The Structure of Discourse-Pragmatic Variation

Studies in Language Variation

The series aims to include empirical studies of linguistic variation as well as its description,
explanation and interpretation in structural, social and cognitive terms. The series will
cover any relevant subdiscipline: sociolinguistics, contact linguistics, dialectology, historical
linguistics, anthropology/anthropological linguistics. The emphasis will be on linguistic
aspects and on the interaction between linguistic and extralinguistic aspects — not on
extralinguistic aspects (including language ideology, policy etc.) as such.
For an overview of all books published in this series, please see http://benjamins.com/catalog/silv

Editors
Peter Auer Frans Hinskens Paul Kerswill
Universität Freiburg Meertens Instituut & University of York
Vrije Universiteit,
Amsterdam
Editorial Board
Jannis K. Androutsopoulos Peter Gilles K. K. Luke
University of Hamburg University of Luxembourg The University of Hong Kong
Arto Anttila Barbara Horvath Rajend Mesthrie
Stanford University University of Sydney University of Cape Town
Gaetano Berruto Brian Joseph Pieter Muysken
L’Università di Torino The Ohio State University Radboud University Nijmegen
Paul Boersma Johannes Kabatek Marc van Oostendorp
University of Amsterdam Eberhard Karls Universität Meertens Institute & Leiden
Tübingen University
Jenny Cheshire
University of London Juhani Klemola Sali Tagliamonte
University of Tampere University of Toronto
Gerard Docherty
Newcastle University Miklós Kontra Johan Taeldeman
University of Szeged University of Gent
Penny Eckert
Stanford University Bernard Laks Øystein Vangsnes
CNRS-Université Paris X University of Tromsø
William Foley
Nanterre
University of Sydney Juan Villena Ponsoda
Maria-Rosa Lloret Universidad de Málaga
Universitat de Barcelona

Volume 13
The Structure of Discourse-Pragmatic Variation
by Heike Pichler
The Structure of
Discourse-Pragmatic Variation

Heike Pichler
Newcastle University

John Benjamins Publishing Company

Amsterdam / Philadelphia
 TM
The paper used in this publication meets the minimum requirements of
8

the American National Standard for Information Sciences – Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Pichler, Heike.
The structure of discourse-pragmatic variation / Heike Pichler.
p. cm. (Studies in Language Variation, issn 1872-9592 ; v. 13)
Includes bibliographical references and index.
1. Pragmatics. 2. Discourse markers. 3. Linguistic change. I. Title.
P99.4.P72P434   2013
401’.41--dc23 2012048981
isbn 978 90 272 3493 3 (Hb ; alk. paper)
isbn 978 90 272 7218 8 (Eb)

© 2013 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
 Table of contents

List of figures xi
List of tables xiii
List of abbreviations and typographical conventions xv
Key to transcription conventions xvii
Glossary of dialect words xix
Acknowledgements xxi

Part I
chapter 1
Introduction 3
1.1 Variationist sociolinguistics and discourse-pragmatic features  3
1.2 Discourse-pragmatic features: Definition of scope and terminology  4
1.3 Reasons for the neglect of discourse-pragmatic features
in variationist research  6
1.4 Arguments in favour of the quantitative analysis
of discourse-pragmatic features  10
1.5 Aims and focus of the book  17
1.6 Organisation of the book  18
chapter 2
Data, methodology and theoretical framework 21
2.1 Introduction   21
2.2 Data   22
2.2.1 Corpus   22
2.2.2 Speaker sample   25
2.2.3 Summary   27
2.3 Variationist sociolinguistics   27
2.3.1 Premises of variationist sociolinguistics   27
2.3.2 Discourse-pragmatic variables   28
2.3.3 Quantitative methods   32
2.3.4 Summary   35
vi The Structure of Discourse-Pragmatic Variation

2.4 Grammaticalisation   35
2.4.1 Mechanisms of change in grammaticalisation   35
2.4.2 Grammaticalisation and discourse-pragmatic features   38
2.4.3 Grammaticalisation in synchronic data  41
2.4.4 Summary   41
2.5 Conversation analysis (CA)   41
2.5.1 Choice of CA   42
2.5.2 Key concepts of CA   42
2.5.3 Functional domains   45
2.5.4 Quantifying multifunctionality   47
2.5.5 Validating qualitative analyses   48
2.5.6 Summary   48
2.6 Conclusion   49
chapter 3
The BwE verb negation system 51
3.1 Introduction   51
3.2 The evolution of verb negation and negative particles   52
3.3 Negative auxiliaries in BwE   53
3.3.1 Negative auxiliary variants in BwE   53
3.3.2 Geographical distribution and categorisation of negative particle/
negative auxiliary variants   55
3.4 Previous research on negator and negative auxiliary variation   56
3.4.1 Extra-linguistic distribution of variants  56
3.4.2 Intra-linguistic distribution of variants  57
3.5 Quantitative analysis of negative particle and negative 
auxiliary variation   59
3.5.1 Distributional analysis  59
3.5.2 Multivariate analysis  63
3.5.3 Summary of results   65
3.6 Conclusion  65
 Table of contents vii

Part II
chapter 4
The construction i don’t know  69
4.1 Introduction   69
4.2 Previous research on i don’t know  71
4.2.1 Qualitative studies: Discourse-pragmatic functions
of i don’t know   71
4.2.2 Quantitative studies: Variation and change in the use
of i don’t know  73
4.2.3 Aims and contribution of this chapter   76
4.3 The variable context and data coding   77
4.3.1 The variable and the envelope of variation   77
4.3.2 The dependent variable: Variants of i don’t know in BwE   78
4.3.3 Independent variables: Data coding   80
4.4 Qualitative analysis of i don’t know   81
4.4.1 Functions performed by unbound i don’t know   82
4.4.2 Functions performed by bound i don’t know with phrasal
or clausal complements   93
4.4.3 Functions performed by i don’t know
with dependent wh-words   97
4.4.4 Summary of functions   98
4.5 Quantitative analysis of i don’t know  100
4.5.1 Distributional analysis   100
4.5.2 Multivariate analysis   108
4.5.3 Summary of results   115
4.6 Discussion   116
4.7 Conclusion   122
chapter 5
The construction i don’t think  123
5.1 Introduction   123
5.2 Previous research on i don’t think  125
5.2.1 Qualitative studies: Discourse-pragmatic functions
of i don’t think  125
5.2.2 Quantitative studies: Variation and change in the use
of i don’t think  128
5.2.3 Aims and contribution of this chapter   130
viii The Structure of Discourse-Pragmatic Variation

5.3 The variable context and data coding   131

5.3.1 The variable and the envelope of variation   131
5.3.2 The dependent variable: Variants of i don’t think in BwE   134
5.3.3 Independent variables: Data coding   135
5.4 Qualitative analysis of i don’t think   136
5.4.1 Functions performed by i don’t think   137
5.4.2 Summary of functions   148
5.5 Quantitative analysis of i don’t think  150
5.5.1 Distributional analysis  150
5.5.2 Multivariate analysis   156
5.5.3 Summary of results   162
5.6 Discussion  162
5.7 Conclusion   168
chapter 6
Negative polarity question tags (neg-tags) 169
6.1 Introduction   169
6.2 Previous research on question tags   171
6.2.1 Qualitative studies: Discourse-pragmatic functions
of question tags  172
6.2.2 Quantitative studies: Variation and change in the use
of question tags  174
6.2.3 Aims and contribution of this chapter   178
6.3 The variable context and data coding   179
6.3.1 The variable and the envelope of variation   179
6.3.2 The dependent variable: neg-tag variants in BwE  182
6.3.3 Independent variables: Data coding   185
6.4 Qualitative analysis of neg-tags   186
6.4.1 Functions performed by neg-tags   187
6.4.2 Summary of functions   192
6.5 Quantitative analysis of neg-tags  193
6.5.1 Distributional analysis   194
6.5.2 Multivariate analysis   202
6.5.3 Summary of results   205
6.6 Discussion   206
6.7 Conclusion   212
 Table of contents ix

Part III
chapter 7
Discussion & conclusion  215
7.1 Introduction   215
7.2 Synthesis of the results   216
7.3 Implications of the results   220
7.4 Conclusion   224
chapter 8
Challenges for the future 227
8.1 The state of the art and beyond   227
8.2 Dealing with low token frequency   228
8.3 Developing reliable analytical methods   230
8.4 Exploring intra-linguistic constraints
on discourse-pragmatic variation   231
8.5 Exploring patterns of discourse-pragmatic change   232
8.6 Investigating the range of discourse-pragmatic variables   234
8.7 Conclusion   236

References 237
Appendices 259
1. Inventory of functions of unbound i don’t know
in the BwE corpus  261
2. Inventory of functions of bound i don’t know
in the BwE corpus  264
3. Inventory of functions of i don’t know + wh-word
in the BwE corpus  265
4. Inventory of functions of i don’t think
in the BwE corpus  266
5. Inventory of functions of neg-tags
in the BwE corpus  269

Index 271
 List of figures

Figure 2.1  Location of Berwick upon Tweed, Northumberland 23

Figure 3.1  Overall distribution of negator and localised negative auxiliary variants
in productive constructions 59

Figure 4.1  Overall distribution of the variants of i don’t know 101

Figure 4.2  Distribution of variants of i don’t know across speaker sex and age 102

Figure 4.3  Distribution of variants of i don’t know across syntactic contexts

and functional categories 104

Figure 4.4  Normalised frequencies of referential uses of i don’t know

across individuals  107

Figure 4.5  Normalised frequencies of pragmatic uses of i don’t know

across individuals 107

Figure 5.1  Overall distribution of the variants of i don’t think 151

Figure 5.2  Distribution of variants of i don’t think across speaker sex and age 152

Figure 5.3  Distribution of variants of i don’t think across functional domains

and complement types 154

Figure 5.4  Normalised frequencies of i don’t think across individuals 155

Figure 5.5  Distribution of variants of i don’t think across micro-functions 160

Figure 6.1  Distribution of neg-tag variants across speaker sex and age 197

Figure 6.2  Distribution of neg-tag variants across syntactic-semantic contexts 199

Figure 6.3  Distribution of neg-tag variants across functional categories 201

Figure 6.4  Normalised frequencies of neg-tags across individuals  202

 List of tables

Table 2.1  Speaker sample 25

Table 3.1  Inventory of negative particle and negative auxiliary variants in BwE 56

Table 3.2  Distribution of negator variants across auxiliaries and distribution

of localised negative auxiliary variants  61

Table 3.3  Distribution of negator and localised negative auxiliary variants

across clause types  62

Table 3.4  Contribution of external and internal factors to the probability

of -nae, no and divn’t  64

Table 4.1  Inventory of variants of the construction i don’t know in BwE  80

Table 4.2  Coding schema for i don’t know functions 99

Table 4.3  Distribution of i don’t know across syntactic configurations  103

Table 4.4  Frequency in per cent of unbound i don’t know across functional
categories by age and sex  105

Table 4.5  Frequency in per cent of bound i don’t know across functional
categories by age and sex 105

Table 4.6  Contribution of external and internal factors to the probability

of I don’t know  109

Table 4.7  Contribution of external and internal factors to the probability of I dunno  111

Table 4.8  Contribution of external and internal factors to the probability of I dono 113

Table 4.9  Contribution of external and internal factors to the probability

of I divn’t knaa  114

Table 5.1  Inventory of variants of the construction i don’t think in BwE 135

Table 5.2  Coding schema for i don’t think functions 149

xiv The Structure of Discourse-Pragmatic Variation

Table 5.3  Contribution of external and internal factors to the probability

of I don’t think, I doØ think and I divn’t think  157

Table 6.1  Inventory of neg-tag variants in BwE  185

Table 6.2  Coding schema for neg-tag functions 193

Table 6.3  Overall distribution of neg-tag variants  195

Table 6.4  Inventory and frequency of auxiliary-negative-pronoun combinations

in non-localised canonical tags  198

Table 6.5  Contribution of external and internal factors to the probability of innit
and non-localised canonical tags  204
 List of abbreviations
and typographical conventions

BNC British National Corpus

BwE Berwick English
CA conversation analysis
COLT Corpus of London Teenage Speech
CPP Cognitive Prominence Principle
CTRP complex Transition Relevance Place
LIC Linguistics Innovator Corpus
neg-tags negative polarity question tags
OED Oxford English Dictionary
p proposition
SAP Social Agreement Principle
TAM tense-aspect-modality
TCU turn-constructional unit
TQ tag question
* denotes grammatically incorrect sentences
small capitals denote the discourse-pragmatic variables analysed in Part II
the primary verbs be, have and do as lexical items
lower case italics denote the formal variants of the discourse-pragmatic variables
analysed in Part II
discourse-pragmatic variables other than the ones
analysed in Part II
grammatical forms of the primary verbs be, have and do
 Key to transcription conventions

[ ], [[ ]] overlap
== latching
= turn continuation
- false start, truncation
“ ” quoted speech
(h), (.h) inbreath, outbreath
@ laughter
+ tisking
% sniffle
$ swallowing
<@ @> produced with laughter
<£ £> produced with smiley voice or suppressed laughter
< > increased tempo
> < reduced tempo
CAPITALS louder than surrounding talk
underlining emphatic stress
º º soft speech
superscript font higher than usual pitch
subscript font lower than usual pitch
:, :: syllable lengthening
(.), (..), (…) short, medium, long pause
. final intonation contour
, continuing intonation contour
? rising intonation contour
(text) uncertain transcription
(?) undecipherable words
((text)) extra-linguistic information
italics used in examples to highlight the variable/variant discussed
in the text
 Glossary of dialect words

aggro bother, inconvenience

auld old
aye yes
bog toilet
cannae can’t
cannit can’t
Cockney person from/dialect of London
dear expensive
dinnae don’t
divn’t don’t
doylem idiot
fither father
frae from
gie give
Geordie person from/dialect of Newcastle/Tyneside
hae have
hantle head
heed head
jougle dog
kecks trousers, underpants
ken know
knaa know
knacker strenuous work
mind remember
mingin dirty, gross
-nae -n’t
nay no
no not
nowt nothing
pet love (endearment form)
skint short of money
twang language, dialect
wa our
whae who
wi with
winnae won’t
youse you
 Acknowledgements

Writing this book has been both challenging and rewarding, and I count myself
lucky to have learned from ongoing discussions with some of the very best: Jenny
Cheshire, Alex D’Arcy, Stephen Levey, Sali Tagliamonte and Elizabeth Traugott.
Many other fellow linguists have provided invaluable support and advice at vari-
ous stages in the writing process: Alex Bellem, Oliver Bond, Lynn Clark, Karen
Corrigan, Liesbeth Degand, Damien Hall, Daniel Ezra Johnson, Yuni Kim, Nick
Roberts, Jane Setter, Dominic Watt and many others who, though not mentioned,
have not been forgotten. Gisle Andersen, Elizabeth Traugott, Suzanne Evans
Wagner and two anonymous reviewers have provided detailed and helpful com-
ments on parts or the whole of earlier drafts of this book for which I am grateful.
Of course, any remaining shortcomings of the book remain my own responsibil-
ity. Anne Sherwin came to the rescue when my stress levels failed to cope with
EndNote, and Jeff Wilson helped with technical issues at the very end. I thank Paul
Kerswill for inviting me to submit the manuscript to SiLV. At John Benjamins, the
series editor Peter Auer, the acquisition editor Anke de Looper and the produc-
tion coordinator Patricia Leplae have patiently provided guidance throughout the
writing and production process.
To my family in Austria I must apologise for the numerous occasions where
I was too busy to return their calls or thank them for their unconditional supply
of chocolates. Thank you to Sue Fox, Erika Fulop, Iris Gruber-La Sala, Martha
Marizzi, Martina Pfeiler, Annu Rist and Fran Sanchez for keeping me on-line
company during insane hours and for providing encouragement when I needed
it. Cheers to Olly Bond and Yuni Kim for entertainment and laughter. Andy
Gordon has been my rock of love and support through all of this. Thank you for
everything.
And, finally, thank you to the people of Berwick for supplying the discourse-
pragmatic variables celebrated in this book.
part i
 chapter 1

Introduction

1.1 Variationist sociolinguistics and discourse-pragmatic features

Variationist sociolinguistics emphasises the importance of analysing language in

its social context and treats language use as the central object of study. The sys-
tematic focus on analysing actual language use has produced an ever-growing
body of empirical evidence to support three principal tenets: (1) that language
use is highly variable, and that this variability evinces “structured heterogeneity”
(Weinreich et al. 1968: 99–100) along multiple extra- and intra-linguistic factors;
(2) that language use is in a constant state of flux, and that these fluctuations are
regular, rule-governed and observable in synchronic language variation; (3) that
language use conveys social meanings, and that these social indexicalities are re-
flected in structured patterns of linguistic variation and change. Support for these
tenets has been provided mainly by analyses of phonological and, to a somewhat
lesser degree, morpho-syntactic features. Although the use of discourse-pragmatic
features, i.e., polyfunctional linguistic items and constructions such as oh, just,
so, I mean, mind you, stuff like that, is also variable, changeable and accessible
for creating social indices, the number of studies which have systematically cor-
related their use with contextual predictors is comparatively modest. This has led
Macaulay (2002a: 298) to lament that “the study of discourse variation is still at an
elementary stage” (see also Pichler 2010: 582).
The present investigation supports Macaulay’s (2002a, 2005) call for a fuller
integration of discourse-pragmatic features into the variationist research agenda.
It develops an innovative, multidimensional methodology which accommodates
the full complexity of variation and change in the use of the conventionalised
polyfunctional elements introduced above. It applies this methodology to explor-
ing patterns of variation and change in the use of three selected variables, each of
whose variants are derived from the same source construction but differ in terms of
their morpho-phonological encoding: i don’t know, i don’t think and negative
polarity question tags (henceforth neg-tags). Principled and accountable analyses
of these variables in a corpus of interview data collected in Berwick upon Tweed,
north-east England, reveal important new insights into: the creation and organisa-
tion of discourse-pragmatic variability; the role of discourse-pragmatic variables in
shaping synchronic language variation; and the way in which discourse-pragmatic
4 The Structure of Discourse-Pragmatic Variation

variants acquire their social indexicality. By way of uncovering new findings which
enrich and illuminate current models of language variation and change, this study
demonstrates the importance of discourse variation analysis for testing and devel-
oping current theories of language.
This chapter provides an introduction to the variationist analysis of discourse-
pragmatic features more generally and to the present investigation more specifi-
cally. Section 1.2 outlines how the class of discourse-pragmatic features and the
boundaries of discourse variation analysis have been delimited for the purpose of
this investigation. It demonstrates how the term ‘discourse-pragmatic features’ is
used here in a non-traditional way to refer to the category of items and construc-
tions that elsewhere tend to be subsumed under the labels ‘discourse markers’ or
‘pragmatic particles,’ and provides a rationale for the choice of terminology used.
Section 1.3 explores the theoretical and methodological reasons for the neglect
of discourse-pragmatic features in the variationist literature. The descriptive and
theoretical value of extending the scope of variationist research to encompass
discourse-pragmatic variables is detailed in Section 1.4. Finally, Sections 1.5 and
1.6 provide an overview of the aims and the structure of this book.

