere in the line (Bridges 1921: 40–41; Sprott 1953: 102; Youmans 1983: 76, 1989: 370–372; Hayes, Wilson, and Shisko 2012: 702, 708; Li 1995: 224–225, 2004: 316–322). Chomsky and Halle’s Nuclear Stress Rule (1968: 102–103) also accounts for this strong-closure principle. 7 *CLASH appears in studies of metrical phonology (Nespor and Vogel 1989, reprinted 2007, Prince 1983, Selkirk 1984), and so does MATCH (Hayes and MacEachern 1998, Hayes 2000, Fitzgerald 2007). The latter replicates Kiparsky’s MWC to dispel stress in polysyllables from occupying W positions.

118

Xingzhong Li

SYL RM ! FCS CNT BFES LH SNTX x x . x . x b. ☞ Al peynted was the wal, (KnT 1970) . x ⇐ The wal

. x . x in length e and bred e

*

. x x . . x . x was al peynted in length e and bred e

*

*

*

**

Comparatively, LH in (16b) alone does the work of both MATCH and *CLASH and is hence a more economic yet more powerful constraint. In fact, there are 124 (33%) of the 373 lines in the corpus that do not call for MATCH and *CLASH but BFES and LH, with BFES alone explaining 108 (29%) of the 373 sampled lines. Three more representative examples follow. (17)

SYL RM ! FCS CNT BFES LH SNTX . x . x . x . . . x a. ☞ Ye wolden on me rew e, er that I deyd e! ‘You would take pity on me before I died!’ (Tr1.460, Coghill 1971: 19) . x . x . . b. ⇐ Ye wolden rew e on me,

. . . x er that I deyd e!

*

*

*

*

**

Two points emerge in lines like this. First, BFES plays a critical role of avoiding a weak and irregular ending of the first hemistich while tolerating the loose beginning of the second hemistich in (17a). Second, the BFES -driven inversion promotes the stressless enclitic on in (17b) to a stress stronger than its neighbors in (17a). Indeed, inversions frequently play such a stress-promoting role, here and elsewhere.8 Now, look at (18). (18) x x . x . x . x . x a. ☞ Out brest the blood with stierne stremes red e; ‘And out in harsh red floods the blood now spurts’ (KnT 2610, Wright 1985: 66)

SYL RM ! FCS CNT BFES LH SNTX * *

8 See Li (1995: 219–220) and Youmans (1996: 194) for more information about clitic promotion.

119

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

. x x x b. ⇐ The blood brest out

. with

x . x . x stierne stremes red e;

. x x x c. ⇐ The blood brest out

. with

x . rede

x . x stierne strem es;

. x . x . x . x . x d. ⇐ The blood with rede stierne stremes brest

x out;

*!

*

*!

*

*

*

*

*

*

*

*

*

The tableau shows that (18d) violates all five higher-ranked constraints and that (18c) is only slightly better – it does not violate SL CT. By contrast, BFES plays a decisive role between (18a) and (18b). Without it, Chaucer’s actual line in (18a) would have fought a losing battle. BFES , however, can sometimes be violated in actual lines, as illustrated in (19). (19)

SYL RM ! FCS CNT BFES LH SNTX . x . x x . . x . x And blisful God prey ich with good entent e, ‘And I pray God in all sincerity’ (Tr2.1060, Coghill 1971: 82) . . ⇐ And ich

x x . x . x . x prey blisful God with good entent e,

*

*

**

*

**

This line illustrates two points, one old and one new. The old is that the rewrite would be selected as optimal without FCS , once more showing it to be an independent and indispensable semantic constraint. The new point is that BFES can be violated, albeit infrequently, as the first hemistich in the rewrite has an irregular beginning, whereas the same hemistich in the actual line has an irregular ending. Speculatively, Chaucer might elide prey ich and leave a momentary pause between the stressed God and the elided preyich with an initial stress to offset the falling rhythm. Lines such as this, however, merely indicate that constraints, unlike rules, are violable. In my corpus, the violation rate of BFES is pretty low: about 8%. By contrast, the violation rate of LH is higher: about 12%. Given the 8:12 ratio, ranking BFES above LH best catches Chaucer’s metrical practice. Thus far, the five higher-ranked constraints are all well supported by statistics of their violation rates, summarized in (20): (20)

a.

RM !, FCS > SL CT > BFES > LH

b.

0%, 0% > .2% > 8%

> SNTX

> 12% > ?

To fill up the unknown percentage under SNTX , we are required to scrutinize syntactic constraints governing inversions.

120

Xingzhong Li

4.4 Syntactic constraints Since Chaucer employs syntactic inversions in service of higher-ranked constraints, it will be interesting to investigate, first, whether that service is also statistically supported and, second, whether certain syntactic constraints outrank any higher-ranked constraints, as the two earlier studies in (5) and (6) above differentially claim. If yes, does Chaucer interleave constraints in the same way other poets do or in different ways? To establish syntactic constraints, we must first rely on a theoretical framework of syntax to categorize unmarked word order. Minimalist syntax and linguistic typology provide such a combinative framework. In keeping with the tenets of the framework, I classified syntactic inversions into three constraints – SPC L (specifier-left), HD L (head-left), and AJT R (adjunct-right) by virtue of word order, guided by the structural representation of phrases in (21) that I put together. (21)

A Skeletal Structure of Phrases9

9 A more complete skeletal representation than (21) would have to include more phrasal categories (e.g., NumP, vP, etc.). Also, depending on verb subcategories, their complements may

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

121

Then, I translated (21) into (22), (23), and (24) below, with examples showing unmarked word order. I assumed that once constraints governing unmarked word order were established, violations of the constraints in inversions with marked word order would naturally emerge. (22)

S PC L-constrained examples, where left-most specifiers are underlined: a. Spec-C’ (preposed WH-phrase as Spec): That if gold ruste, what shal iren do? ‘If gold can rust, then what will iron do?’ (GP 500, Wright 1985: 13) b. Spec-I’ (subject as Spec): He may nat fleen it, thogh he sholde be deed ‘Though it should cost his life, there’s no escaping’ (KnT 1170, Wright 1985: 30) c. Spec-Neg’ (negator as Spec): I am so confus that I kan noght seye ‘I am so confused I can only say’ (KnT 2230, Wright 1985: 57) d. Spec-V’ (pre-v adv as Spec): After that day we hadden never debaat. ‘From that day on we had no more debate’ (WBT 822, Wright 1985: 239) e. Spec-D’ (DegP as Spec): So strange a creature unto his make ‘So strange a foreign creature for his mate’ (MLT 700, Wright 1985: 133) f. Spec-N’ (attrib. adj as Spec): “Iwys, my deere herte, I am nought wroth, ‘I am not angry with you, love; have done!’ (Tr3.1110, Coghill 1971: 150) g. Spec-P’ (DegP as Spec): Right as an aspes leef she gan to quake, ‘Lay trembling, like an aspen leaf she shook’ (Tr3.1200, Coghill 1971: 153)

differ from DP. For our purpose, (21) is sufficient. For more information concerning NumP, see Cinque (1994: 87); for more information about DP, see Abney (1987); for detailed discussions of minimalist syntax, see Radford (2004), among others.

122 (23)

Xingzhong Li

HD L-constrained samples, where left-most heads are underlined: a. C-CP (Aux as head): This false world – allas! – who may it leve? ‘But O, there’s no believing this false world’ (Tr2.420, Coghill 1971: 59) b. I-IP (I as head): And she shal out; thus seyden here and howne. ‘“And she shall go” they shouted up and down’ (Tr4.210, Coghill 1971: 186) c. Neg-NegP (pre-v Neg as head): I ne aughte nat for that thing hym despise ‘Can I despise him? I don’t think I should’ (Tr2.720, Coghill 1971: 70)10 d. V-VP (V as head): And seyde, “Nece, but ye helpe us now, ‘And, “Niece,” he said, “if you don’t help us now . . .”’ (Tr3.1100, Coghill 1971: 150) e. D-DP (D as head): So myrie a fit ne hadde she nat ful yoore; ‘It’s years since she had had so good a bout’ (RvT 4230, Wright 1985: 107)11 f. N-NP (N as head): Now faire brother, beth al hool, I preye! ‘I say, my handsome brother must not come to harm!’ (Tr2.1670, Coghill 1971: 104)

10 In certain aspects, Chaucerian negation differs from that in Modern English. Whereas in Modern English the node Neg is left empty in a syntactic tree and the negator not serves as Spec projecting to NegP, Chaucer used nat (or not, naught, nought, noght, nevere) as the adverbial SPEC of NegP and ne as the head of NegP. Then, along with the verb-raising process, ne is also raised together to I, forming the ne-v-nat construction, as in I ne aughte nat for that thing hym despise (Tr2.720). Coupled with other negative words as quantifiers, conjunctions, correlative conjunctions, negation in Chaucer is more complex than presented here. For more information, see Travis (1984); Chomsky (1995: 129–166); Radford (2004: 162–163). 11 . DPs like so myrie a fit contains a phrasal specifier so myrie named DegP. See Abney (1987) and Radford (2004) for more information.