1.2 Discourse-pragmatic features: Definition of scope and terminology

As a result of their formal and functional heterogeneity and the fact that they have
been studied within a diversity of analytical and theoretical frameworks (see, for
example, Schiffrin 2001; Schourup 1999 for overviews), there is as yet no general
consensus on such fundamental issues as the terminology applied to the category
of linguistic elements referred to in this book as discourse-pragmatic features, the
inventory of elements to be included in this category, or the linguistic properties
uniting them as a category (see, for example, Brinton 1996: Chapter 2; Fischer 2006;
Jucker & Ziv 1998; Schourup 1999 for overviews of relevant debates). Therefore, it
is necessary to specify at the outset how this study defines the scope of the category
and the terminology used.
Although this approach is not without its sceptics (see, for example, Fischer
2006: 5–7), the present study adopts the widespread practice to delineate the cat-
egory of linguistic elements studied in this book on functional-pragmatic grounds.
In this definition, discourse-pragmatic features constitute a formally heterogene-
ous category of syntactically optional elements which make little or no contribu-
tion to the truth-conditional meaning of their host units and – depending on their
scope, linguistic co-text as well as sequential, situational and cognitive context –
perform one or more of the following macro-functions: to express speaker stance;
to guide utterance interpretation; and to structure discourse. In recognising the
 Chapter 1.  Introduction 5

potential for discourse-pragmatic features to have variable scope and to perform

multiple pragmatic and/or procedural functions, sometimes simultaneously, the
current definition of the category is much broader than the seminal definitions
provided by Blakemore (1987: 125, “expressions used to indicate how the relevance
of one discourse segment is dependent on another”), Fraser (1999: 931, items that
“signal a relationship between the interpretation of the segment they introduce,
S2, and the prior segment, S1”), Östman (1981: 5, items that “‘implicitly anchor’
the utterance in which they function, to the speaker’s attitudes towards aspects of
the on-going interaction”), Schiffrin (1987: 31, “sequentially dependent elements
which bracket units of talk”), or Schourup (1999: 234, items “used to relate [an]
utterance to the immediately preceding utterance”). The breadth of the current
definition and the preference for the label ‘discourse-pragmatic features’ as an
umbrella term for the linguistic elements discussed in this book challenge the
terminological and typological distinction made by Fraser (1990) between ‘dis-
course markers,’ which perform a structural role, and ‘pragmatic particles,’ which
express speaker stance. Fraser’s distinction is untenable in light of the evidence
provided in Chapters 4 to 6 which shows that individual discourse-pragmatic fea-
tures rarely perform only one of these broad functions. Regular use in this book of
the alternative labels ‘discourse-pragmatic variables’ or ‘formulaic constructions’
reflects the methodological and conceptual approach it takes to the analysis of what
are more neutrally called discourse-pragmatic features. The term ‘construction’ is
used in a pre-theoretical way to refer to syntagmatic strings, and does not reflect
a Construction Grammar view of the data (Goldberg 1995; Kay & Fillmore 1999);
‘formulaic’ is used in opposition to ‘productive’ to emphasise the conventionalised
and non-compositional nature of discourse-pragmatic features (Wray 2002).
The current definition of discourse-pragmatic features further diverges from
those presented elsewhere for the category of conventionalised polyfunctional
elements described above in that it does not recognise as a unifying property of
its members formal-syntactic specifications such as shortness, phonetic reduc-
tion, lack of stress and sentence-initial or -external positioning (see, for example,
Brinton 1996: 33–34; Jucker & Ziv 1998: 3; Schiffrin 1987: 328). Strong adherence
to these criteria would entail excluding from the category of discourse-pragmatic
features comment clauses such as I (don’t) know, I (don’t) think, as you know, what’s
more and clause-final tags such as don’t you, isn’t that so, you know what I mean,
or something like that because they do not consistently meet all of the phono-
logical, prosodic and syntactic properties stipulated in the literature cited above.
Considering, however, their strong resemblance to more prototypical discourse-
pragmatic features in terms of their functional versatility and diachronic develop-
ment (see further Brinton 2006, 2008; Hoffmann 2006), the exclusion of comment
clauses and clause-final tags from the category of discourse-pragmatic features
6 The Structure of Discourse-Pragmatic Variation

is theoretically unjustified and counter-productive to developing a comprehen-

sive theory of linguistic elements that function primarily in the non-referential
domains of language use.
The scope of discourse variation analysis in this book is delimited to the quan-
titative analysis of patterns of variation and change in the use of discourse-prag-
matic features as defined above. Unlike other scholars (e.g. Jucker & Taavitsainen
2012; Macaulay 2002a; Schneider & Barron 2008), I do not include in my dis-
cussion and analysis of discourse-pragmatic variation and change the variable
realisation or functionality of: pragmatic units such as speech acts; interactional
strategies such as turn-exchange, topic-selection and back-channelling; or se-
quence structures such as greeting and leave-taking rituals. As shown inter alia
by Barron (2008), Coupland (1983) and Johnstone (1990), these phenomena are
also variable and mutable, and their variationist analysis is very valuable. The
exclusion of these phenomena from the present investigation is dictated purely
by considerations of scope.

1.3 Reasons for the neglect of discourse-pragmatic features

in variationist research

At least two reasons can be identified why to date discourse-pragmatic features

have not figured as prominently on the variationist research agenda as phonologi-
cal or morpho-syntactic features. First, because their status in the linguistic system
is different from that of other linguistic elements, discourse-pragmatic features are
often described in negative terms and marginalised in traditional frameworks of
grammar and linguistic analysis. Second, due to their unique semiotic, syntactic
and functional characteristics, discourse-pragmatic features present a number of
methodological challenges for variationist analysis which are not easily overcome.
This section explores in more detail the theoretical and methodological reasons
for the relative neglect of discourse-pragmatic features in (variationist) linguistic
research, and examines recent shifts in how they are theorised and quantified.
Discourse-pragmatic features derive from diverse syntactic sources (including,
amongst others, subject-verb combinations, adverbials, conjunctions and interjec-
tions), and they do not occupy a fixed syntactic position. Therefore, they cannot
be described as constituting a homogeneous word class whose members share a
set of morpho-syntactic properties. In addition to defying syntactic classification,
discourse-pragmatic features eschew lexical definition. They do not generally com-
municate referential content but function to encode pragmatic and procedural
meanings which are not easy to specify in lexical terms. Their indeterminate gram-
matical status and general lack of truth-conditional meaning have routinely been
 Chapter 1.  Introduction 7

adduced as evidence that discourse-pragmatic features are extra- or a-grammatical

elements of language which are not part of sentence grammar (Goldberg 1980;
Stein 1985: 99) and which can be marginalised as meaningless verbal fillers and
superfluous hesitation markers (Brown 1977; Lakoff 1973). As such, they have
been regularly dismissed as pseudo-linguistic devices not warranting scholarly
investigation.1
Evidence from speech perception experiments (e.g. Parton et al. 2002; Russell
et al. 2008), the media (e.g. Petersen 2004; the quotes in Levey 2003) as well as
websites created by self-proclaimed communication experts (e.g. Berkley 2002)
suggests that although speakers may be aware of the specific pragmatic meanings
associated with individual features (Fox Tree 2007), negative perceptions and pre-
scriptive stigmatisations of discourse-pragmatic features similar to those outlined
above continue to be widespread among non-linguists, even those who make fre-
quent use of such features themselves (Watts 1989). Amongst linguists, by contrast,
a dramatic re-orientation in the evaluation and theorisation of discourse-pragmatic
features has taken place since the 1980s. With the growing interest in “language
above the sentence or above the clause” (Stubbs 1983: 1) and in “the purpose or
functions which [linguistic] forms are designed to serve in human affairs” (Brown
& Yule 1983: 1), scholars have begun to investigate the role of discourse-pragmatic
features in the production, comprehension and interpretation of discourse. The
seminal works on discourse-pragmatic features published in the 1980s (Blakemore
1987; Erman 1987; Östman 1981; Schiffrin 1987; Schourup 1985) as well as the
subsequent proliferation of qualitative studies investigating the distribution and
functionality of individual features have demonstrated that widespread assump-
tions about their extra-grammatical status and superfluous nature are erroneous.
Scholars investigating the syntactic distribution, linguistic properties and
multi-functionality of discourse-pragmatic features have convincingly demon-
strated that they are not overt manifestations of verbal dysfluencies and inarticu-
lateness, and that they indubitably constitute integral and indispensable elements
of the core linguistic system. For example, Levey (2006: 431–432) argues that dis-
course like is not symptomatic of production problems or poor syntactic plan-
ning because its rate of occurrence with aborted utterances or false starts is fairly
low. Traugott (2003a; see also Traugott & Dasher 2002: 158–159) postulates that
discourse-pragmatic features occupy highly constrained sequential and syntactic
positions. Recent variationist studies provide quantitative evidence in support of

1. Not all frameworks relegate discourse-pragmatic features to the periphery. Construction

grammar, for example, does not distinguish between ‘core’ and ‘peripheral’ phenomena of gram-
mar (Gisborne & Trousdale 2008). (See also further below on Kaltenböck et al.’s (2011) thetical
grammar approach to theorising discourse-pragmatic features.)
8 The Structure of Discourse-Pragmatic Variation

this premise, demonstrating that the syntagmatic order of discourse-pragmatic fea-

tures and other clausal constituents is far from random (Andersen 2001: 272–286;
D’Arcy 2007: 480–482; Tagliamonte 2005). Observations such as these show that
“in constituent structure terms [discourse-pragmatic features] are [core elements]
of the structure of a sentence” (Traugott 2003a: 643) which cannot be disregarded
in linguistic analysis on the basis of being marginal or peripheral to sentence gram-
mar. Additional support for viewing discourse-pragmatic features as legitimate
grammatical phenomena is provided by proponents of non-traditional conceptions
of grammar. Kaltenböck et al. (2011) draw on the distinct linguistic properties of
what they call ‘theticals’ (e.g. their prosodic and syntactic independence, positional
mobility, non-restrictive meaning) to argue that they belong to thetical grammar, a
domain of discourse grammar that is structurally independent from but not sub-
ordinate to sentence grammar. Other scholars, including Brinton (2006), Diewald
(2006), Traugott (2003a) and the contributors in Degand & Simon-Vandenbergen
(2011), argue in favour of extending the traditional notion of grammar beyond the
level of the clause so that grammatical function is understood as an open category
which comprises not just syntactic functions such as subject, object, complement,
etc. but also pragmatic and procedural functions such as stance expression and
discourse organisation. On this basis, discourse-pragmatic features are conceptu-
alised as performing “genuine grammatical functions” (Diewald 2006: 424) which
may be optional in terms of constituent structure but which are obligatory in terms
of their pragmatic and interactional contribution to discourse.
Although some linguists continue to cite the syntactic flexibility and optional-
ity as well as the lack of referential meaning of discourse-pragmatic features as evi-
dence for their peripheral status in the linguistic system (see, for example, Eckardt
2012), the general consensus today is that discourse-pragmatic features are core
elements of grammar which constitute bona fide objects of linguistic enquiry. As
implied above, variationist sociolinguistics has the potential to corroborate claims
about the grammatical status of discourse-pragmatic features through provision
of quantitative empirical evidence of their usage and distribution in discourse.
However, attempts to provide such evidence have been complicated by methodo-
logical issues.
The variationist paradigm was originally developed by Labov (1963, 1966)
for the analysis of phonological variation and has been successfully applied to
the analysis of morpho-syntactic variation. Yet because of fundamental differ-
ences in the nature and use of lower- and higher-level linguistic features, scholars
have questioned the feasibility of extending the Labovian paradigm to the analysis
of variation and change in discourse (and syntax) (see inter alia Cheshire 1987;
Lavandera 1978; Romaine 1984; Winford 1984). Discourse-pragmatic features do
not easily satisfy the criteria set out by Labov (1972) for the linguistic variable, the
 Chapter 1.  Introduction 9

principal methodological tool in variationist analysis. Firstly, the operation of se-

mantic-pragmatic and interactional-situational constraints on the use of discourse-
pragmatic features regularly affects the frequency critical for quantitative analysis
(Labov 1972: 8). Secondly, their unique semiotic nature, intrinsic multifunctional-
ity and great context-sensitivity prohibit most discourse-pragmatic variants from
being identified on the basis of semantic equivalence (Labov 1972: 271). Thirdly,
the syntactic mobility of discourse-pragmatic features as well as their multi-faceted
meanings and functional overlap with features from other components of language
cause difficulty in defining the scope of analyses. This hampers attempts to quantify
discourse-pragmatic variation in an accountable manner (Labov 1972: 71–72). As
a result of the aforementioned factors, discourse-pragmatic features have been
shunned in the early decades of variationist research.
In more recent decades, though, some progress has been made in addressing
the methodological challenges outlined above and bringing discourse-pragmatic
features more firmly under the remit of variationist linguistics. Dines (1980)
and Lavandera (1978) proposed to resolve issues in conceptualising discourse-
pragmatic features as linguistic variables by substituting the condition of seman-
tic equivalence with one of functional comparability between variants (see also
Sankoff & Thibault 1981). Others argued that information about the social or
stylistic distribution of discourse-pragmatic features could be garnered from nor-
malised frequency tabulations, usually given as the number of tokens per 1,000 or
10,000 words (Stubbe & Holmes 1995). Holmes (1986, 1987) devised methods for
providing accounts of discourse-pragmatic variation which consider the variable
functionality of discourse-pragmatic features in signalling social group member-
ship. Finally, D’Arcy (2005) circumscribed the variable context of discourse like on
the basis of syntactic criteria in order to yield insights into its linguistic-structural
conditioning. While none of these studies have succeeded in fully resolving the
range of methodological challenges outlined earlier (see further Pichler 2010 and
Chapter 2.3), they nevertheless represent important milestones in the development
of discourse variation analysis.
As a result of the growing recognition that there is no theoretical reason for
exempting discourse-pragmatic features from variationist analysis and that their
quantitative analysis, albeit challenging and complex, is in fact feasible, recent
decades have witnessed a moderate expansion in discourse variation studies. The
next section will outline the value of variationist discourse analysis for different
subfields of linguistics.
10 The Structure of Discourse-Pragmatic Variation

1.4 Arguments in favour of the quantitative analysis

of discourse-pragmatic features

Some qualitative scholars downplay the value of quantitative discourse studies,

suggesting, for example, that a focus on distributional frequencies may yield
skewed descriptions of the functionality of discourse-pragmatic features (Hansen
2006: 21–22) or that quantitative studies are simply futile (see Tottie 1992: 121).
While it is true that quantification entails an inevitable degree of abstraction, ob-
servations such as these underestimate the methodological rigour of quantitative
discourse studies as well as their descriptive and explanatory value within and be-
yond variationist linguistics. Variationist discourse studies are critical to providing
comprehensive and accurate descriptions of social and geographical dialect varia-
tion and to formulating a holistic theory of language variation and change which,
by definition, must be based on systematic analyses of all components of language,
including discourse-pragmatics. Further, they are instrumental in enriching the
empirical basis required to develop accurate descriptions and coherent theories
of the characteristic linguistic properties and typical evolutionary trajectories of
discourse-pragmatic features.
The variationist discourse studies carried out to date consistently highlight the
fact that discourse-pragmatic features, like features at other levels of the linguistic
system, evince orderly heterogeneity and undergo regular rule-governed change.
In terms of variability, they have shown that correlations between linguistic varia-
tion patterns and socio-demographic categories, amply documented for variables
in phonology and morpho-syntax, also emerge at the level of discourse-pragmatics
(Macaulay 2005; Stubbe & Holmes 1995). Despite their comparative infrequency,
discourse-pragmatic variables may even demonstrate a greater amount of socially
stratified variation than phonological variables (Woods 1991). Third-wave varia-
tion studies indicate that discourse-pragmatic variables resemble other variables
in constituting an important means for indexing social identities: variation in their
use and design contributes to the negotiation of group-specific styles (Bakht 2010;
Drager 2011; Moore & Podesva 2009). Additional evidence for the social salience
of discourse-pragmatic features in the minds of speakers and hearers has been
furnished by experimental studies. For example, Andersen et al.’s (1999) research
into first language acquisition reveals that children as young as four years old are
sensitive to the social meanings conveyed by discourse-pragmatic features: pre-
school children engaged in role-play used different sets of discourse-pragmatic fea-
tures depending on whether they were impersonating lower-status or higher-status
characters. Dines’s (1980: 19–20) small-scale experiment in which middle-class
speakers attributed a text to a working-class author on the basis that it contained
general extenders, i.e., phrase- or clause-final constructions such as and that, or
 Chapter 1.  Introduction 11

something (like that), suggests that discourse-pragmatic features are productive

resources for social stereotyping (see also Watts 1989).2
In terms of change, variationist discourse studies have shown that discourse-
pragmatic features are an important source of linguistic innovations (Tagliamonte
2005, 2008). As is the case with innovative variants at other levels of the grammar,
innovative variants in discourse-pragmatics spread by social and geographical dif-
fusion (Buchstaller 2008; Buchstaller & D’Arcy 2009; Cheshire et al. 2005), and ex-
hibit an adolescent peak in apparent-time change in progress, in line with Labov’s
(2001) logistic incrementation model (Tagliamonte & D’Arcy 2009). Moreover, dis-
course variation studies have demonstrated that changes in discourse-pragmatics
affect both the function and the form of variables. Erman’s (2001) study of you
know in age-stratified datasets of London English reveals that young speakers use
you know largely as a metalinguistic device, whereas adults use it largely as a tex-
tual device. Barbieri’s (2008) exploration of language variation across the lifespan
uncovers striking age-based differences in speakers’ choice of discourse-pragmatic
features to signal stance (see also Brinton [2001: 150–151] who argues that prag-
matic functions endure while their form is continuously renewed). Finally, it has
been shown that age-appropriate use of discourse-pragmatic features is part of
speakers’ sociolinguistic competence (Roth-Gordon 2007).
Yet despite the fact that they broadly conform to fundamental tenets of vari-
ationist sociolinguistics, it must not be assumed that discourse-pragmatic variables
consistently exhibit patterns of sociolinguistic variation and change identical to
those widely attested for phonological and lower-level morpho-syntactic variables.
Cheshire et al.’s (2005) investigation of dialect convergence in contemporary British

2. Written social attitude questionnaires, matched-guise tests and other speech evaluation
experiments have revealed that speakers harbour strong covert attitudes to discourse-prag-
matic features. Their use is generally positively evaluated along social attractiveness/solidarity
dimensions and negatively evaluated along status/competence dimensions (Buchstaller 2006b;
Dailey-O’Cain 2000), so much so that frequent users of discourse-pragmatic features may be
disadvantaged in the job hiring process (Parton et al. 2002; Russell et al. 2008). However, because
experimental methods in discourse-pragmatics remain relatively unsophisticated in comparison
to those recently developed in phonology, it is not clear which of the many (variable) properties
of discourse-pragmatic features it is that trigger listeners’ strong reactions to their use and users:
their frequency; their role in constructing interactional styles; the way they seemingly disrupt
the flow of discourse and the syntactic structure of utterances; or their regular occurrence in
prosodically and pragmatically prominent positions. These are serious shortcomings, for, as
pointed out by Kristiansen (2011: 277), our understanding of language variation and change
depends, at least in part, on our understanding of “whether and how different aspects of language
[use] may be differently available to awareness and/or social evaluation.” To fully understand the
social indexicality of discourse-pragmatic variation, it will be necessary to complement survey
studies with well-designed experimental studies.
12 The Structure of Discourse-Pragmatic Variation

English reveals that whereas the social distribution of the tag variant innit parallels
that of non-standard morpho-syntactic variants undergoing convergence, discourse
like does not differentiate socio-economic groups in the same way as phonological
innovations such as th-fronting. Tagliamonte & D’Arcy’s (2009) study of linguis-
tic innovations in Toronto English also uncovers disparities between patterns of
change in phonology and discourse-pragmatics. Although the patterns of be like,
like and so variation are analogous with those documented for phonological vari-
ables in exhibiting a peak in frequency during adolescence (see above), they do not
mirror the patterns reported by Labov (2001) whereby males lag behind females
in apparent-time incremental change in phonology. Instead, all three discourse-
pragmatic variables (as well as a range of morpho-syntactic variables analysed)
exhibit a lack of gender contrast in the apparent-time trajectory of change (see
also D’Arcy 2005; Denis 2011). I will argue below that due to the unique semiotic
properties of discourse-pragmatic features, their specific role in social interaction,
and the multi-layered nature of their diachronic development, the findings sum-
marised above may point towards more widespread differences in the distribution
of discourse-pragmatic variables as opposed to other linguistic variables.
Phonological variables have no intrinsic referential meaning. Consequently,
we can safely assume that their variants are identical in truth value and that pat-
terns of variation in their rates of use reflect social group membership. Lower-level
morpho-syntactic variants are regularly shown to parallel phonological variants
in deriving their social meaning from their quantitative distribution across so-
cial cohorts (see, however, Levinson 1988: 166). Conversely, even though many
discourse variation studies have drawn conclusions about the social indexical-
ity of discourse-pragmatic variables on the basis of quantifying their frequency
across independent social variables (Dailey-O’Cain 2000; Dines 1980; Dubois &
Crouch 1975; Stubbe & Holmes 1995; Tagliamonte 2005), discourse-pragmatic
variables are unlikely to differentiate social groups in purely quantitative terms.
In contrast to phonological variables, discourse-pragmatic variables have by defi-
nition a meaning, for their use is motivated by their functionality. What scholars
who quantify the use of discourse-pragmatic variables without studying them in
their interactional context ignore is the fact that these meanings are multi-layered:
individual variables encode multiple pragmatic meanings, often simultaneously.
This property of discourse-pragmatic variables is crucial for indexing social group
membership. Speakers exploit the multifunctionality of discourse-pragmatic vari-
ables to meet the specific interactional goals associated with particular speaker
roles (Redeker 1990), to display and enhance their distinctiveness from members
of other social groups (Cheshire 1981; Meyerhoff 1994; Moore & Podesva 2009;
Starks et al. 2008), and to create or reinforce group-specific interactional styles
(Barbieri 2008; Coates 2004: 88–92; Holmes 1986, 1995: 78–101; Macaulay 2002c).
 Chapter 1.  Introduction 13