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

123

g. P-PP (P as head): In remembraunce of hire and in honour ‘In remembrance and honor of her’ (LGW F 530)12 Central in most theories of syntax and semantics is the argument-adjunct distinction. Though the diagram in (21) represents major argument structures, it leaves out adjuncts – structurally dispensable elements of a sentence. A complete account for Chaucer’s syntactic inversions, however, must put adjuncts into consideration. The lines in (24) contain adjuncts of various sorts in unmarked word order, where adjuncts are placed in brackets: (24)

AJT R-constrained examples a. I woot [right wel], thou darst it nat withseyn. ‘Which you dare not deny, as I well know’ (KnT 1140, Wright 1985: 30) b. And, romynge [on the clyves [by the se]], ‘roaming upon the cliffs by the sea’ (LGW F 1470) c. He knew the tavernes [wel] [in every toun] ‘He knew the taverns well in every town’ (GP 240, Wright 1985: 7) d. Come soupen [with hym] [in his hous] [at eve]. ‘Come that very night to sup with him at home’ (a rewrite of Tr3.560, Coghill 1971: 130)

As shown, adjuncts typically appear at the right edge of a sentence, hence the constraint AJT R to govern the placement of modifiers. The placement of adverbial adjuncts is relatively flexible, making AJT R a weak constraint and thus suggesting a high violation rate and penalty score. However, it turns out

12 The embedded PP of hire is an N-complement. Unlikely in Modern English, Chaucer can prepose embedded complement PPs: For [ti of this world] the feyth ti is al agoon (Tr2.410). Then this preposition violates *HD L because the head of the containing DP is D, namely the.

124

Xingzhong Li

not to be the case in my corpus. Only a minority of Chaucer’s lines contain multiple adjuncts, nominal or adverbial; the majority of his lines contain just one adjunct, if any. Perhaps, the length of the ten-syllable line of iambic pentameter restricts a luxurious use of adjuncts because arguments, which are structurally indispensable, must take priority over adjuncts within the boundary of a line. As seen above, minimalist syntax differs from earlier work of generative grammar in several aspects. For example, inverting the order of subject/aux violates *SPC L, so do inversions of subject/verb, adjective/noun, and more. Just as these inversions violate *SPC L, inversions of aux/v, v/o, v/predicate, n/complement, and more, violate *HD L. We will soon provide examples below to show how these constraints work, but first we must solve the issue of how to rank SPC L, HD L, and AJT R and determine whether they can mathematically fit into our OT-Maxent model. Statistics derived out of a detailed analysis of constraint violations in the 373-line data corpus provide a revealing answer in (25) and (26). (25)

Violations of the Three Syntactic Constraints Name *AJT R *SPC L *HD L

number of violations 65 139 189

percentage 17% 35% 48%

Total

393

100%

These percentages stipulate the ranking in (26): (26)

Ranking of Syntactic Constraints *AJT R > *SPC L > *HD L

With this answer, we are ready to look at examples and see how the three constraints fit into the overall OT-Maxent model. The line in (27) contains two inversions: the predicative orloge before the copula is, and the modified noun thorpes before the adjective lyte. The first inversion violates *HD L, and the second *SPC L, in service of higher-ranked constraints. Since there are two inversions, they shift FCS twice.

125

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

(27)

NP/VL, N/Adj

SYL RM ! FCS CNT BFES LH AJT R SPC L HD L

a. ☞ . x . x . x . x . x of thorpes lyte; The kok, that orlog e is ‘The rooster, clock to hamlets at first light’ (PF 350, Kline) . x . . b. ⇐ The kok, that is

x . . x . x orlog e of thorpes lyte;

. x . . c. ⇐ The kok, that is

x . . x . orlog e of lyte

x thorpes;

*

*!

*

***

**

***

*

*

Thus (27c) violates FCS two times, hence the penalty score of two asterisks. In (27b), the violation of SPC L reduces the penalty score of FCS down to one. In (27a), the violations of the two lower-ranked syntactic constraints remove all the violations of the three relevant higher-ranked constraints, thus making (27a) emerge as the optimal choice. Particularly worth noting are inversions of quantifier/noun and ordinal/ noun, as in (28). (28) noun/quantifier

SYL RM ! FCS CNT BFES LH AJT R SPC L HD L

. x . x . x . x . x a. ☞ With shipes sevene and with no mor e navy e;13 ‘With seven ships, and no larger fleet’ (LGW 960)

*

. x x . . . . x . x b. ⇐ With sevene shipes and with no mor e navy e;

*

. x . x . . . . x . x c. ⇐ With sev ene shipes and with no mor e navy e;

*

***

*

*

13 This line contains two sonorant sequences that substantiate elisions: sev ene and; no mor e navye. See Halle and Keyser (1971) for more information about position-sonorant sequence matching (169). In addition, the line-final word navye, loaned from Old French and rhyming with hye, is stressed on its second syllable.

126

Xingzhong Li

(29) noun/ordinal

SYL RM ! FCS CNT BFES LH AJT R SPC L HD L

. x . x . x . x . x a. ☞ That now, on Monday last, I saugh hym wirche. ‘One who, on Monday last, I saw at work’ (MilT 3430, Wright 1985: 87) . x . x b. ⇐ That now, on last

x . . x . x Monday, I saugh hym wirch e.

*

*

*

**

In (28), the quantifier sevene serves as the head of the DP sevene ships, so the D-Comp inversion violates *HD L. In that connection, the ordinal last, the head of the DP last Monday in (29), also inverts with its complement Monday, violating *HD L. Examples like these show the tremendous power of *HD L since one violation of it can save all other violations of higher-ranked constraints sometimes. No wonder up to 48% of all inversions violate *HD L and come naturally under Chaucer’s pen. First- and second-person possessive pronouns can also invert with their following nouns in Chaucer, but never the third-person possessive pronouns. The table below summarizes the behaviors of possessive pronouns in my data corpus. (30)

Potentiality of Inversions between Possessive Pronouns and Nouns Person

number of inversions

total cases

percentage

1st

4

27

15%

2nd

1

16

6%

3rd

0

45

0%

The third-person possessive pronouns are not the only word category impervious to inversion; two other subcategories of DP–art+n and dem+n–are also so impervious. In the 373-line sample are 222 DPs containing articles, demonstratives, possessive pronouns, quantifiers, and ordinals, and their behaviors differ in whether they are open or closed to inversion, as evidenced in (31):

127

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

(31)

Potentiality of Inversions between Possessive Pronouns and Nouns Subcategory q+n PossPro1,2 + n Total

Cases 20 43 63

Inverted 45% 12% 22%

Un-inverted 55% 88% 78%

dem + n PossPro3 + n art + n Total

35 45 79 159

0% 0% 0% 0%

100% 100% 100% 100%

PossPro1 = first person; PossPro2 = second person; PossPro3 = third person. These statistics evidentially point to a bifurcation of DPs in Chaucer’s hand. That is, invertible DPs violate *HD L; un-invertible DPs respect HD L. For clarification, I name the respected HD L DAP HD L! (DAP for dem., art., poss. pron.). Since DAP HD L! is never violated, suggesting that it is inviolable, I add “!” to indicate that its violation generates a fatal effect. Consequently, it must be equally ranked with RM ! and FCS . To this point, we have established nine constraints, which must be ranked as in (32) below: (32)

Final Rank of the Nine Constraints DAP HD L!,

0%

RM !,

FCS ! > SLCT > BFES > LH

= 0% = 0% > .2%

> 8%

> AJT R > SPC L > HD L

> 12% > 17% > 35% > 48%

This rank is firmly backed up by a gradient continuum of mathematical values from low to high. Again, the more weighted a constraint is, the lower its violation rate is. That is just the hallmark of this OT-Maxent model. As a final test, I apply (32) to lines with multiple inversions. The line in (3c) above makes a good candidate since it contains three inversions violating all three syntactic constraints. (33)

noun/ordinal DAP HD L!

SYL RM ! FCS CNT BFES LH AJT R SPC L HD L

x x . . x x . x . x a. ☞ Ful prively two harneys hath he dight, ‘Prepares two sets of armour secretly’ (KnT 1630, Wright 1985: 42) . . b. ⇐ He hath

x x dight two

x . x x . . harneys ful prively,

*3 = 3 violations; *6 = 6 violations.

* ***

*

*3

**

*6

*

*

*

128

Xingzhong Li

Note that (33b) violates FCS thrice since it does not reflect the three new foci the three distinct inversions induce. It also violates BFES twice and LH six times. In aggregate, it generates as many as twelve violations of the four higher-ranked constraints. By contrast, (33a) reduces the penalty score not only down to seven but also under the last five lower-ranked constraints whose strength gradiently diminishes. Criseyde could not wait to assess: Of harmes two, the lesse is for to chese; Of foules twelve, the fewer is for to blesse.

5 Concluding remarks In this article, we have been concerned with discovering Chaucer’s metrical principles through scrutinizing the causes of his syntactic inversions. We began by setting up a systematic data corpus of 3,020 sampled lines and identified 1,450 lines (48%) that contain syntactic inversions. Dropping lines (56, 4%) unusable as tests, we analyzed lines (1,021, 73%) whose inversions are constrained by RM ! and lines (373, 27%) whose inversions are driven by other factors. Finally, we came up with a rank of nine interleaved constraints crossing core linguistic areas – phonology, semantics, metrics, and syntax. Due to different names of constraints (though they are readily convertible), (5), (6), and (32) can hardly be directly compared. Due to the limit of space, I won’t compare them in detail. Suffice it to say that the three ranks progressively cover increasing numbers of constraints, offer an increasing amount of information, provide more and more detailed descriptions of metrics, and hence have increasingly stronger predictive power. To reiterate, our final product in (32) is strictly ranked according to the mathematical values from our data corpus to express rates of constraint violations under an integrated OT-Maxent model. Capable of catching the underlying causes of Chaucer’s syntactic inversions, our list of strictly ranked constraints represents Chaucer’s intricate metrical system, a crucible in which both violable and inviolable constraints crossing multiple linguistic areas are sorted and alchemized, and through which Chaucer *his verse endows with *structures innovative, *rhythm pleasant, *rhyme catchy, *focus prominent, yet *freedom enormous.

References Abney, Steven P. 1987. The English noun phrase in its sentential aspect. Cambridge, MA: Massachusetts Institute of Technology PhD dissertation. Benson, Larry D. (ed.) 1987. The riverside Chaucer, 3rd edn. Boston: Houghton Mifflin Company.

An integrated OT-Maxent approach to syntactic inversions in Chaucer’s verse

129

Bridges, Robert. 1921. Milton’s prosidy: With a chapter on accentual verse. London: Oxford University Press. Chaucer Concordance. 2007. University of Maine at Machias. http://ummutility.umm.maine. edu/necastro/chaucer/concordance/. Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Cinque, Guglielmo. 1994. Evidence for partial N-movement in the Romance DP. In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi & Raffaella Zanuttini (eds.), Towards universal grammar: Studies in honor of Richard Kayne, 85–110. Washington DC: Georgetown University Press. Coghill, Nevill. 1971. Chaucer. Troilus and Criseyde. New York: Penguin. Fitzgerald, Colleen M. 1994. Prosody drives the syntax. In Susanne Gahl, Andy Dobey, & Christopher Johnson (eds.), Proceedings of the twentieth annual meeting of the Berkeley Linguistic Society, 173–183. Berkeley, CA: Berkeley Linguistics Society. Fitzgerald, Colleen M. 2007. An optimality treatment of syntactic inversions in English verse. Language Sciences 29. 203–217. Gascoigne, George. 1868 [1575]. Certain notes of instruction. In Edward Arber (ed.), English Reprints, 10–12. London. Golston, Chris. 1995. Syntax outranks phonology. Phonology 12. 343–368. Golston, Chris. 1998. Constaint-based metrics. Natural Language and Linguistic Theory 16. 719–770. Halle, Morris & Samuel J. Keyser. 1971. English stress: Its form, its growth, and its role in verse. New York: Harper and Row. Hayes, Bruce. 1989. The prosodic hierarchy in meter. In Paul Kiparsky & Gilbert Youmans (eds.), Phonetics and phonology I: Rhythm and meter, 201–260. San Diego: Academic Press. Hayes, Bruce. 2000. Gradient well-formedness in Optimality Theory. In Joost Dekkers, Frank van de Leeuw & Jeroen van de Weijer (eds.), Optimality Theory: Phonology, Syntax, and Acquisition, 88–120. New York: Oxford University Press. Hayes, Bruce & M. MacEachern. 1998. Quatrain form in English folk verse. Language 74. 473– 507. Hayes, Bruce & Colin Wilson. 2008. A maximum entropy model of phonotactics and phonotactic learning. Linguistic Inquiry 39. 379–440. Hayes, Bruce, Colin Wilson & Anne Shisko. 2012. Maxent grammars for the metrics of Shakespeare and Milton. Language 28 (4). 691–731. Kiparsky, Paul. 1977. The rhythmic structure of English verse. Linguistic inquiry 8. 189–248. Kiparsky, Paul. 1989. Sprung rhythm. In Paul Kiparsky & Gilbert Youmans (eds.), Phonetics and phonology I: Rhythm and meter, 305–340. San Diego: Academic Press. Kline, A. S. (trans.). 2007. Geoffrey Chaucer. The Parliament of Fowls. www.poetryintranslation. com/PITBR/English/Fowls.htm (accessed 30-31 March 2016). Li, Xingzhong. 1995. Chaucer’s meters. Columbia, MO: University of Missouri (unpublished) PhD dissertation. Li, Xingzhong. 2004. A central metrical prototype for English iambic tetrameter verse: Evidence from Chaucer’s octosyllabic lines. In Anne Curzen & Kimberly Emmons (eds.), Studies in the history of the English language II: Unfolding conversations, 315–341. Berlin: Mouton de Gruyter.

130

Xingzhong Li

Li, Xingzhong. 2008. Metrical evidence: Did Chaucer translate The Romaunt of the Rose? In Susan M. Fitzmaurice & Donka Minkova (eds.), Studies in the history of the English language IV: Empirical and analytical advances in the study of the English language change, 155–179. Berlin: Mouton de Gruyter. Mieder, Wolfgang, Steward A. Kinsbury & Kelsie B. Harder (eds.). 1992. A dictionary of American proverbs. New York: Oxford University Press. Minkova, Donka. 1987. The prosodic character of early schwa deletion in English. In Anna Giacalone-Ramat, Onofrio Carruba, & Giuliano Bernini (eds.), Papers from the seventh international conference on historical linguistics, 445–459. Amsterdam: John Benjamins. Minkova, Donka. 1991. The history of final vowels in English. Berlin: Mouton de Gruyter. McCully, C. B. & J. J. Anderson (eds.). 1996. English historical metrics. Cambridge: Cambridge University Press. Nespor, Marina & Irene Vogel. 2007. Prosodic Phonology, 2nd edn. Berlin: Mouton de Gruyter. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14 (1). 19–100. Prince, Alan & Paul Smolensky. 2004. Optimality theory: Constraint interaction in generative grammar. Malden, MA: Blackwell. Radford, Andrew. 2004. Minimalist syntax: Exploring the structure of English. Cambridge: Cambridge University Press. Rice, C. 1997. Ranking components: The grammar of poetry. In Geert Booij & Jeroen van de Weijer (eds.), Phonology in progress – Progress in phonology, 321–332. The Hague: Holland Academic Graphics. Selkirk, Elisabeth O. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Sprout, S. Ernest. 1953. Milton’s art of prosody. Oxford: Blackwell. Travis, Lisa. 1984. Parameters and the effects of word order variation. Cambridge, MA: Massachusetts Institute of Technology (unpublished) PhD dissertation. Wright, David (trans.). 1985. Geoffrey Chaucer. The Canterbury Tales. Oxford: Oxford University Press. Youmans, Gilbert. 1983. Generative tests for generative meter. Language 59. 67–92. Youmans, Gilbert. 1989. Milton’s meter. In Paul Kiparsky & Gilbert Youmans (eds.), Phonetics and phonology I: Rhythm and meter, 341–379. San Diego: Academic Press. Youmans, Gilbert. 1996. Reconsidering Chaucer’s prosody. In C. B. McCully & J. J. Anderson (eds.), English Historical Metrics, 185–209. Cambridge: Cambridge University Press. Youmans, Gilbert. 2009. For all this werlde ryche: Syntactic inversions as evidence for metrical principles in the alliterative Morte Arthure. In Judith Jefferson & Ad Putter (eds.), Approaches to the Metres of Alliterative Verse, 115–133. [Leeds Texts and Monographs: New Series 17]. Leeds: Leeds Studies in English. Youmans, Gilbert & Xingzhong Li. 2002. Chaucer: Folk poet or littérateur? In Donka Minkova & Robert Stockwell (eds.), Studies in the history of the English language: A millennial perspective, 153–175. Berlin: Mouton de Gruyter.

Mark Davies and Don Chapman

The effect of representativeness and size in historical corpora: An empirical study of changes in lexical frequency 1 Introduction There are at least two basic considerations in the creation and use of a corpus, especially historical corpora. First, the corpus should be large enough to accurately reflect what was really happening in the language at a particular point in time. It would probably not make sense to create a corpus from just ten or twenty texts – maybe 20,000 to 40,000 words of text in total – and expect such a small corpus to accurately reflect the entirety of the language. Secondly, the texts will ideally be “representative” of the entire universe of texts for a particular period – and hopefully even the actual language of that period, including informal genres like spoken language (see Biber 1990, 1993; Leech 2006). If we create a corpus that is composed strictly of newspapers, for example, then we may find out a great deal about the language of newspapers over time, but this may have very little to do with other genres, or the language as a whole. In this paper, we will consider both the issue of size and representativeness by looking at (primarily) lexical data from three historical corpora. We will first examine the issue of size. Can a relatively small but well-designed corpus yield data on lexical frequency that is similar to that of a much larger corpus? Second, we will compare the lexical data from the two large corpora. Do they yield similar lexical frequency data, in spite of the fact that one corpus is designed to be representative while the other is not? The answers to these questions may be surprising to some researchers.