It follows that the social embedding of discourse-pragmatic variation differs from

that of phonological and lower-level morpho-syntactic variation in that it hinges
on functionality at least as much as on frequency.
Variationist studies of sound change have plotted variants on a linear age-scale
in order to establish the direction and social embedding of potential changes. The
consistency of sex-differentiated patterns of change uncovered in these studies
led Labov (1998) to formulate general principles of sound change which highlight
the role of women as linguistic innovators: women lead in the importation of
new prestige forms from outside the speech community, the elimination of forms
that have become stigmatised, and changes that operate within the speech com-
munity (see, however, Labov 1963; Milroy & Milroy 1998). Foulkes & Docherty
(1999: 16) and Milroy et al. (1994) also draw attention to sex-differentiated patterns
in the spread of sound change, proposing the following generalisations: women
are instrumental in the spread of supra-local phonological variants, i.e., those
that have a wide geographical distribution, while men favour the use of localised
phonological variants, i.e., those that are characteristic of given localities. Due to
deep-rooted differences in the linguistic conditioning of changes across lower-
and higher-level linguistic variables, the above generalisations may not extend
to discourse-pragmatic change in any straightforward way. Discourse-pragmatic
features are generally understood to result from grammaticalisation which involves
multiple changes affecting multiple levels of language use: phonetic reduction,
decategorialisation, semantic-pragmatic shift (see further Chapter 2.4.1). These
changes, although together constituting grammaticalisation, are separate proc-
esses which may not always co-evolve synchronously (Bisang 2004; Bybee 2010;
Romaine 1995). Consequently, different processes of change may be advanced by
different social groups at different points in time (see also Zilles 2005). Further,
since the development of discourse-pragmatic features is not by necessity unidi-
rectional or unidimensional (see, for example, Buchstaller [2004] who conceptu-
alises the multifunctional uses of discourse like as a radial functional network),
different social groups may initiate and advance different functional developments
of individual variables (see D’Arcy 2005: 222–224). Finally, because of the open-
ended nature of the functional categories underpinning the conceptualisation of
discourse-pragmatic features as linguistic variables (see further Chapter 2.3.2),
different social groups may simultaneously propel the diffusion of different new-
comer variants (see Tagliamonte 2008). It follows that, unlike the case in sound
change, in discourse-pragmatic change it may not be possible to attribute the role
of innovators to a single social group.
Beyond sharing with phonological and morpho-syntactic variables a capacity
for diachronic change and social variation, discourse-pragmatic variables resemble
other variables in being subject to geographical dialect variation. Recent discourse
14 The Structure of Discourse-Pragmatic Variation

variation studies have revealed the following findings. Firstly, the occurrence of
discourse-pragmatic variants may be geographically restricted (see, for example,
Macaulay [2009] on the use of be all as a quotative complementiser in Glasgow
English; Sowa [2009] on the use of adjective + as as an intensifier in Australian
and New Zealand English). Secondly, their relative frequency and sociolinguistic
distribution may vary across geographical space (see, for example, Cheshire et al.
[2005] on the use of innit across three English towns; Denis [MS] on the use of and
stuff in the English town of York and the Canadian city of Toronto). Thirdly, and re-
latedly, variants may be more socially diffused and more grammaticalised in some
varieties than others (see, for example, Buchstaller & D’Arcy [2009] on quotative
be like in British, American and New Zealand English). Nonetheless, systematic
accounts of dialect variation in discourse-pragmatics are virtually non-existent.
For example, recently published overviews of British English dialect variation,
including Hughes et al. (2012) as well as the contributions in Britain (2007) and
Kortmann et al. (2004), focus almost exclusively on phonological and morpho-
syntactic variation. Discourse-pragmatic variation is mentioned only in passing, if
at all, thus leaving a large proportion of speakers’ linguistic repertoires virtually un-
accounted for.3 The cause for this neglect does not lie with the authors or the editors
of these volumes. Systematic analyses of discourse-pragmatic variation are in such
short supply that attempts to reconcile the findings from individual studies into
a coherent overview of geographical dialect variation reveal little else but glaring
gaps in the canon of variables as well as localities that have been investigated. The
problem is exacerbated by the fact that meaningful accounts of dialect variation
in discourse-pragmatics cannot be produced based on cross-variety comparisons
of variables’ or variants’ frequency rates. In order for such accounts to be of any
descriptive or theoretical value at all, they require systematic comparison of the
variable grammar underlying variables’ or variants’ use. Only if this information
is readily available can scholars establish whether variables and their variants have
the same social and functional meanings across varieties and whether they have
been grammaticalising along identical paths, at identical speeds and to identical
degrees. Alas, even where a sufficient number of studies are available that investi-
gate discourse-pragmatic features’ variable grammars, their results are not easily
comparable due to a lack of uniform methodologies (see further Pichler 2010).
It follows that the current dearth of systematic and accountable quantitative dis-
course studies is a serious shortcoming in descriptive sociolinguistics.

3. Fischer (2000: 11) and Rudolph (1991: 208) estimate that discourse-pragmatic features con-
stitute between ten to 24 per cent of all words spoken in informal interactions.
 Chapter 1.  Introduction 15

In addition to being indispensable for establishing the precise nature and full
extent of geographical dialect variation and for probing the robustness of long-
standing principles of language variation and change that are based largely on
the analysis of phonological and morpho-syntactic variables, discourse variation
studies are of great value for advancing theories and descriptions of discourse-
pragmatic features themselves. In stark contrast to their finite coverage in vari-
ationist linguistics, discourse-pragmatic features have figured very prominently
in qualitative research paradigms. Qualitative discourse studies have investigated
the linguistic attributes, functional versatility, context-boundedness and dia-
chronic evolution of discourse-pragmatic features as well as, to a more limited
extent, their cross-linguistic resemblance, use in bilingual interactions, and ac-
quisition by children and second language learners (see, for example, Aijmer &
Simon-Vandenbergen 2011; Brinton 1996: Chapter 2; Fischer 2006; Fox Tree 2010;
Schiffrin 2001; Schourup 1999 for overviews). I will argue below that the patterns
and regularities discovered in quantitative discourse studies can bear on a number
of these issues, serving in particular to verify current and construct new descrip-
tions of the usage, distribution and evolution of discourse-pragmatic features.
Through adherence to the “principle of accountability” (Labov 1972: 72), quan-
titative discourse studies exhaustively report all occurrences of a targeted variable
in a given dataset. (Where it is possible to determine all potential environments for
its occurrence, non-occurrences are also reported.) Because this approach gener-
ates accounts of discourse-pragmatic variables which are firmly grounded in the
complete set of available data, it constitutes a valuable check on any descriptions
provided by qualitative studies that are based on (either randomly or carefully)
selected subsets of available tokens. For example, Andersen’s (2001: 227–271) ac-
countable analysis of the discourse-pragmatic variable like in the Bergen Corpus of
London Teenage Language (henceforth COLT) established that discourse like per-
forms a wide range of subjective, textual and quotative functions, with the former
far outnumbering the latter. Andersen’s findings thus challenge the comprehen-
siveness and adequateness of the descriptions yielded by Jucker & Smith’s (1998),
Miller & Weinert’s (1995) and Romaine & Lange’s (1991) qualitative studies which
have foregrounded individual functions of like at the neglect of others. Moreover,
quantitative methods make possible reliable comparison of the functional profile
and proportional or relative frequency of discourse-pragmatic features across di-
verse social and interactional-situational contexts. As shown inter alia by Fuller
(2003) and Schleef (2008), such comparisons have the potential to provide numeri-
cal evidence for claims made in the qualitative literature about the sensitivity of
discourse-pragmatic features to external constraints.
Further insights which are relevant to modelling regularities in the usage and
distribution of discourse-pragmatic features can be gleaned from the stepwise
16 The Structure of Discourse-Pragmatic Variation

multiple-regression procedure incorporated in the variable rule programme (see

further Chapter 2.3.3). By determining the relative importance of multiple contex-
tual predictors on the distribution of variants when all predictors are considered
simultaneously, this procedure yields fine-grained and reliable descriptions of
the social, discourse-functional and linguistic-structural properties of discourse-
pragmatic­ variables and their variants. For example, Pichler & Levey’s (2011)
multivariate analyses of general extenders in Berwick English discredit previ-
ous claims in the literature that general extenders function first and foremost to
implicate a more general set (Dines 1980: 22–23), or that short variants occur by
definition in extended morpho-syntactic contexts (Aijmer 2002: 249). When car-
ried out on socially stratified real- or apparent-time data, multivariate analyses
can also uncover subtle shifts over time in the sociolinguistic conditioning of dis-
course-pragmatic features which may serve to verify claims from the qualitative
literature about features’ diachronic development. For example, D’Arcy’s (2005)
multivariate analysis of discourse like in Toronto English models the gradual
and systematic diffusion of like across syntactic contexts. The results challenge
developmental trajectories of like depicted as preposition/conjunction > discourse
marker (Meehan 1991) and contest claims that like underwent abrupt scope ex-
pansion (Romaine & Lange 1991). They postulate instead that like developed as a
sentential adverb before it developed as a discourse marker. Moreover, variationist
discourse studies of the kind described above can shed light on the social forces
driving discourse-pragmatic­ developments. As shown inter alia by Andersen
(2001) and Tagliamonte & D’Arcy (2007), these studies thus complement more
traditional grammaticalisation research which focuses on the structural dimen-
sion and language-internal motivation of language change phenomena (see, for
example, Brinton 1996, 2008).
The preceding discussion in no way negates the intrinsic value of qualitative
discourse studies. Qualitative studies reveal patterns that may remain hidden in
quantification, such as the clustering or scattering of discourse-pragmatic fea-
tures at certain points of an interaction (Bell & Johnson 1997). Importantly, the
distributional properties of discourse-pragmatic features cannot be interpreted
accurately without qualitative investigation of their functionality (see further
Chapter 2.5). The fact remains, though, that in addition to providing the empiri-
cal evidence required to advance current knowledge of language variation and
change mechanisms, systematic, exhaustive and accountable quantitative analy-
ses have the potential to test, validate and refine current models of discourse-
pragmatic features.
 Chapter 1.  Introduction 17

1.5 Aims and focus of the book

This book reports the results of an in-depth investigation into patterns of variation
and change in the morpho-phonological encoding of three discourse-pragmatic
variables in the variety of English spoken in Berwick upon Tweed, a small market
town in the far north-east of England. The variables chosen for analysis are the
negative polarity constructions i don’t know, i don’t think and neg-tags. They
are construed as linguistic variables by virtue of the fact that their respective vari-
ants (differentiated by the variable realisation of the negative auxiliary and/or the
optional fusion of adjacent morphemes) are derivations of the same linear string
of components (see further Chapter 2.3.2). Research investigating these variables
in other varieties has shown that their formal variability is closely constrained by
extra- and intra-linguistic predictors (Andersen 2001; Bybee & Scheibman 1999;
Cheshire 1981). The principal objective of the current investigation is to establish
whether similar constraints on variant choice are also operative in the present da-
taset of Berwick English (henceforth BwE) where non-localised variants co-exist
alongside supra-local and localised variants in the pool of available variants. By
establishing the sociolinguistic mechanisms that give rise to observed patterns of
variability in the form, frequency and strategic use of the selected variables, this
project uncovers important new insights into the creation and organisation of
synchronic language variation. It thereby demonstrates the important contribution
discourse variation analysis can make to current theories of language.
To achieve the objectives set out above and capture the complex nature of
discourse-pragmatic variation and change without sacrificing accountability, the
study develops an innovative methodology which combines the theoretical in-
sights and empirical methods from variationist sociolinguistics, grammaticali-
sation studies and conversation analysis. Close integration of these approaches
combined with new ways of conceptualising discourse-pragmatic variables, clos-
ing the envelope of variation, and quantifying multifunctionality ensure reliable
and intersubjective results as well as accountable and theoretically relevant ex-
planations of these results. Through full articulation and detailed illustration of
the methodology, the study provides a framework in which future studies can be
planned and executed.
The viability of the approach is demonstrated by the empirical and theoretical
insights it affords. The results of the analysis demonstrate that the availability of a
range of formal variants to encode discourse-pragmatic variables is strategically
exploited by speakers: the variation between non-localised variants of i don’t
know and i don’t think signals important meaning differences; the use of local-
ised/supra-local variants of i don’t know, i don’t think and neg-tags indexes
social identities which are closely linked to variables’ and/or variants’ functional
18 The Structure of Discourse-Pragmatic Variation

compartmentalisation in the variety. These correlations highlight the centrality

of discourse-functional constraints in accounting for formal and social variation
patterns in discourse-pragmatics, and demonstrate the complex and potentially
isomorphic nature of discourse-pragmatic variability. Moreover, the analysis un-
covers that the different realisations of negative periphrastic do in BwE do not have
the same social and functional meanings across the formulaic and productive con-
structions investigated in this book. These differences highlight the prominent role
of formulaic constructions in shaping linguistic variation, and demonstrate that the
diachronic processes creating synchronic variation affect constructions heteroge-
neously. Close inspection of variant distributions affords far-reaching hypotheses
about the social and interactional causes of this heterogeneity. In sum, the inves-
tigation shows that systematic and accountable analyses of discourse-pragmatic
features are crucial to advancing current understanding of the sociolinguistic em-
bedding of language variation as well as the sociolinguistic mechanisms creating it.

1.6 Organisation of the book

Part I consists of three chapters which provide the context for the empirical analysis
of the selected discourse-pragmatic variables in Part II. Chapter 1 has reviewed the
current state of discourse variation analysis and has outlined reasons for the fuller
integration of discourse-pragmatic features in variationist research. Chapter 2 in-
troduces the corpus data on which this analysis is based and articulates the innova-
tive methodology developed for this project to uncover patterns of variation and
change in the formal encoding of i don’t know, i don’t think and neg-tags.
To contextualise patterns of variation and change in the use of these variables,
Chapter 3 introduces the verb negation system of BwE and explores its variabil-
ity in productive constructions. This chapter also introduces in more detail the
discourse-pragmatic variables chosen for in-depth analysis in Part II.
Part II constitutes the analytical heart of the book. It is divided into three
chapters, one for each of the selected variables. The three chapters are organised
uniformly to facilitate comparison of the analyses and results. Each chapter be-
gins with a short general introduction and a thorough review of previous studies
investigating the targeted variable. There follows an outline of how the variable
was circumscribed and how the data were coded for quantitative analysis. A con-
cise description of the multiple functions performed by each variable in the BwE
data precedes the quantitative analysis which tests the combined impact on the
observed variation of multiple contextual factors. In the discussion section, the
findings obtained from the quantitative analysis are compared with those obtained
in related studies, and the mechanisms giving rise to the variable’s synchronic
 Chapter 1.  Introduction 19

distribution patterns are examined. This is necessary to understand the nature and
origin of the formal and functional variation in the data, and to appreciate the com-
plex mechanisms underlying discourse-pragmatic variability which are explored
further in Part III. Although Chapters 4 to 6 build on each other to cumulatively
demonstrate the complex nature of discourse-pragmatic variation and change as
well as the central role of discourse-pragmatic variables in the linguistic system, the
chapters are organised in such a way that they can act as self-contained standalone
studies. Inevitably, this leads to some repetition of methodological information
across the three chapters.
Part III assesses the methodological and theoretical implications of the results
obtained in Part II as well as the future of the field. Chapter 7 synthesises the
results produced in Part II with a view to: (1) demonstrating their far-reaching
methodological and theoretical implications for grammaticalisation studies and
variationist sociolinguistics; (2) illuminating why discourse-pragmatic variables
do not correspond fully to patterns of variation and change reported for other
levels of the linguistic system; and (3) illustrating the fully grammatical status of
discourse-pragmatic variables in the linguistic system. Finally, Chapter 8 identi-
fies and explores some of the challenges discourse variation analysis still faces in
its attempt to provide comprehensive accounts of discourse-pragmatic variation
and change. Some suggestions are provided as to how these challenges might be
overcome and how the field might be advanced beyond its current state.
 chapter 2

Data, methodology
and theoretical framework

2.1 Introduction

The present investigation explores the sociolinguistic mechanisms underlying

patterns of variation and change in the morpho-phonological encoding of the
discourse-pragmatic variables i don’t know, i don’t think and neg-tags with a
view to providing new insights into the structure of discourse-pragmatic variation
and demonstrating the descriptive, explanatory and theoretical value of discourse
variation analysis. Achievement of these objectives is contingent on a methodol-
ogy which computes discourse-pragmatic variation and change in a principled and
accountable manner. Alas, this criterion is not easily met, for discourse-pragmatic
features eschew easy definition as linguistic variables. Identifying all contexts in
which a discourse-pragmatic feature can occur and closing off the set of compet-
ing variants are far from straightforward tasks. Matters are complicated by the
fact that discourse-pragmatic features are variable and changeable on multiple,
closely interlinked levels: their frequency, form and functionality as well as their
distribution across morpho-syntactic contexts, interactional-situational settings
and social groups. What is needed, then, to provide a comprehensive and illumina-
tive account of discourse-pragmatic variation and change is a multidimensional
methodology which accommodates the full complexity of synchronic discourse-
pragmatic variation.
The innovative methodology developed for the present investigation and
outlined in part in Pichler (2010) meets this criterion. It combines the empirical
methods and theoretical insights from variationist sociolinguistics, grammati-
calisation studies and conversation analysis (henceforth CA) in order to uncover
and account for patterns of formal discourse variation and change in synchronic
dialect data. Insights from variationist sociolinguistics inform the sampling deci-
sions and data collection methods; the quantitative methods associated with the
variationist paradigm are employed to determine the joint effect of contextual fac-
tors on observed patterns of variation. The grammaticalisation framework provides
hypotheses about the nature and trajectory of linguistic changes; these hypotheses
inform the coding decisions as well as the interpretation of the synchronic distribu-
tion patterns of discourse-pragmatic variants. Finally, CA theories and methods
22 The Structure of Discourse-Pragmatic Variation

are drawn on in the qualitative data analysis to establish the functional repertoires
of the selected variables and allocate each token of the variables to a functional
category; these categories constitute factors of the factor group functionality which
is included in the quantitative data analyses to determine its impact on variant
choice. The analyses in Part II will demonstrate that a multidimensional approach
like the one introduced here is a methodological necessity if our aims are: to cap-
ture the complex nature of discourse-pragmatic variation and change, and to fully
understand its underlying mechanisms.
The remainder of this chapter introduces in more detail the theoretical and
analytical framework that underpins the present investigation as well as the data
on which the investigation is based. Section 2.2 describes the corpus and provides
details about the design of the speaker sample. It includes a critical assessment of
the strengths and limitations of interview data for variationist discourse studies
as well as a rationale for the choice of extra-linguistic variables used. The theories
and methods associated with variationist sociolinguistics, the grammaticalisation
framework and CA are introduced in Sections 2.3 to 2.5. Section 2.3 is a detailed
outline of the quantitative methods used. It explores the applicability of the vari-
ationist paradigm and the linguistic variable to higher-level linguistic analysis,
and discusses the technicalities and practicalities of multivariate statistical analy-
sis with Rbrul. The grammaticalisation framework is introduced in Section 2.4.
It outlines the processes of change constituting grammaticalisation, explores the
applicability of the framework to the analysis of discourse-pragmatic features and
synchronic data, and presents the methods used for identifying and tracing ongo-
ing grammaticalisation in synchronic cross-generational data. Finally, Section 2.5
introduces the principal concepts of CA which guide the functional data analy-
sis. It also sets out the functional domains in which discourse-pragmatic features
operate, discusses how this study operationalises multifunctionality for quantita-
tive analysis, and summarises the methods employed for validating the qualitative
analyses in Chapters 4 to 6.