2 The corpora In this study, we will compare the following three corpora: – The Brown family of corpora: A Standard Corpus of Present-Day Edited American English (Brown), The Lancaster-Olso/Bergen Corpus (LOB), The Freiburg-Brown Corpus (Frown), and The Freiburg-LOB Corpus of British English (FLOB). These corpora were designed to be very representative of

132

–

–

Mark Davies and Don Chapman

the language as a whole, but they are rather small in size. The four corpora are Brown (American texts from 1961), LOB (British, 1961), Frown (American, 1991), and FLOB (British, 1991). Each of the four corpora contains one million words of text from 500 different texts (2,000 words each), with essentially the same genres and domains – one half of which is “imaginative” (= fiction) and the other half of which is “informational” (= nonfiction). The Corpus of Historical American English (COHA; corpus.byu.edu/coha). COHA was released in 2010, and it was designed to be both large and representative (see Davies 2012a, 2012b). It contains four hundred million words of texts in more than 100,000 different texts from the 1810s to the 2000s, including at least 10 million words in each decade from the 1830s on, and at least twenty million words in each decade from the 1880s on. Overall, it is about one hundred times as large as the four combined corpora in the Brown family of corpora. It is designed to be representative as well. From the 1870s on, each decade has essentially the same proportion of fiction, popular magazines, newspapers, and nonfiction books (for a total of two hundred million words from fiction, a hundred million from popular magazines, forty million from newspapers, and sixty million from non-fiction books). In addition, the corpus was carefully designed to be balanced and representative at the level of sub-genres and domains as well. For example, the sixty million words of text from nonfiction books have texts from twenty distinct Library of Congress categories (e.g., religion, history, “domestic arts”, agriculture, and engineering), and the balance between these categories stays essentially the same from decade to decade. Google Books (N-grams; books.google.com/ngrams). In terms of size, Google Books n-grams (hereafter “Google Books”) dwarfs any other carefully constructed corpus. For American English alone, it is based on more than 155 billion words of text. Whereas COHA is one hundred times as large as the Brown family of corpora, Google Books is about four hundred times as large as COHA. On the other hand, Google Books was not designed to be representative; the Google Books team simply scanned everything they could find in several large university libraries. This resulted in more than fifteen million books scanned, five million of which were processed to form Google Books.

In summary, then, we have three historical corpora, which will form the basis for the comparisons in this study. The Brown family of corpora was designed to be small but representative. COHA was designed to be both large and representative. And Google Books is very large, but it makes no pretension of being representative.

An empirical study of changes in lexical frequency

133

3 The importance of size English historical linguistics has a strong tradition of small, well-designed corpora, in the range of one to five million words each. These include the Brown family of corpora – one million words each in Brown (US 1960s), LOB (UK 1960s), Frown (US 1990s), and FLOB (UK 1990s). They also include ARCHER (1.8 million words, 1650–1999), CONCE (Corpus of Nineteenth Century Texts) (1 million words, UK 1800s), and the Helsinki Corpus of English Texts (1.6 million words, Old English through the early 1700s). These are all “general” historical corpora – they cover a wide range of genres and (like COHA) are balanced by genre from decade to decade. There are also many small corpora of particular genres, such as letters, newspapers, or court proceedings. These small corpora have certainly proven their value in research on highfrequency syntactic constructions, such as modals and other auxiliaries, pronouns, and prepositions, where even in one million words there might be hundreds or even thousands of tokens. But much less has been done – or can be done – in terms of lexical change, where there are just a handful of tokens for most words. A few studies have attempted to use these smaller corpora in looking at changes in lexis (Hofland and Johansson 1982; Leech and Fallon 1992; Oakes and Farrow 2007; Baron, Rayson and Archer 2009; and Baker 2010, 2011). But as one of the most active researchers in this field notes (Baker 2011: 70): Leech and Fallon (1992) point out that the corpora in the Brown family contain only about 50,000 word types in total, which is relatively small for lexical research, and that the majority of words will be too infrequent to give reliable guidance on British and American uses of language. For that reason, this study focuses only on frequent words in the corpora. It was stipulated that for a word to be of interest to this study, it would need to occur at least 1,000 times when its frequencies in all four corpora were added together. Three hundred eighty words met this criteria, but a number of high frequency words (e.g., class, miss, black, true, and English) were excluded because they missed the cutoff.

In this section, we will continue Baker’s line of research and show empirically what types of lexical data we can extract from a small 2- to 4-million-word historical corpus compared to a much larger corpus like COHA. As a test case, we will briefly consider adjectives that have (at least) doubled in (normalized) frequency in COHA from the 1960s to the 1990s, and then we will examine how well the one million–word Brown and Frown corpora (US, 1960s and 1990s) provide comparable evidence for this increase in frequency. In other words, in the data below we will be considering adjectives like overall, emerging, and motivated, whose charts in COHA are shown in Figure 1.

134

Mark Davies and Don Chapman

Figure 1: COHA: Adjectives doubling in frequency, 1960s–1990s

Table 1 shows that in COHA there are fifteen adjectives that have a combined frequency of between 800 and 1,600 tokens in COHA in the 1960s and 1990s (words such as overall [shown above], amazing, long-term, and alternative) and which have at least doubled in frequency during this time. There are another 127 types with a frequency of between 200 and 400 tokens in COHA in these two decades (e.g., emerging [shown above], compelling, indoor, preferred, and unclear) and 394 types with a frequency of between 50 and 100 tokens (e.g., motivated [shown in Figure 1], first-time, blurry, impaired, viral, obnoxious, and luscious). Table 1: Evidence for increase in adjective frequency, COHA, and Brown family COHA: Token range

800–1600

200–400

50–100

COHA: # of types

15

127

394

0

0

8

114

1–9

1

46

264

# Brown/Frown tokens

Brown/Frown “correct”

> = 10

Support

6

50

12

> = 10

???

5

15

0

> = 10

Contradict

3

8

4

0.39

0.03

0.40

Table 1 shows that for the fifteen COHA adjectives that have at least doubled in frequency and which have a combined token frequency of 800 to 1,600 in COHA in the 1960s and 1990s, all of these occur at least once in Brown/Frown, which is encouraging. One word occurs between one and nine times in Brown/Frown, and the other fourteen occur at least ten times (e.g., three tokens in Brown and seven tokens in Frown), which is perhaps enough to show an increase from the 1960s to the 1990s. Of these fourteen adjectives that occur at least ten times, six do show frequency that has doubled from the 1960s to the 1990s (e.g., Brown,

An empirical study of changes in lexical frequency

135

six; Frown, twelve, which is shown as “Support” (COHA) in Table 1 above). Another five adjectives show an increase but less than the doubling in COHA (e.g., Brown, six, and Frown, seven; shown as “? ? ?” above). And in three cases, the Brown/Frown data actually shows a decrease from the 1960s to the 1990s (e.g., Brown, seven, Frown, four; shown as “Contradict” above). Overall, then, six of the fifteen types (40%) of these high-frequency adjectives in Brown/Frown show the same doubling in frequency that is shown in the robust data (800– 1,600 tokens) in COHA. The situation is a bit less encouraging for the 127 medium-frequency adjectives (token count of 200–400 for the 1960s–1990s in COHA). Of these, eight do not occur at all in Brown/Frown, and forty-six occur just one to nine times, which is probably too few to see an increase. Of those occurring ten times or more in Brown/Frown, fifty show a doubling, fifteen show a smaller increase, and eight show a decrease. The situation with lower-frequency words is very poor. Remember, these are adjectives like first-time, blurry, impaired, viral, obnoxious, luscious, and motivated – less common to be sure but certainly still the type of adjectives that most speakers of English would be familiar with. Of the 394 types in COHA with a frequency of between fifty and 100 and which have at least doubled in frequency, 114 of these do not occur at all in Brown/Frown, and another 264 occur less than ten times – probably too few to be useful. As a result, Brown/ Frown provides evidence for doubling in frequency for only about 3% of all of these lower-frequency adjectives from COHA. We should also realize that for some types of searches, the situation is much worse than the searches just described, where we are simply looking at the frequency of a given word or phrase over time. For example, Figure 2 shows the collocates (nearby words) of gay in the 1800s and 1900s, and we can use these collocates to find evidence for semantic change during these two periods. For example, the older meaning of “happy, cheerful” is seen in lines 1 and 2, while the newer meaning related to sexual orientation is found in lines 4 and 6.

Figure 2: COHA: Collocates of gay, by decade

136

Mark Davies and Don Chapman

With COHA, it is also possible to compare the collocates in different periods. For example, Figure 3 shows the collocates of gay in the 1830s–1910s (left) compared to the 1970s–2000s (right):

Figure 3: COHA: ADJ/NOUN collocates near the noun gay

Another example of comparing collocates is Figure 4, which shows the adjectival collocates preceding women in the 1830s–1890s (left) and the 1960s–2000s (right) and how women are represented and portrayed in the two periods:

Figure 4: COHA: Adjectival collocates of women

The important issue for our purposes here is the fact that collocates are extremely sensitive to corpus size. In a one-million-word corpus (1/400th the size of COHA), virtually none of the collocates of gay or woman would occur more than one or two times. In summary, we argue that a corpus of one million words – while perhaps useful for high-frequency grammatical changes – is simply too small to examine lexical changes with the vast majority of the words in the language. In addition to size, one other problem with some of these small corpora is the issue of granularity. For example, the Brown family of corpora have texts from 1961 and 1991 (and work is proceeding on a similar corpus from 1931 and then 1901). But because there are texts from only every thirty years, any changes that take place in between these years are essentially “invisible”, and in terms of lexical change, this is often too long of a gap.

An empirical study of changes in lexical frequency

137

Let us briefly consider two examples related to granularity, which are representative of tens of thousands of words. First, let us consider the frequency for groovy in COHA (Figure 5):

Figure 5: COHA: Groovy

Imagine that our two corpora contained texts thirty years apart – from 1955 and 1985. In this case, it would appear (based on the COHA data from the 1950s and the 1980s) that groovy is on the increase. While it has increased slightly in these 30 years, we would miss entirely the steep decrease from the 1960s/1970s to the 1980s. Second, consider the case of normalcy (Figure 6):

Figure 6: COHA: Normalcy

This word was famously “rescued” from obscurity by President Warren G. Harding in 1920, who (according to purists) mistakenly used it instead of the more “correct” normality. The word caught on with a public tired of World War I and other foreign involvements, and Harding went on to win the election. But imagine that we had only two corpora from 1901 and 1931 (as with the planned extensions in the Brown family of corpora). There would obviously be a large increase in frequency between 1901 and 1931, but there would be no way to know if that predated Harding, whether his campaign caused the increase in usage, or whether it was after his time. Corpora that have texts that are spaced decades apart may be adequate for looking at much more gradual grammatical change, but they are much more problematic when looking at lexical change, which can occur quite suddenly.

138

Mark Davies and Don Chapman

In summary, we would agree with Baker (2011) that small one to four million–word corpora – while useful for high-frequency grammatical constructions – are in most cases inadequate for lexical studies (especially historical lexis), except for perhaps a handful of extremely frequent words.

4 Representativeness We have now seen the crucial importance of size for historical corpora. In this section, we will consider the issue of representativeness. As we mentioned previously, a cardinal belief in corpus linguistics is that a corpus needs to be representative – meaning that the texts are carefully selected to produce a meaningful sample of the entire population. Ideally that population would be all English for a given time period, but in historical corpora, where we are limited to the writing that has survived, the population might more properly be said to be all writing that has survived. And since size is important, the population may need to be qualified once more to be all writing that has survived in sufficient quantities to be useful. A key consideration for achieving representativeness, whatever the population, is widespread sampling among the groups that make up the population. Since the differences between groups will usually be greater than the differences within a group, the more groups that are included, the greater the chance that the sample will not be skewed by the idiosyncrasies of one group. For similar reasons, balance across groups will be important as well: if one group (say newspapers) predominates in a corpus, the sample may be skewed toward the idiosyncrasies of that group, even if other groups are sampled. Size can also be framed as an issue of representativeness, in that the sample needs to be large enough for relevant linguistic features to show up. Biber (1993) calls this “linguistic representativeness”. All three considerations – widespread sampling, balance, and size – were important in the design of COHA. The range of sampling for COHA is seen partially in its four broad register divisions – fiction, nonfiction, magazines, and newspapers – but more emphatically in the wide variety of subtypes of each of these registers (http://corpus.byu.edu/coha/ > 400 MILLION WORDS, 1810-2009). Under fiction, for example, are samples from drama, movie scripts, novels, poetry, and short stories. Under nonfiction are samples from all twentyone divisions of the Library of Congress classifications system. Under magazines are samples from 127 different magazines, representing a range of styles and

An empirical study of changes in lexical frequency

139

subjects, from general middle-brow magazines like Harper’s to more specialized magazines like National Geographic. Under newspapers are newspaper samples from a range of geographic areas and subgenres within the newspapers, such as editorials and letters to the editor. When the subtypes are considered, it is clear that COHA samples from a wide variety of subtypes. The balancing in COHA and the size have already been demonstrated. In contrast, the collection making up the Google Books corpus was assembled without any consideration of representativeness. As Michel et al. (2011) explain, the Google Books creators basically just went into large university libraries and scanned everything they could find. From the point of view of corpus linguistics, this is heresy. We should never expect the data from such a haphazardly created corpus to produce the same quality of data as a well-designed corpus like COHA. But as we will see in this section, that is precisely what happens (at least in terms of lexis), and this should come as quite a surprise to historical and corpus linguists. Let us first examine some single-word and ad-hoc evidence for the similarity of (lexical) data from COHA and Google Books, after which we will carry out a much more systematic comparison. First, consider cases in Figures 7–12, which show the frequency of the words bosom, steamship, and teenager in both COHA and Google Books. (Note that all of the frequency charts for Google Books in this paper actually come from the googlebooks.byu.edu version rather than the standard interface at books.google.com/ngrams, although both versions are based on the same underlying frequency data.)

Figure 7: Frequency of bosom in COHA

Figure 8: Frequency of bosom in Google Books

140

Mark Davies and Don Chapman

Figure 9: Frequency of steamship in COHA

Figure 10: Frequency of steamship in Google Books

In both corpora, the frequency starts decreasing in the 1850s and is almost uniformly consistent since that time.

Figure 11: Frequency of teenager in COHA

Figure 12: Frequency of teenager in Google Books

In both corpora, there are very few tokens before the 1940s, but there has been a consistent increase in frequency – decade by decade – since that time. We should note that the similar frequencies over time are limited not just to single words like bosom, steamship, and teenager. They also extend to phrases, such as “so ADJ as to VERB” (e.g., so good as to tell me) or “have quite VERB-ed” (he had quite forgotten her name), as seen in Figures 13–16.

An empirical study of changes in lexical frequency

141

Figure 13: Frequency of so ADJ as to VERB in COHA

Figure 14: Frequency of so ADJ as to VERB in Google Books

Figure 15: Frequency of have quite VERB-ed in COHA

Figure 16: Frequency of have quite VERB-ed in Google Books

And although the focus of this paper is a comparison of lexical frequency, we should also mention that there is typically very good similarity in terms of syntactic constructions as well. To take just one quick example, consider the frequency of “NEED NEG VERB” over time in Figures 17 and 18 (e.g., you needn’t worry; cf. the alternative you don’t need to worry):

Figure 17: Frequency of NEED NEG VERB in COHA

142

Mark Davies and Don Chapman

Figure 18: Frequency of NEED NEG VERB in Google Books

We see an increase throughout the 1800s, peaking in the late 1800s or the first decade of the 1900s, and then a fairly sustained decrease since that time. All of the preceding, however, is merely anecdotal, and perhaps we are just “cherry-picking” the best examples to show a similarity between COHA and Google Books. What we need is a much more systematic comparison of the two corpora. The following is a description of the process we used to carry out a comparison of lexical frequency for thousands of different lexical items in the two corpora. We first created a list of all nouns, verbs, adjectives, and adverbs that occurred at least fifteen times in COHA, which amounted to 116,630 different word forms – statues, remarked, massive, particularly, and others. We then took a sample of every tenth word (11,663 words total) and found the frequency by decade in both COHA and Google Books. Table 2 provides an example of these frequencies and shows the partial data for the word steamship. For reasons of space on this printed page, we show only the data for the 1870s–1950s, but in the study we looked at the frequency in each of the twenty decades from the 1810s–2000s. This table shows the raw frequency (e.g., COHA #) in both COHA and Google Books, as well as the normalized frequency (per million words) in each corpus (e.g., COHA PM). Table 2: COHA/Google Books correlation 1870s

1880s

1890s

1900s

1910s

1920s

1930s

1940s

1950s

COHA #

29

86

191

236

246

219

221

137

96

COHA PM

1.56

4.23

9.27

10.68

10.84

8.54

8.98

5.63

3.91

GB #

3,421

9,327

13,860

23,645

46,966

38,558

21,807

21,003

18,251

GB PM

1.20

2.12

2.46

3.14

4.66

5.44

3.76

3.41

2.25

For each of the 11,663 words, we then computed the Pearson correlation between the normalized frequency in the twenty decades of COHA and the equivalent twenty decades in Google Books. For example, in the case of steamship, the correlation is 0.89, which is extremely high (and a glance at the frequency charts from COHA and Google Books in Figures 9 and 10 show this as well).

An empirical study of changes in lexical frequency

143

The general rule of thumb with the Pearson correlation coefficient is that anything over about 0.50 is considered to be a moderate to high correlation. Let us take a look at the word shuddering, which has a correlation coefficient of 0.55, in Figures 19 and 20.

Figure 19: Frequency of shuddering in COHA

Figure 20: Frequency of shuddering in Google Books

We notice here that both corpora show a general decrease over time, but in COHA it is a fairly sustained decrease since the early 1800s, whereas in Google Books there is an increase through the mid-1800s. It is not nearly as similar in the two corpora as it is with steamship (Figures 9 and 10), which had a correlation coefficient of 0.89. Often the Pearson correlation coefficient is lower than we would expect, based on a quick examination of the frequency charts in the two corpora. For example, Figures 21 and 22 show the frequency of the word sense over time.