2.2 Data

2.2.1 Corpus

Successful variation studies require access to language as it is actually used by

speakers: empirical data constitute the basis for testing hypotheses, performing
analyses and formulating theories. To meet this requirement, the present investiga-
tion into discourse-pragmatic variation and change is based on a corpus of some
260,000 words of transcribed speech which I collected between 2003 and 2005 in
 Chapter 2.  Data, methodology and theoretical framework 23

Berwick upon Tweed. Berwick is a small market town of some 13,040 inhabitants
located in the far north-east of England, only three miles (five kilometres) south
of the Scottish-English border (see Figure 2.1).4

EDINBURGH BERWICK UPON

SCOTLAND MIDLOTHIAN TWEED

SCOTTISH
BORDERS

NEWCASTLE
ENGLAND NORTHUMBERLAND UPONTYNE

NORTH TYNESIDE
GATESHEAD SOUTH TYNESIDE
SUNDERLAND

DURHAM
STOCKTON-ON-TEES HARTLEPOOL
CUMBRIA REDCAR &
CLEVELAND

DARLINGTON MIDDLESBROUGH

Figure 2.1  Location of Berwick upon Tweed, Northumberland*

* ©Crown copyright 2013. All rights reserved. Reproduced with permission from Watt & Ingham (2000).

The data were gathered using the interview protocol designed for the Survey of
Regional English to solicit information about informants’ use of local dialect words
and attitudes towards their locality and dialect (Llamas 2007a). Scholars have ex-
pressed reservations not only about the efficacy of sociolinguistic interviews to
overcome the Observer’s Paradox and elicit informants’ vernacular style “in which
the minimum attention is given to the monitoring of speech” (Labov 1972: 208)
but also about the ability of non-community members and non-native speakers
to tap into the vernacular (see in particular Douglas-Cowie 1978; Hazen 2000;
Wolfson 1976). I do not deny the possibility that any potential effects of the re-
cording situation on the nature of the data elicited may have been exacerbated by
my status as a non-native speaking outsider to the community. Yet the quality of
the data suggests that such effects were successfully attenuated through the use of

4. The word count includes false starts, truncations, filled pauses and minimal response par-
ticles, but excludes the interviewer’s contributions. It equals some 35 hours of recorded speech.
24 The Structure of Discourse-Pragmatic Variation

various fieldwork techniques including conducting interviews with self-selected

pairs in informants’ homes. As demonstrated in Examples (1)–(7) below, the data
manifest vernacular structures and inherent variability characteristic of casual and
unmonitored speech styles (Labov 1972: 85–99).5
(1) And you know for a fact where they are, what they’re up to, and other people’s
looking out for them because everybody knows everybody else. It’s no like
living in a city where you divn’t know who lives next door to you, you know.
We might no know their name but we know what they look like.
(2) I divn’t knaa. I cannae mind.
(3) Well, I don’t know whether you realise it or no, but it’s dearer to live in Scotland
than what it is to live in England. See, we’re on a thing called stay-warm for the
heating. Between wa gas and electric. We pay every, I pay every month for my
stay-warm. Now, if I lived up in Scotland, I’d be paying at least another fifteen
or seventeen pound a month more than what I do by living in England.
(4) I ken his dad, you know.
(5) It depends on whae you was talking to. If I was talking to him. He’s a Spittaler
and, but if we was talking to someone that’s not from Berwick, I would just
say we’re Berwickers.
(6) I haven’t got nowt down for that. I can’t think of anything.
(7) I seen a bit of a programme last year Ø was on about Berwick. Something to
do wi, I think, something to do with the parking all the cars and that. Some
people reckon they’ve ruined it putting these things down. The shopkeepers
is losing money and that now. And obviously cars isn’t allowed to park on the
street. And well, they’re no going into the shops. The street’s nice-looking but
th- th-, you know, at one time there was all cars up there. And people would
go into Boots and Woolworths but they cannae do that now.

These examples also illustrate the hybrid nature of BwE, which exhibits charac-
teristics of both Northumbrian and Lowland Scottish dialects (see further Llamas
et al. 2009; Watt & Ingham 2000; Watt et al. 2012). For example, in Extracts (1)–
(2), speakers use negative auxiliary variants that are typically associated with
Northumbrian varieties (divn’t for ‘don’t’) alongside ones that are typically associ-
ated with Scottish varieties (cannae, ’s no, might no for ‘can’t, ’s not, might not’).
In Extracts (2)–(6), they use lexical items that are usually associated with Scottish
varieties such as mind (‘remember’), ken (‘know’) and whae (‘who’) as well as lexi-
cal items that are usually associated with Northumbrian and Tyneside varieties
such as knaa (‘know’), dear (‘expensive’) and wa (‘our’). This linguistic hybridity

5. Page xix provides a glossary of all the non-standard dialect words that are contained in the
verbatim data extracts given here and elsewhere in the book.
 Chapter 2.  Data, methodology and theoretical framework 25

is due to the town’s border location, its turbulent history as a place which was
repeatedly sacked and recaptured by Scottish and English kings, its continued
importance as a communication point between England and Scotland as well as
BwE’s shared history with dialects spoken on both sides of today’s border (Murray
1873: 5–6; Wales 2006: 49).
An important advantage offered by sociolinguistic interviews is that the data
obtained are comparable in content and context: the same topics are discussed in
the same setting with all speakers. This reduces the likelihood that any intra- and
inter-group differences in the frequency and functionality of i don’t know, i don’t
think and neg-tags reported in Chapters 4 to 6 are an artefact of interactional-
situational constraints on their use (see Pichler 2010: 584–586). However, due to
the strong context-dependency of discourse-pragmatic features, any conclusions
drawn in this book about the targeted variables’ functional profiles and typical rates
of occurrence may not generalise to other speech events (Fuller 2003; Lam 2009).
This is not to distract from the value of the present investigation which offers im-
portant new insights into the functional versatility of i don’t know, i don’t think
and neg-tags and thereby improves currently available descriptions of their use.

2.2.2 Speaker sample

Table 2.1 gives a breakdown of the speaker sample constructed for this study. It
includes 36 speakers who are equally stratified across speaker sex and age, and who
represent a socially homogeneous group of working-class speakers, as determined
via a combination of traditional social class indicators (housing, education, occu-
pation) and informants’ own assessment of their social class membership.

Table 2.1  Speaker sample

young (17–23) middle (27–48) old (60–81)
male female male female male female
6 6 6 6 6 6

The stratification of speaker samples across pre-fabricated social categories has

recently come under criticism. It has been argued that social identities are not
monolithic speaker attributes but dynamic products of social interaction which are
locally situated and emerge “from practice, from what people do rather than what
they intrinsically are” (Cameron 2005: 488–489, italics in the original) (see also
Eckert 1997; Eckert & McConnell-Ginet 1992). The present investigation is not
concerned with the performative nature of social identities or with how individuals
26 The Structure of Discourse-Pragmatic Variation

exploit sociolinguistic resources to construct social identities. Its objective is to

explore the sociolinguistic embedding of variation and change in discourse-prag-
matics. It therefore utilises sex and age as broad methodological and exploratory
tools to test the combined effect of extra- and intra-linguistic factors on discourse-
pragmatic variation and change (Milroy & Milroy 1997: 53). This enables us to
draw comparisons with studies of phonological (and morpho-syntactic) variables
which have associated men with the use of non-standard or localised variants, and
credited women with community-internal changes, the elimination of stigmatised
variants as well as the spread of prestige and supra-local variants (Labov 1998;
Milroy et al. 1994).6
In order to identify the existence and direction of any ongoing changes in the
data, I draw on the apparent-time framework as a surrogate for real-time data
(Bailey 2002). The apparent-time construct hinges on the notion that the speech
patterns recorded for different age groups at a single synchronic point in time
represent different diachronic stages in language change. Age-based differences in
synchronic speech patterns can therefore be interpreted as evidence of ongoing,
real-time linguistic change. Inferences about change in progress that are drawn
on the basis of apparent-time data rely on the basic hypothesis that individuals do
not fundamentally alter their speech patterns as they grow older. The empirical
validity of such inferences – and of the apparent-time construct more generally –
has been confirmed by real-time panel studies attesting the stability of vernaculars
throughout adult life (Sankoff & Blondeau 2007) as well as by real-time trend
studies verifying the hypotheses inferred from synchronic data with diachronic
evidence (Pope et al. 2007). To operationalise apparent time and add explanatory
value to age as a dynamic social variable (Eckert 1997), the sample for this study
was stratified across three emically-defined age cohorts reflecting shared life stages:
teenagers and young adults in short-term unsalaried employment or full-time
education and co-habiting with their parents (ages 17 to 23); adults in full-time
salaried employment with dependent children and their own household (ages 27
to 48); retired people with no dependent children (ages 60 to 81).

6. No attempt has been made in this study to include in the analysis social-psychological variables
such as local loyalty or identity. This is not to deny the possibility that speakers’ frequent use of
localised variants such as I divn’t knaa or I divn’t think may reflect their locally oriented (as opposed
to outwardly oriented) identities. The exclusion of such variables from the analysis was motivated
by the aim of the study: to provide an account of the synchronic structure of discourse-pragmatic
variation which can be compared and contrasted with existing accounts of the synchronic structure
of phonological (and morpho-syntactic) variation. Since the latter are generally based on inclusion
of broad demographic variables such as speaker sex and age, it is important for the purpose of
comparison and generalisability that extra-linguistic factors are kept constant.
 Chapter 2.  Data, methodology and theoretical framework 27

2.2.3 Summary

Access to a corpus of BwE makes it possible to investigate discourse-pragmatic

variation and change in a transitional and peripheral variety of British English,
and it also complements previous discourse variation studies which have tended to
focus on more mainstream varieties. The stratification of the speaker sample along
two broad social dimensions, sex and age, affords us the opportunity to establish
whether the sex effect in discourse-pragmatic variation and change is comparable
to that in phonological or morpho-syntactic variation and change, and to trace the
progress of any ongoing discourse-pragmatic changes in apparent time.

2.3 Variationist sociolinguistics

As stated in Chapter 1, a central aspect of the current investigation is to uncover

which contextual factors contribute to the choice process between formal variants
of the variables i don’t know, i don’t think and neg-tags, and to determine
whether the sociolinguistic conditioning of variant choice is changing in appar-
ent time. To this end, quantitative methods of variationist sociolinguistics are
adopted. This section outlines the main theoretical principles of the variationist
paradigm, argues in favour of its extension to the analysis of discourse-pragmatic
features provided that the original definition of the linguistic variable is modi-
fied accordingly, and describes how the choice process can be reliably modelled
with Rbrul.

2.3.1 Premises of variationist sociolinguistics

The quantitative variationist paradigm originates in the work conducted by Labov

(1963, 1966, 1972) on the American east coast in the 1960s and 1970s. It is based
on two fundamental premises. Firstly, that the choice process between variable lin-
guistic forms is systematically constrained by multiple contextual factors, including
“the phonological environment, the syntactic context, the discursive function of
the utterance, topic style, situation and personal and/or socio-demographic char-
acteristics of the speaker and other participants” (Sankoff 1988: 151). Secondly,
that ongoing linguistic changes may be inferred from fluid patterns in the socio-
linguistic conditioning of variant choice across different age groups in a commu-
nity (Labov 1994) (see Section 2.2 above). Support for these premises is provided
through quantitative analysis, which involves examining patterns in the distribu-
tion of a linguistic form across contextual factors and expressing in quantitative
28 The Structure of Discourse-Pragmatic Variation

terms the strength of co-occurrence between them (“principle of quantitative

modelling,” Bayley 2002: 118). Co-occurrence patterns are taken to represent the
social and grammatical structure giving rise to the observed variability which,
viewed in apparent time, may yield insights into the embedding and trajectories
of ongoing changes.

2.3.2 Discourse-pragmatic variables

The key theoretical construct of the variationist framework is the linguistic vari-
able, which was initially developed by Labov (1963, 1966) for the analysis of pho-
nological variation. In its most basic sense, a variable is a linguistic feature with
two or more identifiable realisations, i.e., variants, which differ in their social and
stylistic distribution but are equivalent in referential meaning (“saying ‘the same
thing’ in several different ways,” Labov 1972: 271). In order to fully understand
the mechanisms underlying the choice process, it is necessary to isolate the whole
set of possible variants and calculate each variant’s occurrence out of all possible
contexts of occurrence, i.e., those where it did occur and those where it could
have occurred but did not (“principle of accountability,” Labov 1972: 72). Because
discourse-pragmatic features have unique semiotic and distributional properties,
it is not easy to apply the parameters outlined above to their conceptualisation
as variables or to their quantitative analysis. Firstly, discourse-pragmatic features
are typically semantically bleached and therefore cannot be defined in terms of
semantic equivalence between variants. Secondly, they are typically both refer-
entially and syntactically optional, and thus eschew straightforward reporting as
non-occurrences (see, however, D’Arcy 2005). Consequently, it is not immediately
obvious on what basis to identify co-variants of a discourse-pragmatic variable and
how to produce accountable results.
Due to these difficulties, scholars have often resorted to analysing discourse-
pragmatic variation and change without reference to the linguistic variable. The
most common approach has been to generate (normalised) frequency tabulations
for individual discourse-pragmatic features without any consideration of their
potential co-variants. Through comparison of these frequencies across individu-
als, social groups and/or communities, scholars strove to gauge variables’ social
indexicalities and test their robustness in and across communities (e.g. Macaulay
2002c; Stubbe & Holmes 1995; Tagliamonte 2005). However, this approach ignores
the fact that discourse-pragmatic features may derive their social indexicality not
just from frequency alternations but also from: variation in their form (see, for
example, Cheshire [2007] who found that although working-class and middle-
class adolescents use the general extender variable with comparable frequency,
they favour different general extender variants); variation in their strategic usage
 Chapter 2.  Data, methodology and theoretical framework 29

(see, for example, Erman [1992] who found that women use you know, you see,
I mean to link consecutive arguments while men use these features to draw atten-
tion or signal repair); and variation in their syntactic-semantic distribution (see,
for example, Andersen [2001] who found that ethnic groups are not differentiated
by their use of the tag variant innit per se but by their use of innit in non-third
person singular contexts with be). Furthermore, the approach described above
treats discourse-pragmatic features as if they operated in a systemic vacuum. This
is highly problematic since the overall structure of variation and change cannot be
detected if variable features are analysed in complete isolation from their potential
co-variants (Winford 1996: 189). An alternative but equally problematic approach
has been to quantitatively model the distribution and conditioning of alternative
forms without conceptualising them as co-variants of a linguistic variable (e.g.
Cheshire 2007; Ferrara 1997; Macaulay 1995). Unless the forms included in an
analysis are identified under the umbrella of a variable, it may prove difficult to
establish how these forms are related to each other, what the theoretical basis is
for including them in the same analysis, and what forms are, in fact, subjected to
quantitative analysis. As a result, it may be difficult to replicate these studies or
discern how scholars arrived at their results.
It is clear, then, that the extension of the linguistic variable to the level of dis-
course-pragmatics is a methodological necessity for producing reliable, intersub-
jective and accountable results. Yet despite Sankoff ’s (1973: 44,58) reassurance that
the extension of the linguistic variable to levels “above and beyond phonology [is]
not a conceptually difficult jump,” suggestions to do so fuelled a heated controversy
which continued throughout the 1970s and 1980s. Dines (1980) and Lavandera
(1978), for example, questioned the methodological and theoretical soundness of
scholars’ wholesale transfer of the linguistic variable to the analysis of variable non-
phonological features because referential meaning cannot be employed to establish
equivalence relationships between higher-level variants (see above). Dines’s and
Lavandera’s proposals to modify the defining criterion of the linguistic variable
for the analysis of discourse-pragmatic (and syntactic) features were criticised by
Cheshire (1987) and Romaine (1984) for being made without consideration of the
theoretical assumptions underlying the variable’s original conception. Yet it is hard
to conceive how the mechanisms underlying discourse-pragmatic variation can be
satisfactorily explained without modifying the definition of the linguistic variable
in accordance with the unique semiotic and distributional properties of discourse-
pragmatic features or without catering for the differential motivation for their use.
Dines (1980) and Lavandera (1978) therefore proposed to abandon the cri-
terion of semantic equivalence for defining discourse-pragmatic (and syntactic)
variables and to substitute it for one of functional comparability. For example,
general extenders, i.e., clause- or phrase-final constructions such as and that, and
30 The Structure of Discourse-Pragmatic Variation

stuff like that, or something, were argued to constitute a variable on the basis that all
of its variants perform the common function of “cue[ing] the listener to interpret
the preceding element as an illustrative example of some more general case” (Dines
1980: 22). The function-based conceptualisation of discourse-pragmatic variables
has proven very successful and has been widely adopted in studies of quotatives
(“all strategies used to introduce reported speech, sounds, gesture and thought
by self or other,” Buchstaller 2006a: 5; see also Macaulay 2001; Tagliamonte and
Hudson 1999) and intensifiers (“every option speakers have at their disposition
to reinforce or boost the property denoted by the head they modify,” Rickford
et al. 2007: 7; see also Ito and Tagliamonte 2003; Macaulay 2006). The appeal of
Dines’s (1980) and Lavandera’s (1978) definition is that it allows scholars to treat
the general extender, quotative and intensifier systems as closed sets of variants
and model the distribution of variants within a standard variationist framework.
Moreover, Terkourafi (2011) has recently capitalised on the notion of procedural
meaning as expounded in relevance theory (Sperber & Wilson 1995) in order
to provide a theoretical justification for the function-based conceptualisation of
discourse-pragmatic variables. Discourse-pragmatic features encode procedural
meanings which “constrain the inferential phase of comprehension by indicating
the type of inference process the hearer is expected to go through” (Wilson &
Sperber 1993: 11; see also Blakemore 1987). Thus, Terkourafi (2011: 363) argues,
quotatives instruct hearers to “interpret the speech that follows as attributed to
a speaker other than the speaker’s current self;” and intensifiers instruct them to
“interpret the property denoted by the head as one that the speaker wishes to fore-
ground or reinforce.” The stable inference processes constitute a unifying criterion
for establishing equivalence relationships between quotative variants on the one
hand and intensifier variants on the other hand.
Yet as discussed in detail in Pichler (2010: 587–591), the function- or inference-
based conceptualisation of discourse-pragmatic variables is not universally applica-
ble. Because function is an open category, some functions such as hedging may be
signalled by a vast number of linguistic features from different levels of the linguistic
system. Not only might it then be beyond the remit of individual analyses to close
off the set of variants and report each variant’s actual and potential occurrences, but
including a range of elements from different components of the grammar “in the
same analytic unit […] is surely stretching the concept of the variable beyond all
credibility” (Cheshire et al. 2005: 164). Moreover, postulating discourse-pragmatic
variables on the basis of shared functions or inference processes disregards the
synchronic polyvalence of form-function relationships stemming from grammati-
calisation. Older and newer layers of language use regularly co-exist synchronically
either as variation in form, whereby different forms are used side-by-side for similar
or identical functions (see above), or as variation in function, whereby a single item
or construction serves a diversity of functions (see Chapters 4 to 6). Function- or
 Chapter 2.  Data, methodology and theoretical framework 31

inference-based approaches to defining discourse-pragmatic variables are not easily

reconcilable with the latter layering effect.7 What is needed to quantify variation in
the use of multifunctional discourse-pragmatic features is a definition of discourse-
pragmatic variables which accommodates their polysemic layering in synchronic
data while still ensuring that “the variants are in some way the same, have something
in common” (Dines 1980: 18; italics in the original).
A definition which fulfils these criteria and which is adopted for the present
investigation is one based on variants’ derivational equivalence (see also Pichler
2010). In this definition, i don’t know and i don’t think constitute variables
by virtue of the fact that their respective variants are derived from the same linear
string of components schematised as: (first person singular pronominal subject) +
(negative periphrastic do) + (predicate know)/(predicate think). neg-tags consti-
tute a variable by virtue of the fact that its variants are derived from the same linear
string of components schematised as: (auxiliary) + (negator) + (pronoun) or (aux-
iliary) + (pronoun) + (negator). As detailed in Chapters 4 to 6, the variants of these
variables differ from each other in terms of their morpho-phonological encoding
which primarily affects the realisation of the negative auxiliary as well as the fusion
of adjacent morphemes.8 This derivation-based definition of discourse-pragmatic
variables makes it possible to: delimit the analysis to a finite set of variants that are
derived from the same source construction; and conduct an accountable inves-
tigation into patterns of discourse-pragmatic variation and change. Importantly,
because function is not the unifying denominator, this definition makes it possible

7. Terkourafi (2011) argues that her inference-based postulation of pragmatic variables caters
for both layering effects outlined above: variation in form (e.g. marking discourse-new entities)
and variation in function/procedural meaning (e.g. quotative and intensifier uses of all). As
stated above, Terkourafi (2011: 363) defines quotatives as hearer instructions to “interpret the
speech that follows as attributed to a speaker other than the speaker’s current self;” and intensi-
fiers as instructions to “interpret the property denoted by the head as one that the speaker wishes
to foreground or reinforce.” With quotative and intensifier meanings thus defined, it is not clear
to me how an inference-based postulation of pragmatic variables can unite both the quotative
and intensifier use of the form all under the umbrella of a single variable. All signals different
procedural meanings and alternates with a different set of co-variants depending on its role as a
quotative or intensifier. Unless quotative and intensifier uses of all were argued to be united by
virtue of signalling procedural rather than conceptual meaning, which would not in fact suc-
cessfully delimit the variable context to the form all, I cannot see how Terkourafi’s postulation
of pragmatic variables can be applied to quantifying variation in function.
8. The derivation-based conceptualisation of discourse-pragmatic variables has also been
adopted in recent studies of general extenders. Recognising that function is not a diachroni-
cally stable denominator and that general extenders do not in fact constitute a unitary functional
category, as originally postulated by Dines (1980), Pichler & Levey (2011: 448) define general
extenders “as semi-fixed constructions which prototypically share the following schematic pat-
tern: (connector) (modifier) (generic noun/pro-form) (similative) (deictic).”
32 The Structure of Discourse-Pragmatic Variation

to operationalise function as an independent variable and test its effect on observed

patterns of formal variability.9

2.3.3 Quantitative methods

With the variable contexts thus defined, all variants of the selected variables are
isolated to determine the conditioning of variability in the data and identify any
ongoing changes in the distribution of variants. For each of the variables ana-
lysed, the frequency of individual variants is reported as the proportion of the
total number of instantiations of the variable. The overall distribution patterns to
emerge from these tabulations establish the proportional composition and robust-
ness of the variable systems investigated in Chapters 4 to 6. They make it possible to
situate the results obtained for individual variables in relation to previous research,
and allow us to compare the relative frequency of variants (e.g. localised vs. non-
localised) across variables.
However, overall distributions do not reveal the mechanisms producing the
alternation between competing variants within the variable context. To uncover
these mechanisms, I follow for each of the variables investigated in Part II a set of
standard variationist procedures: operationalising as independent variables, aka
extra- and intra-linguistic factor groups (or predictors), any hypotheses derived
from the literature about variant choice (e.g. gender; preceding subject); coding
each instantiation of the variable for one of the categories, aka factors, which to-
gether constitute the set of possible or reasonable divisions of factor groups (e.g.
male vs. female; person, number and type of preceding subject); and examining
in detail the quantitative distribution of variants across the factors of each fac-
tor group (Bayley 2002; Guy 1993). Because the choice process is generally con-
ditioned by multiple extra- and intra-linguistic variables (“principle of multiple
causes,” Young & Bayley 1996: 253) and because the data under investigation are
unbalanced (see Chapters 4.5.1, 5.5.1 and 6.5.1), multivariate analyses are con-
ducted with Rbrul (Johnson 2009). Like GoldVarb (Sankoff et al. 2005), Rbrul uses

9. Torres Cacoullos & Walker (2009) demonstrate that functional differences between morpho-
syntactic variants can be identified even where the variable context is delimited in terms of
function. They delimited their research area by postulating that will, going to, the Present and the
Present Progressive constitute variants of one variable on the basis that they function to refer to
events or states that occur after speech time. They then investigated the linguistic conditioning
of variants to refine their analysis, and established that the variation between will and going to
is not determined by each variant having an invariant semantic reading. Rather, each variant
occupies different lexical, syntactic and pragmatic niches. Future research will establish whether
this approach can be extended to the analysis of discourse-pragmatic variables.
 Chapter 2.  Data, methodology and theoretical framework 33

a stepwise multiple-regression procedure to assess the contribution of each factor

group included in the analysis to the observed variation when all factor groups are
considered simultaneously. However, unlike GoldVarb, Rbrul incorporates mixed-
effects modelling which takes into account inter-speaker and intra-group variation.
It is thus able to account for the fact that individual speakers may contribute dif-
ferent amounts of data, and that they may favour or disfavour individual variants
as well as particular factor levels to a greater or lesser degree than a fixed model
would predict. This reduces the risk that a single speaker will bias the results for
external and internal predictors, and thus produces more accurate results (Baayen
2008: Chapter 7).
The multivariate analyses in Chapters 4 to 6 yield a number of different results
that help us interpret the data. Firstly, they give the input value which indicates
the overall probability of variant occurrence regardless of contextual constraints.
Secondly, they give log-odds, i.e., raw co-efficients which range from negative
infinity to positive infinity: the larger the number, the bigger the effect size; the
smaller the number, the more moderate the effect size. To facilitate comparison
with results obtained from analyses with GoldVarb, Rbrul converts log-odds to
centred factor weights, with values between 0 and 1. Like log-odds, they indicate
the likelihood of variant occurrence with each factor of a factor group. Values
>.5 signal favouring effects; values <.5 signal disfavouring effects. To make the
results accessible to a wide audience, this study reports both log-odds and factor
weights when reporting the outcome of multivariate analyses. In addition to pro-
viding input values and log-odds/factor weights, the multivariate analyses provide
three lines of evidence that help us uncover the structure of variability in the data
(Poplack & Tagliamonte 2001: 93–94): (1) statistical significance at the 0.05 level,
which reveals which of the factor groups included in a data run make a significant
contribution to the choice of a variant (non-significance of factor groups is indi-
cated by enclosing their log-odds and factor weights in square brackets); (2) the
range, or magnitude of effect, which indicates the relative strength of factor groups
included in a run and situates factor groups with respect to each other (the greater
the range, the greater the effect of the factor group on variant choice); and (3) the
constraint hierarchy, which is the ordering of factors within a factor group and
represents the grammar underlying the surface variability. Comparison of the rela-
tive strength of identical factor groups and constraint rankings of individual factors
across the variables analysed in Part II makes possible comparison of their under-
lying grammars. Where variables occur with sufficient token numbers, apparent-
time analyses are conducted of the data. Comparison of the significance, strength
and especially the ordering of constraints across the age groups represented in the
corpus makes it possible to identify the existence and direction of any ongoing
changes to the structure of variation.
34 The Structure of Discourse-Pragmatic Variation

The fact that the discourse-pragmatic variables analysed in Part II do not

occur with very high frequency may affect the robustness of the results obtained.
Statistical significance of weaker conditioning effects is less easily achieved with
small samples because they tend to involve greater levels of random fluctuation in
the distribution of variants than larger samples (Young & Bayley 1996: 258–259).
Following Poplack & Tagliamonte (2001: 93), the constraint hierarchies yielded
through variable rule analysis are therefore construed as a window on the under-
lying structure of variability even when factor groups are not selected as statisti-
cally significant. To enable readers to evaluate for themselves the robustness of the
results, all graphs and tables in Part II provide the raw token numbers on which
the results are based. Where tokens occur in quantities insufficient to permit
variable rule analysis, i.e., lower than ten to 30 (Guy 1980: 20), no statistical tests
are conducted and results are reported on the basis of percentages and raw token
numbers only. While such results can only yield tentative hypotheses about the
conditioning of linguistic variability, the discussion in Chapter 7 demonstrates
that they nevertheless have important implications for current theories of lan-
guage variation and change.
Despite their acknowledged limitations (see Section 2.3.2), normalised fre-
quency tabulations are generated to supplement the quantitative analyses de-
scribed above and establish the inter- and intra-group robustness of the selected
variables in the BwE data. Normalised frequencies are gauged by adopting the
approach propagated by Macaulay (2005): the raw number of occurrences of a
discourse-pragmatic feature is divided by the total number of words produced
by a speaker or social group; this figure is then multiplied by 10,000 to provide a
normalising measure for comparing frequencies.10 I acknowledge that the overall
frequency of i don’t think and neg-tags could be reported as proportions of
the total number of clauses. Unlike with i don’t know, the syntactic positioning
of these variants is fairly fixed and predictable. I refrain from taking this approach
for two reasons. Firstly, because speech is replete with dysfluencies, discerning
what constitutes a clause is a complex and potentially subjective task. Secondly,
by consistently quantifying the frequency of all discourse-pragmatic variables per
10,000 words, scholars are in a much better position to draw reliable comparisons
of their frequencies across individuals, social groups and varieties.

10. As stated earlier, false starts, truncations, minimal response particles and filled pauses are
included in the word counts. Readers are referred to Pichler (2010: 594) for a justification of this
approach.
 Chapter 2.  Data, methodology and theoretical framework 35

2.3.4 Summary

This study conceptualises the discourse-pragmatic features chosen for analysis

(i don’t know, i don’t think, neg-tags) as linguistic variables on the basis that
their variants are derived from the same source construction. This allows inclusion
of function as a factor group in the multiple regression analyses conducted with
Rbrul. Rbrul tests the combined effect of multiple predictors on variant choice and
includes speaker as a random effect in the analysis. It thus provides a reliable mir-
ror on the grammar underlying observed patterns of formal variability in the use
of the targeted variables. Viewed in apparent time, the output of the variable rule
analyses helps uncover any ongoing changes in the use of the selected variables.

2.4 Grammaticalisation

Synchronic language structures arise from historical developments (Bybee

2010: Chapter 6). Consequently, synchronic patterns of variation in the morpho-
phonological encoding, morpho-syntactic distribution and functional versatility
of i don’t know, i don’t think and neg-tags may find a natural explanation in
grammaticalisation, a diachronic process of change which generally affects multiple
levels of the linguistic system (phonetics, morpho-syntax, semantics-pragmatics).
This section introduces the set of changes that constitute grammaticalisation, and
explores the applicability of the grammaticalisation framework to accounting for
the evolution of discourse-pragmatic features. In addition to briefly addressing the
status of grammaticalisation as a distinct process and coherent theory of language
change, it outlines how the progression or product of grammaticalisation can be
systematically observed in synchronic data.

2.4.1 Mechanisms of change in grammaticalisation

Grammaticalisation has been defined as “the change whereby lexical items and
constructions come in certain linguistic contexts to serve grammatical functions
and, once grammaticalised, continue to develop new grammatical functions”
(Hopper & Traugott 2003: 18), or as “the process by which a frequently used se-
quence of words or morphemes becomes automated as a single processing unit”
(Bybee 2003: 603). In what follows I will unpack these definitions and draw on
illustrative examples from the literature to introduce the effects of grammaticalisa-
tion on language use and structure.
Although the language change processes associated with grammaticalisation
have been widely discussed in the eighteenth and nineteenth centuries, Meillet
36 The Structure of Discourse-Pragmatic Variation

(1912) is generally credited with the first use of the term ‘grammaticalisation’.
Meillet conceived of grammaticalisation in two ways: (a) the change whereby free
or pragmatically motivated word orders (e.g. in Latin) develop into fixed or more
restricted word orders (e.g. in French); (b) the change whereby independent lexi-
cal items (e.g. nouns, verbs, adjectives) develop into grammatical function words
(e.g. prepositions, adverbs, auxiliaries) which may then grammaticalise further
into affixes. The evolution of grammatical elements from major lexical to minor
grammatical categories is sketched by Hopper & Traugott (2003: 7) as a “cline of
grammaticality” of the following type:
(8) content item > grammatical item > clitic > inflectional affix

From a diachronic point of view, a cline such as (8) depicts a path along which lexi-
cal items or content words progress as they develop functional roles. The develop-
ments are generally understood to be: (i) continuous: the labels on the cline refer
to cluster and focal points along the trajectory of change; (ii) unidirectional: items
tend to shift from the lexical to the grammatical end of the cline but not generally
vice versa (Hopper & Traugott 2003: 99–139); (iii) gradual: innovative uses co-
exist alongside older uses during intermediate stages of development (see Hopper’s
[1991] notion of layering).11 The developmental clines depicted in (8)–(10) are
also called “grammaticalisation chains” (Heine 1993) and “paths” or “pathways”
of grammaticalisation (Brinton 2006). From a synchronic point of view, they can
be understood as continua: fuller, lexical items and constructions cluster towards
the left end of a continuum and reduced, grammatical items and constructions
towards its right end (Hopper & Traugott 2003: 6).
Initially, the focus of grammaticalisation studies was largely on morpho-
syntactic­changes (see in particular Lehmann 1995 [1982]). With Traugott (1982),
the focus of interest shifted to semantic-pragmatic aspects of grammaticalisation.
Traugott argued that pragmatic meanings develop from referential (or proposi-
tional) meanings, a hypothesis that she conceptualised as the unidirectional cline
in (9). More recently, Traugott & Dasher (2002) have postulated a more complex
unidirectional model of semantic-pragmatic change whereby linguistic material
becomes increasingly pragmatic, procedural and meta-textual as it progresses
along the multiple developmental paths depicted in (10).

11. Kaltenböck et al. (2011) propose that what they call ‘theticals’ arise at first through instan-
taneous co-optation of units taken from sentence grammar and only later undergo gradual
formulaicisation. This view is not supported by variationist studies that have tracked the pro-
gression of grammaticalising discourse variables in apparent time (see, for example, Romaine
& Lange 1991).
 Chapter 2.  Data, methodology and theoretical framework 37

(9) propositional (> textual) > expressive12

 (Traugott 1982: 256)
(10) truth-conditional > non-truth-conditional
content > content/procedural > procedural
nonsubjective > subjective > intersubjective
 (Traugott & Dasher 2002: 40)

As implied above, the grammaticalisation of linguistic material involves a com-

plex set of interrelated mechanisms which affect all levels of the linguistic system.
According to Heine (2003: 579), these mechanisms include: desemanticisation,
extension, decategorialisation and erosion. Desemanticisation, or semantic
bleaching, refers to the loss of referential content. Typically, bleaching tends to
occur only in later stages of grammaticalisation as items develop more pragmatic
meanings to replace their original content meaning (Traugott & König 1991).
Extension, or generalisation, obtains when items, as a result of the conventionali-
sation of conversational inferences and the associated increase in their polysem-
ies, progressively come to be used in new contexts where they were not used
before. Decategorialisation refers to the process whereby a lexical or grammatical
form loses the morpho-syntactic properties that are characteristic of its source
category. Finally, erosion, or phonological attrition, refers to the reduction in
the phonological representation of a form and/or fusion of adjacent words/mor-
phemes. It is caused by the fact that as their inferences conventionalise and their
uses generalise, frequently occurring items and constructions gradually weaken
their prosodic emphasis and their referential contribution to utterance meaning
(Bybee 2003, 2006).
Bybee (2003, 2006) argues that the changes outlined above are driven by the
high token frequency of items and constructions undergoing grammaticalisation.
With increased frequency and repetition, co-occurring morphemes or lexemes
lose their internal structure and compositionality, and come to be stored and proc-
essed as single units. As a result, they grow autonomous from their composite
morphemes and their non-grammaticalised sources. However, frequency is not
just a trigger but also a result of grammaticalisation. As grammaticalising items
and constructions undergo context expansion and acquire more polysemies, they
dramatically increase their frequency (see the notion of extension above).
The above outline demonstrates that the individual processes constitut-
ing grammaticalisation (desemanticisation, extension, decategorialisation, ero-
sion) are closely interrelated by virtue of the fact that they are all triggered by

12. ‘Textual’ here refers to meanings that signal intra-clausal truth-conditional cohesion
(Traugott 2003a: 633).
38 The Structure of Discourse-Pragmatic Variation

frequency and that they interact with each other in regular and predictable ways.
Grammaticalisation scholars draw on these observations to refute criticisms ex-
pressed inter alia by Campbell (2001) and Janda (2001): that grammaticalisation is
not a distinct process of change but an epiphenomenon of changes that elsewhere
operate independently of one another. Haspelmath (2004: 26), for example, rejects
these criticisms by arguing that grammaticalisation “is a macro-level phenomenon
which cannot be reduced to the properties of corresponding micro-level phenom-
ena” because its constituting processes are strongly correlated. Heine (2003: 583)
posits that the different components of grammaticalisation together constitute an
explanatory framework for elucidating the origins and development of grammatical
forms, thereby rebutting suggestions that grammaticalisation cannot be considered
a theory in its own right (see in particular Newmeyer 1998). Heine’s (2003) view is
supported by Bybee (2009: 353) who argues that grammaticalisation is “a diachronic
theory of language.” It describes how grammar arises and develops through lan-
guage use, and furnishes strong hypotheses about synchronic language structure.
Grammaticalisation thus constitutes a useful framework for variationist studies
which explore the factors conditioning variation in synchronic language use.

2.4.2 Grammaticalisation and discourse-pragmatic features

A growing number of studies demonstrate that many of the processes defining

grammaticalisation account for the diachronic development and synchronic prop-
erties not just of morpho-syntactic but also of discourse-pragmatic variables (see
inter alia Brinton 1996, 2006, 2008 for diachronic studies; Andersen 2001; D’Arcy
2005; Ferrara & Bell 1995; Tagliamonte & Hudson 1999; Thompson & Mulac 1991b
for synchronic studies). Close inspection of the relevant literature reveals that
grammaticalising discourse-pragmatic features commonly undergo the following
morpho-syntactic changes:
– decategorialisation, i.e., the loss of morpho-syntactic properties characteristic
of the original form (see Section 2.4.1): for example, the inability of I mean
to be modified by adverbials or to take phrasal and clausal complements
(Brinton 2007);
– change from major (open) to minor (closed) word class (Lehmann 1995
[1982]): for example, the change of I think from a complement-taking matrix-
clause to an epistemic adverb (Thompson & Mulac 1991a);
– freezing or ossification of the feature in one grammatical construction or
form (Lehmann 1995 [1982]): for example, its fossilisation in the imperative
when say occurs before numerical expressions to mean ‘about, approximately’
(Brinton 2005).
 Chapter 2.  Data, methodology and theoretical framework 39

In addition, there is evidence that grammaticalising discourse-pragmatic features

commonly undergo the following semantic-pragmatic changes:
– desemanticisation or semantic bleaching, whereby items lose their full mean-
ing and assume less concrete meanings (see Section 2.4.1): for example, their
loss of concrete visual perception meaning when you see and see come to ex-
press speakers’ subjective emotion (Brinton 2008: Chapter 6);
– shift from referential to non-referential meaning (see clines (9) and (10)
above): for example, the use of well as part of an agreement signal (Macaulay
2005: 60);
– conventionalisation of conversational implicatures (Traugott & König 1991):
for example, the use of the temporal adverb now as an emphatic topic-changer
(Aijmer 2002: Chapter 2);
– subjectification, i.e., the process whereby lexical items increasingly signal
speaker beliefs and attitudes (Traugott 1995, 2003b): for example, the use of
like as a focusing device and quotative introducer (Romaine & Lange 1991).
However, previous studies also show that the evolution of discourse-pragmatic
features may diverge from canonical cases of grammaticalisation, especially with
regard to Lehmann’s (1995 [1982]) parameters of grammaticalisation. Firstly, dis-
course-pragmatic features do not generally undergo coalescence whereby items
lose their independent word status through affixation or morphological fusion
(see, however, Traugott [2003a: 642] on the morphological bonding of in + deed
and any + way). Secondly, instead of being subject to condensation or scope reduc-
tion, the grammaticalisation of discourse-pragmatic features generally involves
a unidirectional expansion of scope (scope within a proposition > scope over a
proposition > scope over discourse). Finally, not all discourse-pragmatic features
manifest fixation. For example, while oh always occurs in turn- or utterance-initial
position (Macaulay 2005: 60), I think has acquired increased syntactic mobility as
a result of its grammaticalisation from a complement-introducing matrix clause to
an epistemic adverbial. It consequently occurs not just utterance-initially but also
utterance-medially and -finally (Thompson & Mulac 1991b).13