Figure 21: Frequency of sense in COHA

Figure 22: Frequency of sense in Google Books

144

Mark Davies and Don Chapman

Both corpora show that the frequency has been relatively flat for the last 150 years or so. But where there might be a slight increase between two decades in COHA, there is perhaps a slight decrease in Google Books (or vice versa). As a result, the Pearson correlation coefficient for this word is only 0.20. Overall, the correlation between COHA and Google Books was 0.51, which shows a fairly large correlation between the two corpora. But we also wanted to see if there was some relationship between the frequency of a word and the COHA/Google Books correlation. Table 3 shows the correlation for ten different word-frequency bands – the top 10% most frequent of the 11,663 words (words 1–1,166 words; frequency-band 1), the next highest 10% (1,167–2,332; band 2), and so on. Table 3: COHA/Google Books correlation by frequency band Frequency band

Words

Correlation

1 2 3 4 5 6 7 8 9 10

1–1,166 1,167–2,332 2,333–3,498 3,499–4,664 4,665–5,830 5,831–6,996 6,997–8,162 8,163–9,328 9,329–10,494 10,495–11,660

0.620 0.591 0.548 0.530 0.489 0.468 0.436 0.401 0.383 0.379

This same data can also be represented in Figure 23, which shows the average Pearson correlation coefficient (Y axis) for the ten frequency bands (X axis).

Figure 23: COHA/Google Books correlation by frequency band

An empirical study of changes in lexical frequency

145

What this shows, perhaps not surprisingly, is that the most frequent words in COHA correlate the best with the same words in Google Books. On the other hand, words at the bottom of the frequency band (which occurred just 15–20 times overall in COHA), have only a 0.379 correlation with Google Books. This makes sense, because with just 15–20 tokens, the data is too sparse and the frequency data is too “spikey” in COHA. For example, consider Figure 24, which shows the frequency for humpbacks, which occurs only seventeen times in COHA, in contrast to Figure 25. The data (especially in COHA) is just too sparse to get a good correlation between the two corpora.

Figure 24: Frequency of humpbacks in COHA

Figure 25: Frequency of humpbacks in Google Books

Perhaps the best evidence for a strong overall correlation between COHA and Google Books comes from looking at words that have a strong overall trend toward increasing or decreasing use in the language. For example, consider grieved, which is clearly decreasing in frequency over time, in Figure 26:

Figure 26: Frequency of grieved in COHA

The question is how well COHA and Google Books agree on these words that are strongly trending one way or the other. In order to measure this, we again looked at all 11,663 words in the study. We looked for words that changed in the same “direction” in four successive decades, spaced four decades apart. For example, we looked for the frequency of each word in the 1860s, 1900s, 1940s, and 1980s or in the 1880s, 1920s, 1960s, and 2000s. If the frequency was higher in the 1880s than in the 1920s, and more in the 1920s than in the 1960s (and so

146

Mark Davies and Don Chapman

on), then we would say that the word was in overall “decrease”. And of course the overall frequency could be increasing as well. The question, then, is whether words that had an overall increase in COHA had the same overall increase in Google Books, and the answer is that they clearly did. Table 4 shows that 354 words have a strongly increasing frequency in both COHA and Google Books, whereas 399 words have a strongly decreasing frequency in both corpora. There were only five words that were increasing overall in COHA but decreasing in Google Books, and only one that was decreasing in COHA but increasing in Google Books. Clearly, the two corpora are in very good agreement with each other, even though COHA was designed to be representative and Google Books was not. Table 4: COHA/Google Books correlation: Overall increasing and decreasing COHA

Google Books

increase decrease

increase

decrease

354 5

1 399

5 Discussion and conclusion In this study, we have looked empirically at the effect that corpus size and representativeness have on lexical robustness and accuracy (as measured by the frequency of words over time). The first important conclusion of this study is that size really does matter. Previous studies involving COHA (e.g., Davies 2012a, 2012b, 2012c) have looked at syntactic phenomena, and they show that a 400-million-word corpus like COHA allows for research on a wide range of syntactic phenomena that could never be studied effectively with a small 1- to 5- million-word corpus like ARCHER or the Brown family of corpora. No previous study, however, has looked carefully at the effect of corpus size on lexical phenomena, such as the accuracy of word frequency over time. Small corpora have not been used much for lexical studies, and researchers perhaps have intuitively known that small corpora would not be robust enough for lexical studies (especially the frequency of words over time). But ours is the first study to empirically compare lexis in small and large historical corpora. As we have seen in our comparison of COHA and the Brown family of corpora, the Brown family of corpora (four million words) provides poor data on lexical frequency for “medium”-frequency words like emerging, compelling,

An empirical study of changes in lexical frequency

147

indoor, preferred, and unclear. And it provides very poor data for lower-frequency words, such as motivated, first-time, blurry, impaired, viral, obnoxious, and luscious. For these words, the Brown family was able to show evidence for overall increases in the frequency of the word from the 1960s to the 1990s in only about 3% of all cases. In addition, the Brown family of corpora are very poor in terms of “granularity” – the ability to track the frequency of words at, for example, the level of individual years. But as we have seen, lexical frequency can change dramatically in just a few years. Because COHA has at least two million words each year for every year since the 1870s, it can track such changes quite effectively. These findings regarding the importance of corpus size for lexical studies are perhaps not surprising, and they probably confirm what others have suspected but never actually measured carefully. What is probably much more surprising are the findings regarding the importance of representativeness and the “correct” design of a corpus, at least as far as being able to accurately measure lexical change. As we have discussed, a primary tenet of corpus linguistics is that representativeness is absolutely critical in corpus design. Corpus creators need to design a corpus so that it accurately reflects the target “population” of text or speech in the “real world”. It is not enough to have a large corpus if it is not representative, and one of the key factors for making a corpus representative is to select from a wide range of texts and text-types. Proportion and balance from type to type and from decade to decade are also important considerations. Yet Google Books has disregarded these principles of representativeness. The creators of this collection simply scanned everything available in several large university libraries. Unlike COHA, there was no attempt to sample from multiple genres or to balance the selection across groups and decades. And yet Google Books provides data on lexical change (as measured by lexical frequency) that is very similar to that of COHA, which is a well-designed corpus. How can this be? The answer may be simpler than we think. The concept of representativeness says that we should accurately “model” the entire target population of texts in the “real world”. But what if you have, in effect, the entire target population at your disposal, or at least a sufficiently large percentage of it? In this case, modeling is not as important. The variety of text-types will be taken care of by a sample that is large enough to catch that variety. And this is precisely what Google Books has done. Of course, a corpus sampled by genre will be important for investigations into genre differences, but when COHA is used as a single, undifferentiated corpus, it behaves remarkably similarly to a corpus that was never intentionally stratified

148

Mark Davies and Don Chapman

by genre to begin with. While it may still be important to make sure that a corpus representing “general language” is stratified by as many genres as possible, the data seem to indicate that it works just as well to come in and scan everything and not worry about genre. This is, of course, assuming that we have a corpus as massive as Google Books, where we have “the whole”, rather than the case where we are merely sampling “the whole”. This conclusion might suggest that – at least in terms of historical data – corpus linguists ought to quit worrying about designing corpora and just include everything that is available. For periods covered by Google Books, we already have a corpus that performs as well as a carefully designed historical corpus like COHA. But this conclusion is too simplistic. In the first place, the limitations of our interface with Google Books provides some serious obstacles to research. As we have discussed elsewhere (Davies 2014), Google Books is very good at tracking lexical change over time. And with the right architecture and interface (such as googlebooks.byu.edu), it can even be pressed into service to look at many types of syntactic change. But crucially, Google Books is not a real “corpus”, in the sense that it contains sentences and paragraphs. It is composed of just n-grams – one-, two-, three-, four-, and fiveword sequences. There is no context and no ability to search beyond those n-grams. This means that it is very difficult to extract collocates, even with an improved interface like googlebooks.byu.edu. In COHA, on the other hand, it is quite easy to find collocates and even to compare collocates in different historical periods (and this is an important method for examining semantic change; see Davies 2012a, 2012b, as well as Figures 2–4 above). Even more seriously, there is an aspect of Google Books that few researchers seem to be aware of. Only those n-grams that occur forty times or more in the underlying corpus are searchable by end users, whether in the “standard interface” at books.google.com/ngrams or in alternative interfaces like googlebooks. byu.edu. Even in a massive corpus like Google Books, the vast majority of all two-, three-, four-, and five-word strings might occur just ten or twenty or thirty-nine times in the underlying corpus, but all of these would be “invisible” to the end user since they don’t occur at least forty times. In COHA, on the other hand, all of the data is available, even if a string only appears one or two times. This makes COHA much more useful for looking at syntactic change (see Davies 2014). In the second place, assembling a large-scale collection of texts for periods not covered by Google Books may be less practical than creating a smaller, widely sampled corpus. Perhaps university libraries will contain enough different kinds of books that the genre effects will not be large when the sample is large,

An empirical study of changes in lexical frequency

149

but for older texts, what similarly large collections do we have available? Will the Early English Books On-line (EEBO) collection, for example, be sufficiently stratified to ignore genres in selection? Or will it be idiosyncratic enough that ignoring other genres will make a difference? It would be interesting to see. And what about collections for even earlier stages of English? For those stages, where so little writing has survived, will it be possible to compile a corpus large enough that it will unintentionally sweep in enough genres to be representative of all language? As we go further back in time, perhaps the only claims for representativeness we can make is that the corpus is representative of what has survived, and in that case, a corpus like the Dictionary of Old English Corpus (DOE) will essentially be the whole. Finally, we might turn the basic conclusion regarding COHA and Google Books on its head. Some might argue that “smaller” corpora like the 400-millionword COHA corpus are now “obsolete”, with the availability of massive data sets like Google Books n-grams. But we could also argue that – at least in terms of researching lexical frequency – it was perhaps not necessary to spend hundreds of millions of dollars (and perhaps millions of man-hours) to create Google Books, when similar data can be found in COHA – which was created on a much more modest budget and by just one person. The results of these comparisons are intriguing. We know that language varies in noticeable ways from genre to genre, but what happens when we mix all the genres together and look at averages across genres? Does it make a difference whether the sample was carefully stratified by genre or not? This paper has provided evidence that a large collection of books will sweep in enough different kinds of language use that the average does not look very different from a corpus that is designed to sweep in many different types of language use. It would be interesting to see if these same kinds of similarities between corpora show up for other kinds of linguistic structures. Figures 13–18 have shown that several lexico-grammatical constructions behave similarly in COHA and Google Books. Will that be the case with more such lexico-grammatical constructions or with other grammatical constructions? Would it make a difference if a corpus were stratified by even more genres than are present in COHA? For now, this present study has raised questions about the relative importance of widespread sampling and sample size. When it comes to examining lexical frequency over time, corpus size is crucial. Given a large enough corpus (such as Google Books), however, wide-spread sampling may not be as important as we have traditionally thought. Finally, we will eventually want to do more than look at the frequency of words over time, and in this case a real corpus like the Corpus of Historical American English (COHA) is invaluable.

150

Mark Davies and Don Chapman

References Baker, Paul. 2010. Will Ms ever be as frequent as Mr? A corpus-based comparison of gendered terms across four diachronic corpora of British English. Gender and Language 4 (1). 125– 129. Baker, Paul. 2011. Times may change but we’ll always have money: A corpus driven examination of vocabulary change in four diachronic corpora. Journal of English Linguistics 39. 65–88. Baron, Alistair, Paul Rayson & Dawn Archer. 2009. Word frequency and key word statistics in corpus linguistics. Anglistik 20. 41–67. Biber, Douglas. 1990. Methodological issues regarding corpus-based analyses of linguistic variation. Literary and Linguistic Computing 5. 257–269. Biber, Douglas. 1993. Representativeness in corpus design. Literary and Linguistic Computing 8. 243–257. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman grammar of spoken and written English. London: Pearson ESL. Davies, Mark. 2012a. Expanding horizons in historical linguistics with the 400 million word Corpus of Historical American English. Corpora 7. 121–157. Davies, Mark. 2012b. The 400 million word Corpus of Historical American English (1810–2009). In Irén Hegedus & Alexandra Fodor (eds.), English historical linguistics 2010, 217–250. Philadelphia: John Benjamins. Davies, Mark. 2012c. Examining recent changes in English: Some methodological issues. In Terttu Nevalainen & Elizabeth Closs Traugott (eds.), The Oxford handbook of the history of English, 263–287. Oxford: Oxford University Press. Davies, Mark. 2014. Making Google Books n-grams useful for a wide range of research on language change. International Journal of Corpus Linguistics 19 (3). 401–416. Hofland, Knut & Stig Johansson. 1982. Word frequencies in British and American English. Bergen: Norwegian Computing Centre for the Humanitiesm & London: Longman. Leech, Geoffrey. 2006. New resources, or just better old ones? The holy grail of representativeness. In Marianne Hundt, Nadja Nesselhauf & Carolyn Biewer (eds.), Corpus linguistics and the web. Amsterdam: Rodopi. Leech, Geoffrey & Roger Fallon. 1992. Computer corpora – What do they tell us about culture? ICAME Journal 16. 29–50. Michel, Jean-Baptiste, Yuan Kui Shen, Aviva Presser Aiden, Adrian Veres, Matthew K. Gray, The Google Books Team, Joseph P. Pickett, Dale Hoiberg, Dan Clancy, Peter Norvig, Jon Orwant, Steven Pinker, Martin A. Nowak & Erez Lieberman Aiden. 2011. Quantitative analysis of culture using millions of digitized books. Science 331. 176–182. [Published online ahead of print 12/16/2010]. Oakes, Michael & Malcolm Farrow. 2007. Use of the chi-square test to examine vocabulary differences in English-language corpora representing seven different countries. Literary and Linguistic Computing 22 (1). 85–100.

III Particulars of communicative setting

Peter J. Grund

Seeing is believing: Evidentiality and direct visual perception verbs in Early Modern English witness depositions 1 Introduction1 In the late 1560s, Isabell Ferbeke filed a witness deposition with the church court in Durham, England, in support of her daughter Agnes’s complaint of domestic abuse. In the extract from the deposition in (1), Isabell makes clear that her information about the condition of her daughter’s body is based on direct visual experience (‘she has seen it’). At the same time, she can supply no direct support from firsthand observation that her daughter’s husband (Thomas) caused those injuries, as she ‘never did see him (. . .) beat her’. (1)

she saith yt she haith sein hir said dough~ boddy blake & bloo wth stroke but she yis ex~ neu~ dyd se hym ye said tho~ beit hir/ ‘She says that she has seen her said daughter’s body black and blue with strokes, but she, this examinant, never saw him, the said Thomas, beat her.’ (ETED: Durham 1567–1574: F_1ND_Durham_013)

Similar phrasing, which highlights the deponents’ visual experience (or lack thereof), is found throughout Early Modern English depositions, and they presumably played an important part in emphasizing the source of the deponents’ information. However, although such expressions would seem crucial in the context of a legal process where credibility and convincing evidence are key, they are far from universal and the type of expression varies. This article charts how and in which contexts deponents (or the scribes taking down their testimonies) mark information as deriving from direct visual perception. In other words, it deals with an aspect of evidentiality, or the indication of the source of information (see section 2), and aims to contribute to the general study of this phenomenon in early English, an area which has recently 1 I am grateful to James W. Hartman, the editors, and a reviewer for commenting on an earlier draft of the article. Naturally, any remaining errors are my own. All modern translations/ paraphrases of the ETED examples are my own.

154

Peter J. Grund

received increasing attention. I focus on different verbal constructions (such as see and observe), since I have shown in previous studies that marking of visual perception is primarily indicated through verbs in Early Modern English depositions (Grund 2012: 15, 18; 2013: 331). However, in this study I show how verbal constructions relate to other evidential strategies, especially over time, and discuss the alternation of explicit marking with more implicit indications. Although my study thus includes aspects of diachronic pragmatics, my approach is primarily pragmaphilological, focusing on the sociocultural and situational context of the use of evidentials from a historical-synchronic perspective (Taavitsainen and Jucker 2010: 12). I argue that visual perception verbs were used strategically in different contexts for a number of pragmatic reasons, including expression of certainty and emphasis and the projection of credibility. The communicative setting of the depositions and the court process in general are thus crucial for understanding the deponents’ (or perhaps the scribes’) deployment of these evidential strategies. This finding highlights the more general significance of the broader sociocultural context as well as the narrower communicative purpose in studying linguistic variation and change.

2 Background My study concerns the linguistic realization and contextual usage of evidential markers of direct visual perception. The concept of evidentiality has been treated in various ways in previous research, which has led to conceptual and terminological confusion (e.g., Bednarek 2006: 636–637). I follow studies such as Aikhenvald (2004) and Whitt (2010, 2011) in defining evidentiality as an expression of the source of information. Many studies of English see evidentiality as also encompassing the concept of epistemic modality or the evaluation of the possibility, certainty, or reliability of the information (e.g., Chafe 1986; Biber and Finegan 1988, 1989; Stygall 1994; Taavitsainen 2001). However, there is good reason to separate the two concepts since expressions of source do not always have clear epistemic connections, and some epistemic markers (most notably modal auxiliaries such as may and could) carry no direct indication of the source of the information. That does not mean that epistemic modality is not important for understanding the dynamics of evidentiality, as will be clearly demonstrated in this study (section 4.3), but the connection is secondary (see Grund 2012: 9–10).