13. Traugott (2010) differentiates two models: ‘grammaticalisation as reduction’ and ‘gram-
maticalisation as expansion.’ In the former prototype model, grammaticalisation is conceived
as involving reduction, freezing and obligatorification and as pertaining to the inflectional en-
coding of tense, aspect, mode, case, number, etc. In the latter extended model, grammaticalisa-
tion is conceived as involving structural expansion of various kinds (see Himmelmann 2004)
and as pertaining to discourse-pragmatic features. According to Traugott, the two approaches
are complementary because they address different questions, i.e., the development of morpho-
syntactic forms through grammaticalisation vs. the types of functional changes involved in
grammaticalisation.
40 The Structure of Discourse-Pragmatic Variation

Although some evolutionary differences are to be expected between features

that develop differential grammatical roles (Traugott 2003a: 643), the fact that
discourse-pragmatic features do not unequivocally fulfil all of Lehmann’s (1995
[1982]) grammaticalisation criteria has led to some disagreement over whether
grammaticalisation is indeed an appropriate theoretical framework for explaining
and elucidating these features’ diachronic evolution and synchronic distribution.
The uncertainty over the status of discourse-pragmatic features in a theory of
grammaticalisation has been augmented by claims that the diachronic processes
observed in their evolution yield pragmatic rather than grammatical items and
constructions which have not traditionally been considered part of grammar (see
Chapter 1.3). Despite the implicit suggestion in Meillet (1912) to apply the term
‘grammaticalisation’ broadly (see Section 2.4.1), some scholars propose that the
emergence of discourse-pragmatic features be conceptualised as constituting a sep-
arate process from grammaticalisation: pragmaticalisation (Aijmer 1997; Erman &
Kotsinas 1993). Pragmaticalisation separates the domains of grammar and prag-
matics, and accommodates in its definition the linguistic developments specific to
the emergence of pragmatic features such as lack of bonding, scope expansion and
syntactic mobility (see above). However, the proposal to distinguish the emergence
of pragmatic elements from that of grammatical elements has not been universally
accepted (see in particular the papers in Degand & Simon-Vandenbergen 2011).
Several arguments can be identified for its almost unanimous rejection. First, it
must be acknowledged that Meillet’s (1912) conception of grammaticalisation in-
cludes the syntacticisation of word order (see Section 2.4.1 above), which is rele-
vant to the fixing and freezing of multi-unit discourse-pragmatic variables. Second,
the outline at the beginning of this section demonstrates that the emergence of
discourse-pragmatic and morpho-syntactic features shares far more similarities
than differences, leading Günthner (2000: 439) to argue that there is in fact no real
difference between pragmaticalisation (the emergence of pragmatic functions)
and grammaticalisation (the emergence of grammatical functions). Finally, some
scholars have advocated a comprehensive notion of grammar which encompasses
pragmatic meanings and recognises the status of discourse-pragmatic features as
legitimate and integral elements of grammar (see Chapter 1.3). In light of these
arguments, I adopt a broad conception of grammar inclusive of pragmatic mean-
ings, and frame my account of the diachronic evolution and synchronic properties
of discourse-pragmatic features in terms of grammaticalisation.
 Chapter 2.  Data, methodology and theoretical framework 41

2.4.3 Grammaticalisation in synchronic data

Grammaticalisation is a diachronic process of change. Nonetheless, it is highly

relevant to studying discourse-pragmatic variation in synchronic data. For a
start, grammaticalisation studies provide testable hypotheses about the nature
and trajectory of changes giving rise to synchronic language structure which can
be operationalised as independent linguistic variables conditioning synchronic
variation (Poplack 2011; Torres Cacoullos 2011). Moreover, because grammati-
calisation typically occurs as a series of micro-steps, it can be observed in fluid
patterns of synchronic language variation (Hopper & Traugott 2003; Traugott &
Trousdale 2010). Key to tracking grammaticalisation in synchronic data is cor-
relating the form undergoing change with independent linguistic predictors and
comparing the results of independent multivariate analyses across age groups.
The existence and directionality of any ongoing grammaticalisation changes can
then be gleaned from apparent-time shifts in constraint hierarchies within factor
groups (see Section 2.3.3 above). As shown in Chapters 4 to 6, evidence for the
grammaticalised status of discourse-pragmatic variables can be deduced from the
distribution of variants across intra-linguistic constraints even where no apparent-
time shifts in the distribution of variants are observable or where apparent-time
analyses are not possible due to insufficient data.

2.4.4 Summary

This investigation draws on the framework of grammaticalisation to explore the

synchronic distribution of the selected discourse-pragmatic variables. The frame-
work provides important hypotheses about the distribution of variants in syn-
chronic data, thus guiding the analysis of the data as well as the interpretation of
the results.

2.5 Conversation analysis (CA)

Among the hypotheses generated by the grammaticalisation framework is that

grammaticalising material undergoes phonetic attrition and desemanticisation.
To establish whether formal variation patterns in the use of the targeted variables
(i don’t know, i don’t think, neg-tags) can be correlated with their differential
positioning on a cline or continuum of semantic-pragmatic change, it is necessary
to include function as a factor group in the quantitative analysis and establish for
42 The Structure of Discourse-Pragmatic Variation

each of the targeted variables the factors that constitute this factor group. This
section introduces the key concepts of conversation analysis (CA) which inform
the functional data analysis. It also presents the functional domains in which
discourse-pragmatic variables operate, and describes the measures employed for
reducing the subjective dimension of the analysis and validating its results. First,
however, it provides the rationale for the choice of CA.

2.5.1 Choice of CA

While coding data for independent linguistic-structural and broad extra-linguis-

tic variables is relatively straightforward, there are no universally agreed meth-
ods for coding independent discourse-functional variables. In the literature, the
semantic-pragmatic functions performed by discourse-pragmatic features have
been studied within a wide range of theoretical and methodological frameworks
including, amongst others, coherence theory (e.g. Lenk 1998; Schiffrin 1987), po-
liteness theory (e.g. Algeo 1988, 1990; Holmes 1995) and relevance theory (e.g.
Andersen 2001; Blakemore 1988; Jucker 1993). Studies conducted within these
frameworks have yielded important insights into the role of discourse-pragmatic
features in signalling coherence relations between discourse units, affecting social
rapport between interlocutors, and guiding utterance interpretation. Top-down
approaches such as these have the advantage of providing unified and thus eas-
ily comparable accounts of the functionality of discourse-pragmatic features. Yet,
there is a high risk that the focus on a single theoretical explanation for their use
might yield incomplete descriptions of features’ functional versatility. An induc-
tive bottom-up approach which is not tied to a single theoretical framework and
which is sensitive to features’ strong context-dependency might be better suited to
yielding comprehensive functional taxonomies which reflect the fact that a single
discourse-pragmatic variable is likely to perform multiple and highly diverse func-
tions, sometimes even simultaneously. The present investigation therefore adopts
a conversation analytic approach and examines the chosen discourse-pragmatic
variables in their full interactional, sequential and linguistic contexts of occur-
rence. This will yield comprehensive descriptions of the variables’ functional spec-
tra in the BwE data.

2.5.2 Key concepts of CA

CA is an empirical field of investigation which conceptualises talk as an organised

and ordered social activity, and provides insights into the interactional structure
and sequential organisation of talk-in-interaction (see further Hutchby & Wooffitt
 Chapter 2.  Data, methodology and theoretical framework 43

1998; Schegloff 2007). Chapters 4 to 6 draw on the following principal tenets of

CA to establish broad functional taxonomies for the selected discourse-pragmatic
variables:
– recipient design: Talk-in-interaction is designed in a way that displays speak-
ers’ orientation and sensitivity to their co-participants in interaction (Sacks et
al. 1974: 727).
– preference organisation: In talk-in-interaction alternative but non-equivalent
actions are available to speakers. One action is preferred or expected to be
chosen (e.g. agreement, acceptance); the other is dispreferred (e.g. disagree-
ment, refusal). Preferred next actions are generally performed directly and
without delay (“preferred-action turn shape”); dispreferred next actions are
generally performed indirectly and in a qualified manner, and are gener-
ally delayed between and within turns (“dispreferred-action turn shape”)
(Pomerantz 1984: 64).
– turn-exchange mechanisms14: Turn-exchange is administered by participants
in interaction to proceed smoothly. Usually only one speaker talks at a time,
and transitions are finely co-ordinated to minimise gaps and overlaps. Speaker
exchange is accomplished on a turn-by-turn basis through one of the following
turn-allocation techniques: (i) the current speaker selects a next speaker; (ii) a
next speaker self-selects themselves; or (iii) the current speaker continues to
speak (Sacks et al. 1974).
– topic-proffering sequences: By proffering a topic, speakers make available to
co-conversationalists a particular topic which they expect their co-conversa-
tionalists to embrace or reject. Preferred responses to topic-proffers engen-
der expansion of the topic; dispreferred responses engender topic-closure
(Schegloff 2007: 169–180).
Knowledge of these tenets is important for functional data analysis because they
provide some general indication of the interactional work discourse-pragmatic
features do: conveying speaker and hearer attitudes; determining the preference
status of actions; and signalling turn-exchange as well as topic-closure. In order to
fine-tune the analysis and allocate each token of the targeted variables to specific
functional categories, the qualitative analyses in Part II also exploit the following
key concepts associated with CA:

14. Turns at talk have been defined in a variety of ways (see Edelsky [1993] for a critical appraisal
of the literature). In this book, ‘turn’ refers to an interactional unit whose boundaries are marked
by speaker exchange.
44 The Structure of Discourse-Pragmatic Variation

– sequential implicativeness: Utterances are “context-shaped” and “context-re-

newing” (Heritage 1984: 242), i.e., they occur in response or reaction to some
prior utterance and project a relevant next utterance.
– adjacency pairs: Talk is sequentially organised into adjacency pairs. In their
most basic form, adjacency pairs are (i) composed of two turns; (ii) produced
by different speakers; (iii) adjacently positioned; (iv) ordered, i.e., consisting
of a first pair part which initiates an exchange, and a second pair part which
responds to the prior turn; and (v) pair-type related, i.e., the production of the
first pair part requires the production of a second pair part that is from the
same pair type as the first utterance (e.g. question-answer, offer-acceptance/
refusal, assessment-assessment) (Schegloff & Sacks 1973: 295–296).
– next-turn proof procedure: Because utterances are understood as directed to
prior talk (see above), a current turn’s talk displays a speaker’s analysis and
understanding of the immediately preceding turn’s talk (Heritage & Atkinson
1984: 9–11).
These concepts postulate that the meaning of an utterance is activated by its se-
quential positioning and participation in adjacency pairs. Hence, the data inter-
pretation in Part II is founded on a close reading of the larger and immediate
sequential context in which the variables of interest are used as well as close ex-
amination of the structure and type of adjacency pairs in which they occur. Also,
next-turn talk is carefully examined because it may display how the use of an utter-
ance containing or constituted by a discourse-pragmatic variable was understood
by co-participants in the interaction, thus serving as an important resource for
guiding the allocation of discourse-pragmatic variables to functional categories.
In addition to the tenets and concepts outlined above, the qualitative data
analysis is based on some observations from CA which have not been formulated
as theories but which, due to their interactional importance, impact on utterance
interpretation nevertheless:
– temporal development of interactions: Features such as overlaps and interrup-
tions convey important information about speaker-hearer alignment (Du Bois
et al. 1993: 52).
– false starts, repetitions and filled pauses (em, er, um): False starts and repeti-
tions aid speakers in the planning of discourse (Cameron 2001: 34), and filled
pauses perform a range of pragmatic functions similar to more prototypical
discourse-pragmatic features (Kjellmer 2003; Stenström 1990).
– acknowledgment tokens (mhm, uh-huh, yeah): Depending on their strategic
placement in discourse, these tokens function to signal hearers’ continued
interest and attention to the speaker, or to express agreement and acceptance
(Schegloff 1982).
 Chapter 2.  Data, methodology and theoretical framework 45

– prosodic and paralinguistic features (speech rate, stress, pauses, volume,

duration, pitch movement, intonation contour, voice quality): These fea-
tures contribute to the communicative meaning of utterances (Du Bois et al.
1993: 61–73; Psathas & Anderson 1990: 81–89), and are of great importance
for disambiguating the functions performed by discourse-pragmatic variables
(Wichmann 2011; Yang 2006).
Prosodic and paralinguistic features are dealt with in very broad terms. The dif-
ferentiation between low, high and regular pitch is made impressionistically,
and intonation contours are described in terms of a rough three-way distinc-
tion: continuing, final or rising. (Future work will elucidate the role of prosodic
features in disambiguating the meanings encoded in the variables analysed in
Part II.) In addition to the features listed above, close attention is paid to co-
occurring discourse-pragmatic features and adverbs. While co-occurring elements
may function in isolation of each other with each performing a different func-
tion (Fleischman 1999), more often than not they mutually reinforce each other
(Aijmer 2002: Chapter 2; Fetzer 2011; Stubbe & Holmes 1995: 83). Attention to the
adjacent linguistic context thus constitutes an additional means for differentiating
subtle nuances of meaning.
Finally, the qualitative data analysis draws on Brown & Levinson’s (1987)
politeness model. This model builds on Goffman’s (1967) notion of face, which
Brown & Levinson (1987: 61) define as “the public self-image that every mem-
ber [of a society] wants to claim for himself [sic].” Face consists of two related
aspects: positive face, i.e., members’ desire to be liked and have their actions and
opinions approved of by others; and negative face, i.e., members’ desire to have
their actions unimpeded by others. Because of the mutual vulnerability of face,
participants in interaction routinely employ a series of strategies that function to
redress face-threatening acts, i.e., acts that run counter members’ face wants. As
shown in Part II, discourse-pragmatic features play a central role in mitigating
potential face-threats.

2.5.3 Functional domains

Existing taxonomies of discourse functionality differ in terms of the number and

types of domains or modes they identify. For example, Fischer (2000) and Schiffrin
(1987) identify five domains on which discourse-pragmatic features operate,
Bazzanella (2006) and Erman (2001) only three. This investigation follows Brinton
(1996, 2008) in basing the functional classification of features on Halliday’s (1979)
three functional components of language: ideational, interpersonal, textual. This
46 The Structure of Discourse-Pragmatic Variation

model offers the advantage that scholars can draw as many divisions within the
broad functional components as are necessary for detailed and accurate descrip-
tions of features’ functional versatility. Yet at the same time it makes possible viable
statistical analyses because it allows scholars to collate micro-functions along the
broad components of language use outlined above. With other models, scholars
might end up with too few tokens in each domain to warrant statistical testing, and
collation of individual domains might seem functionally arbitrary.
Some items and constructions that commonly perform discourse-pragmatic
functions may also operate in the ideational mode which is concerned with the
representation of referential content. This is due to the diachronic origin of most
discourse-pragmatic variables in referential material and the functional layering
effect of old and new meanings in synchronic data (see Section 2.4). Where appli-
cable, the ideational mode is included in the functional taxonomies and quantita-
tive analyses in Part II of the book.
The interpersonal mode is concerned with the expression of speaker attitudes
and the coordination of speaker-hearer alignment and relationships. Following
Traugott (2003b), I distinguish in the interpersonal domain subjective from inter-
subjective functions, although the distinction between the two is not always clear-
cut. Variables with a subjective function serve to indicate speakers’ relation and
attitude towards their propositions. They may signal speakers’ subjective stance
towards propositions (e.g. I believe) and/or speakers’ assessment of the reliability
of their propositions (e.g. possibly, I’m convinced that, Absolutely!). Variables with
an intersubjective function make explicit speakers’ attention to and awareness of
their interlocutors. They may serve to request confirmation of assumptions pre-
sumed to be shared by speakers and hearers (e.g. isn’t that so?) as well as to signal
shared assumptions (e.g. Indeed!). Also, intersubjective tokens may contribute to
creating a mitigating effect in conversation by reducing the anticipated unwelcome
effect or illocutionary force of potential face-threats (Fraser 1980; Holmes 1984b).
Although discourse-pragmatic features with interpersonal functions frequently
serve to redress potential face-threats and to mitigate interactional conflict, polite-
ness concerns are not a prerequisite for their use.
The textual mode relates to the text-structuring function of language.
Discourse-pragmatic features operating in this domain signal the relation between
sequentially arranged discourse units, including propositions, turns and topics
(e.g. and, however, as a result). Textual tokens are also instrumental in the overall
development and organisation of discourse. For example, they function to initi-
ate and close discourse (e.g. listen!, right), to mark topic-shifts (e.g. now, anyway)
or to indicate new and old information (e.g. actually, in fact, as I said before). In
the textual mode, then, discourse-pragmatic features contribute to coherence and
 Chapter 2.  Data, methodology and theoretical framework 47

textuality in discourse. In this investigation, the textual domain is divided into

the sub-domains repair marking, turn-exchange, topic-development, transition
marking.

2.5.4 Quantifying multifunctionality

The model outlined above is used in this investigation to describe the functional
versatility of the selected variables and to prepare the inclusion of function as an
independent variable in the quantitative data analyses. However, endeavours to
allocate tokens to functional categories are complicated by the intrinsic multi-
functionality of discourse-pragmatic features: not only do they perform different
functions in different contexts of use but a single instantiation of a feature can
operate across discourse domains and perform multiple functions simultaneously
(Schiffrin 1987). Holmes (1984a) argues that researchers can differentiate on the
basis of the contextual cues discussed in Section 2.5.2 the primary from the sec-
ondary function of multifunctional discourse-pragmatic features and quantify
functionality accordingly. This approach may be useful if the aim of the analysis
is to explore the functional range of variables and to probe the overall salience of
these functions in a given corpus (see Andersen 2001). Yet despite its popular-
ity, this procedure is problematic. Not only does it add an unnecessary layer of
arbitrariness and subjectivity to the analysis but it also yields misleading results.
Conversationalists may exploit the multifunctionality of discourse-pragmatic­
features at strategic points in interaction (Coates 1987: 130), and it is features’
multifunctional, not their unifunctional, uses that are unmarked (Cameron et al.
1988: 77). Holmes’s (1984a) approach fails to reflect this.
The method adopted in the present investigation addresses these shortcom-
ings by incorporating multifunctionality as a parameter in the analysis (see Pichler
2010: 597–598). When tokens perform multiple functions simultaneously (e.g.
initiating or terminating a turn whilst also qualifying its content), these tokens
are categorised as multifunctional tokens performing both functions concurrently.
This approach is preferable to Holmes’s (1984a) since it allows researchers to reflect
in quantitative terms the multifunctional nature of discourse-pragmatic variables.
Also, it is less subjective than Holmes’s method since it does not rely on research-
ers’ intuitive judgments as to which of the multiple functions is more important
in a given context.
48 The Structure of Discourse-Pragmatic Variation

2.5.5 Validating qualitative analyses

In addition to accommodating multifunctionality in the functional taxonomy,

this investigation adopted the following measures to reduce the subjectivity of the
functional data analysis and validate its results. After an initial coding system had
been devised for each variable through the interrogation of existing taxonomies,
the functional categories and the coding of random tokens were cross-checked
with two native speakers of British English. This resulted in some modifications
being made to the initial coding schema. In the case of i don’t know, categories
were also tested with the help of community members who may possess intuitive
knowledge of the prototypical meanings and functions associated with discourse-
pragmatic variables which are not accessible to an outsider (Fox Tree 2007). As
part of a school project, sixth-formers at Berwick Community High School were
introduced to the variable i don’t know and the functions I had identified for
it, and asked to verify the functional categories by questioning their intuitions
about the construction’s use and observing its usage amongst family and friends
over a two-week period. The students’ intuitions and observations confirmed the
accuracy of the initial taxonomic framework. For each variable, three independ-
ent and blind analyses were then conducted at two-month intervals. Tokens that
had not been consistently allocated to the same functional category in all three
passes were discussed with two independent coders. These discussions resulted
in minor modifications to the taxonomy. In the end, the dynamic coding process
established a detailed coding system to capture the function of all tokens of the
selected variables in the data. To ensure the transparency of the analysis and fa-
cilitate its replication, Chapters 4 to 6 discuss illustrative examples of the variables’
functions in some detail and outline precisely which contextual factors gave rise
to the functional interpretations offered.

2.5.6 Summary

This investigation adopts a dynamic, bottom-up approach to functional data analy-

sis. Application of CA theories and methods ensures the empirical discovery of
the full functional range of the selected variables when they are examined in their
interactional, sequential and linguistic context of occurrence. Multifunctionality is
accounted for in the categorisation of tokens, and various methods are employed
for validating the analysis.
 Chapter 2.  Data, methodology and theoretical framework 49

2.6 Conclusion

The methodology developed for this project constitutes a great improvement on

previously used methods. Firstly, it advocates a derivation-based conceptualisation
of discourse-pragmatic variables which caters for the fact that discourse-pragmatic­
features are polysemic elements and that function is not by necessity a stable de-
nominator. Secondly, although the investigation is resolutely variationist, it com-
bines insights from various linguistic models to enhance current understanding of
discourse-pragmatic variation and change. Thirdly, it goes beyond generating over-
all frequency counts and conducts multivariate analyses to target the grammatical
system underlying the choice process between variants. Finally, the methodology
advocated here includes multifunctionality as a parameter in the qualitative and
quantitative data analysis to reflect the interactional and strategic motivations for
the usage of discourse-pragmatic­ features. Together, these measures offer an en-
hanced method for studying discourse-pragmatic variation and change. By cater-
ing for the flexibility, multifunctionality and complexity of discourse-pragmatic
features, they ensure a more accurate account of the mechanisms underlying
discourse-pragmatic variation and change than would be possible by adopting a
unidimensional approach. The full articulation of my innovative methodology in
this chapter will serve as a model for future discourse variation studies.
 chapter 3

The BwE verb negation system

3.1 Introduction

The focus of the present investigation is on exploring patterns of variation and

change in the realisation of three discourse-pragmatic variables: i don’t know, i
don’t think and neg-tags. In addition to operating at the discourse-pragmatic
level and sharing similar developmental properties, the variables are related by
virtue of containing negative auxiliaries. As shown in Example (1) below, negative
auxiliaries are highly variable in the variety of English under scrutiny here.15 In
addition to Standard English variants such as don’t, shouldn’t or doesn’t, the pool
of available variants includes special forms such as divn’t (‘don’t’) in (1a) and (1b),
negative auxiliaries formed with the isolate negator no such as ’m no (‘’m not’) in
(1b), and negative auxiliaries formed with the enclitic negator -nae such as would-
nae (‘wouldn’t’), dinnae (‘don’t’) and cannae (‘can’t’) in (1b)–(1d).
(1) a. Natalie: I divn’t like being called a Berwicker.
HP: No?
Natalie: I don’t like that.
b. Natalie: E:h, I wouldnae feel offended. I just say (.) “Well. I divn’t have
a Geordie accent because (.) I’m no from Newcastle,” you
know
c. Evelyn: Oh! I shouldn’t say it but I dinnae like the Geordie twang.
d. Lori: She says she cannae understand youse at all. She says youse are
hard to understand. But she says she doesn’t know if it’s cos
you’ve got a low voice.

This chapter introduces in more detail the variants available in BwE for effecting
verb negation, with the aim of setting the background for the analyses of the for-
mulaic constructions i don’t know, i don’t think and neg-tags in Part II of the
book, and establishing the robustness and distribution of negative auxiliary vari-
ants in productive constructions, i.e., constructions which have compositionally

15. Unless otherwise stated, examples are taken verbatim from the BwE corpus. All informant
names are pseudonyms; HP is the interviewer. A key to the transcription conventions is provided
on page xvii. A glossary of dialect words is provided on page xix.
52 The Structure of Discourse-Pragmatic Variation

transparent meanings and are assembled on-line from an open category of com-
ponents (Wray & Perkins 2000: 2). The chapter starts with a brief account of the
history of verb negation in English and Scottish varieties in Section 3.2. Section 3.3
introduces the range of negative auxiliary variants found in BwE and discusses their
geographical distribution in the English-speaking world. Before their extra- and
intra-linguistic distribution in BwE is explored in Section 3.5, Section 3.4 reviews
studies investigating their distribution in other UK varieties. Finally, Section 3.6 is
the conclusion to this chapter and the transition to Part II of the book.

3.2 The evolution of verb negation and negative particles

In order to fully appreciate the nature and degree of variation in the form of nega-
tive auxiliaries in BwE, it is necessary to briefly examine the evolution of verb
negation in English. Although scholars disagree on the precise causes and conse-
quences of what is widely referred to as “Jespersen’s Cycle” (see, for example, van
der Auwera 2009; Wallage 2008 for outlines of different accounts), it is generally
assumed that the development of verb negation involves the following stages: the
use of the pre-verbal negative particle ne in Old English, as in (2); the addition
of post-verbal negative particles such as na (‘not a’), nalles (‘not at all’) or noht
(< nawiht, nowiht) (‘not a thing’) in early Middle English, as in (3); the exclusive
use of the negator not (< noht) due to the loss of the pre-verbal negative particle
ne in late Middle English, as in (4).
(2) ic ne secge.  (Jespersen 1917: 9, literally ‘I not say’)
(3) I ne seye not.  (Jespersen 1917: 9, literally ‘I not say not’)
(4) I say not.  (Jespersen 1917: 9)

Jespersen (1917: 9–11) adds as separate stages in the development the introduc-

tion of do-support and placement of the negative marker before the verb in late
Middle English, as exemplified in (5), as well as the subsequent weakening in this
position of not to the clitic -n’t, as exemplified in (6).16 As a result of the develop-
ments sketched out in (2)–(6), modern Standard English has two clause negators:
isolate not and clitic -n’t. In declaratives, they are placed after the auxiliary and
before the predication.
(5) I do not say.  (Jespersen 1917: 11)
(6) I don’t say.  (Jespersen 1917: 11)

16. Following Quirk et al. (1985), small capitals are used throughout this book to refer to the
primary verbs be, have and do as lexical items. Lower case italics are used to refer to their
grammatical forms (e.g. was, is, has, had, did, does) and to modal auxiliaries (e.g. could, would,
will) which are invariable in their form.
 Chapter 3.  The BwE verb negation system 53

In Scottish varieties, negation followed the same cycle of development as in

Standard English, albeit later (Aitken 1979: 88).17 Importantly, in addition to not
and -n’t, the following negative particles were preserved in Scotland: isolate no and
clitic -nae or -na, with the alternative spellings of the clitic representing differences
in the realisation of its vowel ([ne], [nɛ] and [nʌ]). The clitic -na(e) derives from
Old English nā (= ne) (Craigie et al. 1937–2002), which was originally a competi-
tor of not (Jespersen 1917: 17). While it was lost in Standard English early on, it
still prevails in many Scottish varieties as a clitic that is attached to auxiliary verbs
(Macafee 1992). Isolate no is a reduced form of Old English noht (Craigie et al.
1937–2002) and, like clitic -na(e), is used in Scottish varieties to this day alongside
the Standard English particles not and -n’t (Macafee 1992).18
These developments are highly relevant to the present study. As pointed out
in Chapter 2.2, Berwick upon Tweed shares close historical, demographic and
linguistic links with neighbouring areas in Northumberland as well as south-east
Scotland. These links are reflected in the BwE verb negation system.

3.3 Negative auxiliaries in BwE

3.3.1 Negative auxiliary variants in BwE

As in other varieties of English, we find in BwE verb forms that are negated with
the negative markers not and -n’t, as shown in (7) and (8).
(7) It’s not a community like it used to be, Spittal.
He’ll not use very much slang.
We’ve not got any of these high-rise blocks, which is nice.
I might not like the accent but at least it identifies who I am.

17. Modern Scottish speech is commonly described as a continuum with two extreme poles:
(Broad) Scots on one end and (Scottish) Standard English on the other end (Aitken 1984: 519–527;
Corbett et al. 2003: 1–3). The relationship between the two poles of the continuum is far from
clear-cut and it is beyond the constraints of this book to explore it. To avoid implying that certain
linguistic features are situated more towards one end of the continuum or the other, I will through-
out this book refer to modern Scottish speech as ‘Scottish varieties’ or ‘varieties of Scotland.’ All I
mean to imply when I talk about features being typical of Scottish varieties or varieties of Scotland
is that they are commonly associated with Scotland as a geographical area. (There are, of course,
non-Germanic language varieties spoken in Scotland which are not subsumed under my labels.)
18. There are regional differences in the form of the localised negative particles in Scotland.
While isolate no and clitic -nae prevail in southern and western Scotland, their counterparts in
north-eastern Scotland are isolate nae and clitic -na (Macafee 1992; Millar 2007: 76–77; Smith
2000: 239). In the northern northern varieties, a mixed pattern applies (Millar 2007: 132).
54 The Structure of Discourse-Pragmatic Variation

(8) Berwick people aren’t impressed very easily.

They won’t include you in the Borders.
They haven’t got a clue what I’m talking about.
I don’t like football.
We wouldn’t accept it being Scottish.

In addition, the BwE corpus contains verb forms that are negated with the in-
dependent negative particle no and the dependent negative particle -na(e). The
variable pronunciation of -na(e) (see Section 3.2 above) will not be pursued here,
and the particle will henceforth be represented orthographically as -nae. The use
of no and -nae is illustrated in (9) and (10).
(9) I’m no a Scots, you know.
You’ll no get a much better place, you know.
I’ve no finished mine.
(10) He isnae daft.
They cannae even win now.
We’re a relaxed place to be but I wouldnae say we’re a fashionable place to be.

When do and will are negated with the clitic -nae, they are phonetically modified,
yielding the forms dinnae [dənə], [dənɪ] or [dɐnɪ] for ‘don’t’ and winnae [wʊ̃nɪ]
for ‘won’t’, as shown in (11).
(11) a. The people in Berwick dinnae like it when it’s busy.
b. I’m sure you winnae.

In addition to the variants exemplified above, the BwE corpus contains two nega-
tive auxiliary variants that are formed with the negative particles -n’t and not but
which are not generally considered to be part of the repertoire of Standard English.
These forms are divn’t [dɪvǝnt] for ‘don’t’ and cannit [ˈkhanǝt] for ‘can’t.’ In divn’t, a
kit-vowel is preceded by a voiced alveolar plosive and followed by a labiodental
fricative and the negative particle -n’t. The unrounding and fronting of the vowel
in do may have arisen from Northern fronting in Early Middle English (Britton
2002). The form cannit differs from Standard English cannot in that the negative
in cannit is not stressed. In contrast to cannot, cannit can therefore occur in un-
emphatic contexts. The use of divn’t and cannit is illustrated in (12).
(12) a. At the same time you divn’t want the dialect and things to die out.
b. They cannit understand you.

In sum, the BwE data manifest a high degree of variability in the formal encoding
of negative auxiliaries which reflects the historical development of verb negation
in UK varieties as well as Berwick upon Tweed’s location on the Scottish-English
border. Below I will provide details about the geographical distribution of the
variants introduced above.
 Chapter 3.  The BwE verb negation system 55

3.3.2 Geographical distribution and categorisation

of negative particle/negative auxiliary variants

The negative auxiliary variants introduced above differ in terms of their geographi-
cal distribution in England and the English-speaking world. Following Milroy
et al. (1994), geographical distribution is used in this investigation as a basis for
categorising variants.
As Standard English forms, the particles -n’t and not are not confined to
a specific locale but are widely used across England and the English-speaking
world. They are therefore categorised as non-localised or non-localisable vari-
ants. Variants that are used in many varieties of English across the British Isles
but are not generally associated with Standard English usage, such as the tag
form innit discussed in Chapter 6, are subsumed under the category of supra-
local variants.
The usage of the negative particles -nae and no is far more limited than that
of -n’t and not. According to the historical and contemporary literature, -nae and
no are characteristic of Scottish varieties (Anderwald 2002: 53–58; Beal 1997: 370–
371; Grant & Dixon 1921: 115–120; Grant & Murison 1931–1976; Miller & Brown
1982: 3–14; Miller 2003: 87–89; Murray 1873: 216–217; Wilson 1915: 126–127;
Wright 1902). However, restricted use of -nae has also been reported for Ulster
(Anderwald 2002: 50–52) and for the three northern-most counties of England:
Cumbria, Northumberland and Durham (Anderwald 2002: 61; Glauser 1974;
Orton & Halliday 1963; Upton et al. 1994; Wright 1902). In addition, isolated
instances of the independent negator no have been reported for Northumberland
(Orton & Halliday 1963). While the use of -nae and no is widespread in Scotland,
their use in England is localised to varieties spoken close to the Scottish-English
border. For this reason and because their sociolinguistic distribution in BwE is not
identical to that in Scottish varieties (see Section 3.5 below), the negators -nae and
no and auxiliaries negated with -nae and no are here and throughout categorised
as localised variants. Yet due to their strong association with Scotland, these vari-
ants and their north-eastern Scottish equivalents (see footnote 18) are occasionally
referred to as Scottish negators in Section 3.4.
Divn’t and cannit are well-established features of northern/north-eastern
England. Although searches of the Scottish Corpus of Text and Speech (Douglas
2003) yield isolated instances of divn’t in negative polarity question tags in north-
east Scotland, the historical and contemporary literature generally describes divn’t
as a characteristic feature of Tyneside/Northumberland and, to a lesser degree,
Durham and Cumbria speech (Beal 1993: 192–193, 2004: 124; Beal & Corrigan
2005; Orton & Halliday 1963; Upton et al. 1994; Wright 1902). The use of can-
nit has so far been reported only for Tyneside/Northumberland (Beal 1993: 199,
2004: 123; Beal & Corrigan 2005). Seen within the context of England, then, divn’t
56 The Structure of Discourse-Pragmatic Variation

and cannit are localised to its most northern parts. They are therefore categorised
here and throughout as localised variants.
Table 3.1 summarises the BwE inventory of negative particle and negative aux-
iliary variants in productive constructions, with an indication of their geographical
spread in England.19

Table 3.1  Inventory of negative particle and negative auxiliary variants in BwE
non-localised localised localised
negative particles negative particles negative auxiliaries
-n’t -nae divn’t
not no cannit

3.4 Previous research on negator and negative auxiliary variation

The extra- and intra-linguistic distribution of -nae, no, divn’t and cannit has re-
ceived only limited attention in the scholarly literature. To prepare the quantitative
analysis of BwE negative auxiliary variation in Section 3.5, this section reviews the
few extant studies of negative auxiliary variation in Scottish, Northumbrian and
Ulster varieties. It focuses first on accounts of their external distribution before
moving on to accounts of their internal distribution in these varieties.

3.4.1 Extra-linguistic distribution of variants

Based largely on their intuitions as native speakers of Edinburgh speech as well

as data gathered through elicitation tasks and informal observations, Brown &
Millar (1980: 111) and Miller (1993: 114–115) propose that the use of -nae for -n’t
in Edinburgh is stigmatised and avoided by educated speakers in formal contexts.
Macaulay (1991: 50–57) provides empirical evidence for the social stratification of
different negators in his study of socially differentiated language use in Ayrshire,
south-west Scotland. He found that the middle-class speakers in his sample never
used the variants -nae and no. By contrast, the lower-class speakers in his corpus
generally preferred the use of -nae over -n’t and displayed rates of no as high as
93%. Smith et al. (forthc.) report that 96% of all negative auxiliaries in adult-to-
adult conversation in their data from Buckie, north-east Scotland, occurred with

19. The formulaic constructions i don’t know, i don’t think and neg-tags allow additional
variants. These are introduced in Chapters 4 to 6.
 Chapter 3.  The BwE verb negation system 57

the Scottish negators. In interactions between caregivers and their children in the
same community, the rate of use of Scottish negators among caregivers dropped
to 72%; children demonstrated a rate of Scottish negators of 47%, with older chil-
dren displaying higher rates than younger children. Finally, Anderwald (2002: 47)
reports that in Ulster, -nae has developed social significance as a symbol of Ulster
Scots Protestant identity. In sum, these patterns suggest that the use of Scottish
negators is relatively robust in Scottish varieties, and that their use in Scotland and
Northern Ireland is constrained by the following social factors: socio-economic
class, age, geography, ethnicity as well as identity or age of the interlocutor.
Little information is provided in the literature about the relative frequency and
social distribution of the variants divn’t and cannit. In Beal & Corrigan’s (2005)
small-scale study of negation in Tyneside, the use of the localised variant cannit
was found to be marginal compared to that of the non-localised variant can’t (20%
and 80% respectively). No information is provided about the relative frequency
of don’t and divn’t. The majority of tokens of cannit and all tokens of divn’t were
produced by three working-class men with minimal schooling and a negative at-
titude towards education. Beal & Corrigan (2005: 146–147) acknowledge that these
speakers’ use of divn’t and cannit might be due to their having been less inhibited
than other speakers by the recording situation. However, they also propose that
these speakers may have been less influenced than other speakers by the prescrip-
tive stigmatisation of divn’t. While this might explain the non-occurrence of divn’t
amongst other speakers in their data, it does not explain why cannit, which ac-
cording to Beal & Corrigan (2005: 148) is not overtly stigmatised, was used almost
exclusively by working-class men. In sum, these findings indicate that the use of
divn’t and cannit in Tyneside is constrained by the following social factors: speaker
sex and socio-economic class.

3.4.2 Intra-linguistic distribution of variants

In their description of auxiliary verbs in Edinburgh speech, Brown & Millar

(1980: 102–120) point out that in declaratives the isolate negative particle no can
only ever occur with cliticised auxiliaries (e.g. forms of be), but never with non-
cliticised auxiliaries such as do. The particle -nae can cliticise onto all primary
verbs and onto any modal auxiliary except for might. With the modal auxiliary
can, the negative particle -nae is obligatory if it is the modal itself that is negated;
if it is the following verb that is negated, no is used instead. Unlike -n’t, -nae never
inverts over the subject in interrogatives and tags; no, not or -n’t are used instead.
Irrespective of the auxiliary involved, it is not possible for both the auxiliary and
the negative particle to undergo cliticisation. Similar patterns are reported in Miller
58 The Structure of Discourse-Pragmatic Variation

(2003: 87–88, 2004: 50–51) who points out that -nae is attached to all modal aux-
iliaries and to do. No, he argues, is most frequent with will and forms of be and
have. Isolate particles, i.e., no and not, are in fact the norm not just with will, be
and have but also with interrogatives and question tags. By implication, the use
of -nae is limited to declaratives and imperatives.
These accounts, which are largely based on informal observations, elicitations
and the authors’ native speaker intuitions, suggest that auxiliary type and clause
type affect the distribution of Scottish negators in Scotland. A small number of
quantitative corpus-based studies provide empirical corroboration of these ac-
counts. Steele’s (2003) study of caregiver and child speech in Buckie, north-east
Scotland, reveals that do has a lower rate of Scottish negators than other auxilia-
ries (see also Smith et al. forthc.). Also, did and does have higher rates than do of
nae and -na (the Buckie equivalents of BwE no and -nae, see footnote 18 above)
(94%, 89% and 62% respectively). With non-do auxiliaries, forms of present tense
be show the highest and modal auxiliaries the lowest proportion of nae and -na
(92% and 72% respectively). The effect of auxiliary type on the rate of Scottish
negators is further corroborated by Anderwald’s (2002: 50–51) investigation of
verbal negatives in the Northern Ireland Transcribed Corpus of Speech. Anderwald
reports that in Ulster the variation between Scottish and non-localised negators
is restricted to a small number of auxiliaries: no and -nae never occur with am,
are, has, will and only rarely with do, have, could and past-tense be. As regards
clause type, Steele’s (2003) analysis of Buckie reveals that forms of do are more
often negated with Scottish negators in imperatives than declaratives (77% and
46% respectively). With all the other auxiliaries, only declaratives are variable.
Interrogatives are almost categorically formed with Scottish negators, while the
reverse holds for question tags. Similar patterns are evident in Shetland, to the
north-east coast of Scotland, where negative polarity tags are near-categorically
formed with the negative particles -n’t and not (Smith 2009).
The literature review, then, reveals that across the varieties that have been
described or empirically investigated, the occurrence of Scottish negators is con-
strained by clause type and/or auxiliary type. However, comparison of the em-
pirical studies also reveals some cross-variety differences in the frequency and
distribution of negator variants. While Scottish negators occur with all auxiliaries
in the varieties of Buckie and Shetland, they do not occur with do in recordings
from Ayrshire. In Ulster, the variation between Scottish and non-localised negators
is restricted to an even smaller number of auxiliaries. Also, where Scottish nega-
tive particles occur in Ulster, they do so at substantially lower rates than in the
investigated Scottish varieties. In sum, the cross-corpora comparisons suggest that
Scottish negative particles have a different language-internal distribution across
varieties, and that they occur at lower rates outside of Scotland.
 Chapter 3.  The BwE verb negation system 59

3.5 Quantitative analysis of negative particle

and negative auxiliary variation

This section sets out to establish the robustness and distribution of negative particle
and negative auxiliary variants in BwE productive constructions, i.e., in contexts
other than i don’t know, i don’t think and neg-tags. It explores whether social
factors, clause type and type of auxiliary contribute to the distribution of variants
in BwE as they do in other varieties. Knowledge of the constraints on variation in
productive constructions will allow us to contextualise the findings obtained in
Part II of the book and establish the extent to which linguistic context affects the
distribution of negative particle and negative auxiliary variants.