Seeing is believing

155

While many languages have obligatory morphological marking of evidentiality (e.g., Aikhenvald and Dixon 2003), English relies on lexical resources, primarily verbs, and the marking is optional (cf. Palmer 2001: 51). As these lexical resources do not always have evidential functions, it is crucial to distinguish the different usages. As underscored by several scholars (e.g., Anderson 1986: 274; Boye 2010; Whitt 2010: 26, 2011: 349–350), fundamental to the expression of the source of information is that an evidential expression has to have scope over a proposition, or “a meaning unit which can be said to have truth value” (Boye 2010: 290).2 Whitt (2010: 26), for instance, makes a distinction between (2) and (3). (2)

I see the house. (Whitt 2010: 26)

(3)

I see the house burning. (Whitt 2010: 26)

According to Whitt (2010: 26), while (2) simply records the act of perception, (3) records two propositions: the burning of the house and the perception of the burning by the subject I. I see in (3) is thus an evidential construction, while I see in (2) is not. This distinction is in theory straightforward, but context is also crucial here, as illustrated in (4).3 (4)

in y e time the said Anne Saundes and Thomas Greenaway did lodge in this depts house as aforesd she this dept did see them them in naked bed togither dius times ‘At the time that the said Anne Saunders and Thomas Greenway lodged in this deponent’s house as aforesaid, she, this deponent, saw them naked in bed together several times.’ (ETED: London 1627–1628: F_2LD_London_007)

2 Boye (2010) attempts to make a distinction between “states of affairs” (as in “I heard him yell”) and “propositions” (as in “I heard that he was yelling”), arguing that heard is only an evidential marker in the latter case. There is not space here to discuss this distinction fully, but its relevance is not clear, and it does not appear to be supported by examples in context (rather than decontextualized examples as used by Boye) (see Whitt, forthcoming). 3 While Whitt (2010; 2011) appears to exclude all examples such as (2), some of his later research (Whitt 2014: 48n3) acknowledges that in some contexts noun phrase objects can be seen as implying a proposition and hence that such constructions can be seen as evidential (see Grund 2012: 18–19).

156

Peter J. Grund

In the example, the deponent establishes that the information was based on direct visual experience. In this case, the information is not a fully expressed proposition from a syntactic point of view (it is a noun phrase and a prepositional phrase), but the underlying proposition is, of course, “they were naked in bed together” (see also Boye 2010: 301). Instead of simply stating so and leaving the source to be inferred, the deponent underscores that she witnessed it firsthand. However a proposition is construed linguistically, and the notion of two propositions is key for my study, where one proposition, or the “information,” is qualified by another statement that specifies the grounds or source of the information.

3 Material and method My material derives from An Electronic Text Edition of Depositions 1560–1760 (ETED), which contains 905 depositions totaling about 267,000 words transcribed from original manuscript testimonies given by men and women, recorded in various localities around England (see Kytö, Grund, and Walker 2011). These testimonies are statements of a person’s experience in a particular context taken down in writing by a scribe in connection with a court case. These cases range from issues of church taxes, broken marriage promises, adultery, and defamation to theft, battery, murder, and treason. This kind of material comes with a set of challenges in terms of interpreting linguistic data. One challenge in particular is worth highlighting, and that is the role of the scribe. The scribe was rarely a completely passive or absolutely accurate mediator of the words spoken by the deponents; rather, the scribe’s influence can be felt on many levels of language, not least in the reformulation of the deponent’s first-person narration (I saw. . .) into a third-person account (He/she/this deponent saw. . .) (see Grund and Walker 2011; Grund 2011). Considering the potential importance of markers of direct visual perception, it is unlikely that the scribes would insert or leave them out. However, whether they, for example, recorded the actual verb used or provided a near-synonym or an alternative formulation is less clear. As I will show, there are contexts where scribal influence seems particularly likely (section 4.2), while other contexts point to strategic use by the deponents (section 4.3). There are, of course, also ambiguous contexts, though the purpose of the usage seems clear.

Seeing is believing

157

To ensure a comprehensive account of direct visual perception verbs that signal evidentiality in ETED, I approached the topic from two angles. Using the built-in software, I searched for visual perception verbs listed in Visser (1973: 2250–2255, 2340–2346, 2379–2382), complemented by verbs provided in other studies of visual perception verbs in modern English (such as Quirk et al. 1985: 203–204, 1205–1207; Biber et al. 1999: 740–742; Gisborne 2010: 4–13). Secondly, I examined a word list of the full edition to identify verbs that could potentially signal visual perception, checking the ETED context of potential verbs.4 After manual inspection of the instances, a number of (categories of) verbs were excluded. To make clear my methodological decisions, I provide a detailed discussion of these exclusions. Some verbs did not signal visual perception in ETED, such as conceive and apprehend, which can indicate vision in the early modern period (although very rarely), according to the OED (s.vv. conceive, apprehend). Other verbs that did indicate visual perception were nevertheless excluded since they were not found in evidential contexts in ETED (that is, with two propositions involved, as outlined in section 2). This is the case of watch, illustrated in (5). Here, the deponents simply record the act of watching and do not signal information provided as based on the watching. (5)

And the Said John Neale & Soldier went away together & these Informts followed them & in a little time Saw them abt the Middle of a field toward Norwich Arm in Arm & they watcht them which they took notice of ‘And the said John Neal and a soldier went away together, and these informants followed them, and in a little while saw them about the middle of a field toward Norwich arm in arm, and they watched them, which they took notice of.’ (ETED: 1700–1754: F_4EC_Norwich_022)

Verbs such as exhibit and show potentially signal vision, but, as Gisborne (2010: 13) notes, “show[ing] does not entail see[ing]”. From the context of some 4 The following verbs (in various inflected forms and spellings) were considered as potentially signaling perception (although not all were attested): admire, appear, apprehend, attend, behold, catch, conceive, consider, descry, detect, discern, discover, distinguish, espy, eye, exhibit, experience, expose, find, glimpse, look, mark, note, notice, notify, observe, perceive, remark, scan, see, show, sight, spot, spy, survey, view, watch.

158

Peter J. Grund

of the examples, it is reasonably clear that visual perception is involved, but the dynamics of show and exhibit are different in that the subject is not the experiencer (as in the case of a prototypical visual perception verb such as see). Instances of these indirectly visual verbs occur in evidential contexts but mostly carry meanings in addition to visual perception, such as inference or other cognitive processes, as in (6). In the example, it is not clear exactly what the “Examinant” bases her statement about Miss Blandy’s state of mind on. It does not appear to be visual perception alone, as judgments about “Concern or uneasiness” would seem to call for inference or conclusions. These uses are thus similar to instances of appear, look, and seem, which involve (mostly visually based) inference, as in (7) and (8): appearing to be uneasy and seeming to be angry are not matters of direct visual perception, but require some conclusion on the part of the deponents. (6)

This Examinant further Saith That Miss Blandy on her Fathers Death Shewed very little Concern or uneasiness ‘This examinant further says that Miss Blandy, on her father’s death, showed very little concern or uneasiness.’ (ETED: Henley 1751: F_4SC_Henley_005)

(7)

Says that the said Jackson in little while came back out of the said lower room, went upon the said stage, seemed appeared to be very uneasy ‘says that the said Jackson came back out of the said lower room after a little while, went up on the said stage, [and] appeared to be very uneasy.’ (ETED: Lancaster 1700–1760: F_4NC_Lancaster_011)

(8)

wherevpon this Informt came away from the said Ely Harsnet he seeming to be angry, ‘this informant then went away from the said Ely Harsnet, who seemed to be angry.’ (ETED: Somerset 1682–1688: F_3WC_Somerset_048)

The general connection between visual perception and cognition, and the metaphorical extension of visual perception verbs to signal inference or other cognitive processes is well known (e.g., Sweetser 1990: 23–48; see also Alm-Arvius 1993; Gisborne and Holmes 2007). In addition to the verbs cited above, observe, perceive, and see are also used to signal mental activity rather than or in

Seeing is believing

159

addition to visual perception, although examples of straightforward visual perception are more common for these three verbs. Since my study focuses on expressions of direct visual perception and their functions, I exclude all instances (and verbs) that are used to signal perception more indirectly (such as show and exhibit, which require a separate study) and the verbs used in other senses (inferential, cognitive, etc.), including ambiguous examples.5 Finally, perhaps the trickiest case of classification is the verb find. In the first entry to the verb, the OED (in its unrevised OED2 form) notes the following meaning: “To come across, fall in with, meet with, light upon. Primarily of persons, and implying perception of the object encountered”. What is particularly important here is the notion of “implied perception”. Examples of this kind of meaning (or the closely related “discover”) occur frequently in ETED, as in (9). (9)

And when he came thither he founde the saide foster & Collin drinking together ‘And when he came there, he found the said Foster and Collin drinking together.’ (ETED: Winchester 1566–1577: F_1SD_Winchester_015)

The usage in (9) can arguably be construed as evidential: instead of simply stating that Foster and Collin were at the house drinking (a proposition), the deponent makes clear that he has firsthand experience of them being there. He thus avoids the question of how he knew that they were there, although, of course, even without found, readers of the deposition might have assumed that he saw them because the deponent was there (see section 4.3). But as noted by the OED, perception is implied rather than directly stated, unlike verbs such as see or observe, where direct visual perception is the primary meaning. To some extent, then, find resembles show and exhibit in indirectly or secondarily suggesting vision, although in the case of find the subject is the experiencer. Since the meaning is slightly different, I will not consider find here, but its possible evidential use deserves a special study.

5 I realize, of course, that there is a slippery slope here, and that direct perception of a physical object also involves mental activity and processing, though quite different from that involved in examples (6)–(8).

160

Peter J. Grund

4 Findings 4.1 Overall results, diachronic development, and court type variation After these delimitations of the data, the instances and verbs given in Table 1 remained.6 Table 1: Verbs in evidential visual perception contexts in ETED Verb

N

See Observe Perceive Mark Total

200 (90%) 12 (5%) 8 (4%) 1 (