3.5.1 Distributional analysis

Figure 3.1 displays the overall distribution of the negator variants -n’t, not, -nae
and no as well as the localised negative auxiliary variants divn’t and cannit in BwE
productive constructions (N = 3,713). The distribution reveals a clear prevalence of
non-localised over localised variants. Almost three quarters of negative auxiliaries
in productive constructions are formed with -n’t and not. The localised negators
-nae and no combined constitute only one fifth of the data. The occurrence of divn’t
and cannit is not in fact as negligible as suggested in Figure 3.1. Table 3.2 shows
that when their frequency is tabulated in the context of negative do and can only,
divn’t and cannit are far from marginal.
divn’t cannit
6.5% 1%
(N = 241) (N = 30)
no
14%
(N = 528)

-nae
6%
(N = 217)

-n’t
not 62%
10.5% (N = 2307)
(N = 390)

Figure 3.1  Overall distribution of negator and localised negative auxiliary variants
in productive constructions (declaratives, interrogatives and imperatives)
60 The Structure of Discourse-Pragmatic Variation

Initial data runs reveal that speakers do not make full use of the formal variability
available to them for effecting verb negation. Firstly, out of the 36 speakers in the
sample, eight speakers near-categorically employ non-localised negative particle
variants and non-localised negative auxiliary variants (> 95%). Five of these speak-
ers are from the youngest age group; three are from the middle age group. To avoid
obscuring constraints on variation, data produced by these speakers are removed
from further analysis (N = 689). Secondly, variable speakers do not use localised
negators with the full range of auxiliaries. Table 3.2 displays for each auxiliary form
listed in the left-most column the proportion of negator and negative auxiliary
variants listed in the top row of the table. The results show that the use of -nae is
virtually limited to non-third person singular present tense forms of do and the
modal auxiliary can, while the use of no is virtually limited to present tense forms
of be, non-third person singular present tense auxiliary have and the modal will.20
In the BwE corpus, the variation between non-localised and localised negator
variants thus affects only a small number of auxiliaries. Moreover, the patterns
outlined above suggest that the choice between negator variants is closely linked
to contraction and cliticisation strategies: (1) the negator variant -nae is used with
auxiliaries that cannot undergo contraction (do, can); the variant no is used with
auxiliaries that regularly undergo contraction (present tense be, auxiliary have,
will); (2) with auxiliary have and modal will, there is a strong association between
negator cliticisation and the use of the non-localised clitic negator -n’t on the
one hand, and between auxiliary contraction and the use of the localised non-
clitic negator no on the other hand; (3) with present tense be, variation between
non-localised and localised negators is virtually restricted to isolate negators and
contexts of auxiliary contraction.
As implied above, it is not just auxiliary type that exerts a strong conditioning
effect on the occurrence of negator variants but also auxiliary form and function.
The top rows in Table 3.2 show that negative periphrastic do is one of the most
variable auxiliaries in the data. However, the variation only affects do; negative
does and negative did are virtually never realised as divn’t and they only very rarely
occur with no. Conversely, almost half of the instantiations of do amongst variable
speakers are constituted by localised variants: divn’t, dinnae, do no (listed in order

20. Following Quirk et al. (1985: 129–148), I distinguish the following auxiliary functions
for have: main verb uses (e.g. I haven’t a clue) and auxiliary uses (e.g. I haven’t seen it) (see
Huddleston’s [2002: 103] non-core and core uses). The former are henceforth referred to as main
have, the latter as auxiliary have. Auxiliaries, then, refer to a syntactic class of verb forms that
always occur in finite form in a fixed position in a sentence; auxiliary uses refer to the functions
of auxiliaries in a sentence. Included in the latter category are instances of have got (e.g. I’ve
no got nowt to hide) (see Huddleston 2002: 112).
 Chapter 3.  The BwE verb negation system 61

Table 3.2  Distribution of negator variants across auxiliaries (on the left) and distribution
of localised negative auxiliary variants (on the right)
-n’t not -nae no divn’t cannit
 % N  %  N  % N  %  N % N % N TOTAL N

do   52.8 347   0.9 6   9.4 62    1.1 7 35.8 235 – – 657

does   97 164   0.6 1  0 0    0.6 1   1.8    3 – – 169
did   95.4 206   1.9 4  0 0    2.8 6  0    0 – – 216
present be   11   77 28.8 200   0.2 1   60 416 – – – – 694
past be   93.8 106   2.7 3  0 0    3.5 4 – – – – 113
auxiliary have   57   90   5.7 9  0 0   37.3 59 – – – – 158
auxiliary had   81.2   13 12.5 2   6.3 1    0 0 – – – –   16
auxiliary has   57.7   15 34.6 9  0 0    7.7 2 – – – –   26
main have 100   13  0 0  0 0    0 0 – – – –   13
can   31.5   82   0.8 2 55 143    1.9 5 – 10.8 28 260
will   40.5   15   8.1 3   5.4 2   45.9 17 – – – –   37
could   99.2 119  0 0   0.8 1    0 0 – – – – 120
would   98.2 508   0.4 2   1.2 6    0.2 1 – – – – 517
might   14.3    1 42.9 3  0 0   42.9 3 – – – –    7
should   89.5   17  0 0   5.3 1    5.3 1 – – – –   19
may    0    0  0 0  0 0 100 1 – – – –    1
must 100    1  0 0  0 0    0 0 – – – –    1

of frequency). The patterns for be and have confirm the effect of tense and person
on negative auxiliary variation. While present tense forms of be display a high
degree of variation, with the localised variant no constituting the preferred option
amongst variable speakers for effecting present tense be-negation, past tense forms
are virtually categorically negated with the non-localised variants -n’t and not.
Although inferences must remain tentative due to low token numbers, a similar
pattern can be observed with have: all but one token of had are negated with -n’t
and not. Furthermore, have resembles do in terms of the effect of person: have is
far more likely to be negated with localised negators than has. Finally, the pattern
for have reveals the effect of auxiliary function on negator variation. In contrast
to auxiliary uses of have, main verb uses of have occur categorically with the
non-localised negator -n’t. Among the modal auxiliaries, can is the most variable.
Roughly two thirds of all tokens of negative can among variable speakers take a
localised form: cannae, cannit, can no (listed in order of frequency). This is in stark
contrast to the modal auxiliaries could and would which are near-categorically
negated with the non-localised particles -n’t and not. The modal will demonstrates
a tendency to be negated either with non-localised -n’t or localised no. However,
62 The Structure of Discourse-Pragmatic Variation

low token numbers mean that these patterns have to be interpreted with caution.
The same holds for the modals should, might, may and must which occur too in-
frequently in the data to allow any conclusions to be drawn about their variability
in negative contexts. In sum, the context of negative auxiliary variation in BwE is
highly circumscribed by the type, function, tense and person of the auxiliary being
negated. With primary verbs, use of localised variants is virtually restricted to do as
well as auxiliary have and present tense be when they are contracted. With modal
auxiliaries, use of localised variants is virtually limited to can and will.
Table 3.3 compares the distribution of variants across clause type. The re-
sults reveal important parallels with those reported for Scottish varieties (see
Section 3.4.2): the enclitic negator -nae never occurs in interrogatives; the isolate
negator no never occurs in imperatives. No and not evince a propensity to occur
in interrogatives. Together they account for more than 60% of all negative parti-
cles in interrogatives, thus demonstrating that interrogatives are more likely to be
negated with isolate than enclitic negative particles. In contrast to no and not, -nae
and the negative auxiliary variant divn’t evince a propensity to occur in impera-
tives. Considering that negative imperatives are uniformly formed with do, the
results indicate that imperatives are the only clause type that is more likely to be
negated with localised variants (divn’t, dinnae) than with non-localised variants
(don’t) (61.7% vs. 38.3%). Like no and -nae, the occurrence of divn’t and cannit is
also affected by clause type: the former never occurs in interrogatives; the latter
only ever occurs in declaratives. The only negator variant that occurs across the
clause types investigated here is -n’t. In sum, the patterns in Table 3.3 demonstrate
that clause type is implicated in negative auxiliary variation. The use of localised
variants is associated with interrogatives and imperatives.

Table 3.3  Distribution of negator and localised negative auxiliary variants

across clause types
-n’t not -nae no divn’t cannit
% N % N % N % N % N % N TOTAL N
declaratives 59.5 1727   8 232   7.2 208 16.9 492   7.5 218 1 28 2905
interrogatives 38.9    28 16.7   12  0    0 44.4   32  0    0 0  0    72
imperatives 38.3    18   0    0 19.1    9  0    0 42.6   20 0  0    47
 Chapter 3.  The BwE verb negation system 63

3.5.2 Multivariate analysis

Having established which auxiliaries are subject to variation in the BwE data and
how variants are distributed across clause types, this section sets out to establish
the joint contribution of intra- and extra-linguistic predictors to variant choice
among the variable speakers in the sample. The analysis focuses first on the use
of the negator variants -nae and no. It includes other negator variants as non-
application values and is based only on those auxiliaries that have been shown
to be variable in Table 3.2: can, do, present tense be and, with no, auxiliary have.
(Will is not included in the analysis due to low token numbers.) The analysis then
focuses on the variant divn’t. Since its co-variants are formed exclusively with third
person singular present tense do, auxiliary is not included as a factor group in the
analysis for this variant. The only variants included as non-application values in
the run for this variant are therefore: don’t, do not, dinnae and do no. Individual
speaker is included as a random effect in all three multiple regressions in order to
account for any effects of unbalanced data.
Table 3.4 presents the results of the independent multivariate analyses for -nae,
no and divn’t. The em-rule indicates that factors were not included in the analysis
because variants never occur in these environments (see Table 3.3). (More infor-
mation on how to read and interpret the results of multivariate analyses is provided
in Chapter 2.3.3). As evident from the range values, auxiliary type makes the most
important contribution to the occurrence of the non-localised negator variants
-nae and no. The negator -nae is strongly favoured with can and do and strongly
disfavoured with present tense forms of be. Age exerts the second most important
effect on the occurrence of -nae. It is favoured by speakers from the old and middle
age groups and strongly disfavoured by speakers from the young age group, sug-
gesting that its frequency and importance may be waning in apparent time. (See
also the low input probability given at the top of the table which is indicative of the
variant’s marginal status in the BwE negation system). Speaker sex and clause type
were not selected as making a significant contribution to the occurrence of -nae,
even though this negator shows a tendency to occur more often in imperatives
than declaratives. In stark contrast to -nae, which is favoured with auxiliaries that
do not undergo contraction, the isolate negator no is strongly favoured with those
auxiliaries that regularly undergo contraction: auxiliary have and present tense
forms of be. As far as auxiliary is concerned, then, -nae and no are in complemen-
tary distribution. The second most important constraint on the use of no is clause
type. No is favoured in interrogatives and disfavoured in declaratives. Speaker sex
and age do not make a significant contribution to the occurence of no, despite
factor weights indicating apparent-time fluctuations in its use. Finally, the results
Table 3.4  Contribution of external and internal factors to the probability of -nae, no and divn’t
-nae no divn’t
input prob. .025 .133 .365
total N 1563 1723 644
deviance 630.681 1087.787 707.363
factor %  N log odds factor %  N log odds factor % N log odds
weights weights weights
age p = 0.0289 not significant not significant
old .711 16.2   752 −0.900 [.589] 29.8   842 −[0.362] [.455] 32.0 306 [−0.182]
middle .590 15.0   440 −0.364 [.643] 35.3   490 −[0.591] [.487] 33.8 142 [−0.050]
young .220   4.9   371 −1.264 [.278] 16.1   391 [−0.952] [.558] 54.4 196 −[0.232]
range .491
sex not significant not significant not significant
64 The Structure of Discourse-Pragmatic Variation

male [.500] 13.4   949 [−0.002] [.522] 28.5 1026 −[0.089] [.451] 31.7 398 [−0.195]
female [.500] 13.6   614 −[0.002] [.478] 28.0   697 [−0.089] [.549] 44.3 246 −[0.195]
range
clause type not significant p = 0.00216 not significant
declarative [.380] 13.0 1517 [−0.488] .346 27.9 1667 −0.638 [.507] 43.5 598 −[0.028]
interrogative – –  – −– .654 39.3    56 −0.638 – – – –
imperative [.620] 19.6    46 −[0.488] – –  – −– [.493] 36.0   46 [−0.028]
range .308
auxiliary p = 3.96e-109 p = 308e-179
can .977 57.0   251 −3.737 .100   1.9   260 −2.200
do .619   9.6   944 −0.487 .074   1.1   611 −2.528
present be .014   1.0   668 −4.224 .953 59.9   694 −3.001
auxiliary have –  –  – −– .849 37.3   158 −1.727
range .963 .879
speaker random st. dv 1.423 random st. dv 1.323 random st. dv 1.507
 Chapter 3.  The BwE verb negation system 65

for divn’t show that none of the independent variables included in the analysis
significantly constrains its use. The variant occurs with roughly equal probability
across age groups, speaker sex and clause type.

3.5.3 Summary of results

The quantitative analysis of verb negation in BwE productive constructions cor-

roborates the findings obtained in previous studies of negative auxiliary variation
which have shown that the locus of variation is highly circumscribed by the type,
person, tense and function of the auxiliary as well as by the clause type in which
variants occur. Yet despite these broad similarities, the sociolinguistic distribution
of variants is not identical across the investigated varieties. The variation between
localised and non-localised negators is restricted to a smaller number of auxiliaries
in BwE compared to the Scottish varieties discussed in Section 3.4. Also, where
localised negators occur in BwE, they do so with markedly lower rates than in
the Scottish varieties. Moreover, BwE differs from the Scottish varieties discussed
earlier in that the occurrence of -nae is not robust. Yet with the exception of the
effect of speaker age on -nae, social factors are not implicated in the choice of -nae,
no and divn’t amongst those speakers who used the variants during the recordings
made for this study.

3.6 Conclusion

The analysis presented in Section 3.5 has shown that negative auxiliaries in BwE
productive constructions are highly variable in their form and that this variation is
for the most part not random but systematically constrained by language-internal
and, to a much lesser degree, language-external factors. These observations are
highly pertinent to the analysis of discourse-pragmatic variation and change in
Part II of this book. Each of the three discourse-pragmatic variables selected for
analysis contains a negative auxiliary. In i don’t know and i don’t think, nega-
tive periphrastic do occurs between the first person singular pronominal sub-
ject I and the cognitive verb know or think. Variants thus are derived from the
same linear string of components: (first person singular pronominal subject) +
(present tense negative do) + (cognitive verb). However, they differ in the details
of their morpho-phonology, including the realisation of negative do (see further
Chapters 4 and 5). In neg-tags, the negative auxiliary either precedes a pronoun/
existential there, or the pronoun/existential there occurs between the auxiliary
and the isolate negator. Variants thus are derived from the same linear string of
66 The Structure of Discourse-Pragmatic Variation

components: (auxiliary) + (negator) + (pronoun) or (auxiliary) + (pronoun) +

(negator). However, they differ in their realisation of the negative auxiliary and
its adjacent elements (see further Chapter 6). By investigating whether different
negator variants and different variants of negative periphrastic do exhibit compa-
rable frequencies, social indexicalities, linguistic distributions and/or discourse-
pragmatic functionalities across the combined set of productive constructions
investigated above and each of the three formulaic constructions investigated in
Part II of the book, it is possible to establish the extent to which patterns of vari-
ation in the realisation of negative auxiliaries are contoured by linguistic context
and usage constraints.
part ii
 chapter 4

The construction i don’t know*

4.1 Introduction

Corpus linguists draw attention to the fact that i don’t know is the most fre-
quent negative collocation in spoken corpora and as such constitutes one of the
most common epistemic markers in English conversation (Baumgarten & House
2010: 1186; Kärkkäinen 2003: 51; Scheibman 2001: 70–71). A growing number
of studies in the fields of discourse and conversation analysis corroborate these
claims, representing as negligible the role of i don’t know in signalling a cog-
nitive state and emphasising its interactional role within and beyond marking
epistemicity (see further Section 4.2.1 below). In stark contrast to the attention
it has received in qualitative paradigms, in the field of variationist sociolinguis-
tics i don’t know has been largely ignored despite its frequency and highly
variable functional and formal properties. Building on the few extant studies
concerned with variation and change in the use of i don’t know, in particular
Bybee & Scheibman (1999) and Scheibman (2000), this chapter sets out to ad-
dress the current shortage of variationist studies of i don’t know, with the aim
of accounting for the construction’s variable realisation in BwE. The formal vari-
ability is illustrated in the extracts from the corpus in (1). It is depicted here and
throughout by means of variation in orthography, with each orthographic form
(I don’t know, I dono, I dunno, I divn’t knaa, I dinnae ken) representing a different
variant whose phonetic, phonological and morphemic properties are described
in detail in Section 4.3.2.
(1) a. Jerry: I I dono if if the percentage is maybe more one way than the
other, I don’t know.
b. HP: Why not.
Natalie: @ I don’t know? (.) I dunno? I just don’t like being called a Ber-
wicker.

* An earlier analysis of a subsample of the data investigated here was presented in Pichler
(2009). This chapter investigates a larger set of tokens of i don’t know, including those followed
by a phrasal or clausal complement, and offers a more advanced analysis and interpretation of
the data.
70 The Structure of Discourse-Pragmatic Variation

c. HP: Do you sometimes deliberately change the way you speak?

Jerry: (h) I I
Elizabeth: I dunno if em I change (.) deliberately would change the
way I speak. Because (.) (h) like I say, (.) I I don’t know
whether it’s very much of an age thing (.) and that?
d. Luke: I dunno? It’s just something ab- I dunno? It’s just only, I think
only Berwick people can tell. I divn’t knaa what it is. Just
HP: yeah
Luke: = just as if you h-hear someone you say “Ah it’s a Berwick per-
son.” I divn’t knaa how. Don’t have a clue. It’s weird. @
HP: yeah yeah
e. Keith: Well, I dinnae ken. I (.) maybe. (..) I mean, we know who
HP: mhm
Keith: = said it (..) first or the first person I heard (?). I divn’t knaa.
I mean, we know exactly who he’s talking about but I dunno
where it (.) where that is a (.)

Systematic analysis of the construction’s usage and distribution in the socially

stratified BwE dataset provides important new insights into the social, linguis-
tic and interactional mechanisms that give rise to the variation observed in (1).
The analysis uncovers a robust functional split between the most frequent non-
localised variants of i don’t know (I don’t know, I dunno), which is similar to
that reported in Bybee & Scheibman (1999) and is argued to be a product of long-
standing grammaticalisation processes. Moreover, it reveals that social factors
have primacy over linguistic factors in conditioning the occurrence of the most
frequent localised variant (I divn’t knaa) which, unlike the frequently occurring
non-localised variants, does not evince function-specific patterning. Together with
the analyses in other chapters, the analysis presented here addresses the broader
research aims of this project set out in Chapter 1 which include establishing the
structure of discourse-pragmatic variation and demonstrating the important role
of discourse-pragmatic variables in shaping synchronic language variation.
The chapter begins in Section 4.2 with a review of the literature that informs
the current analysis of i don’t know, followed by a more detailed outline of the
aims of this chapter. Section 4.3 describes the decisions regarding the circumscrip-
tion of the variable context and the coding of the data. This section also details the
variants of i don’t know which have been identified in the BwE corpus. By provid-
ing a detailed outline of the variable’s functional repertoire in the present dataset,
Section 4.4 demonstrates how the data were coded for pragmatic function. Section
4.5 presents the results of the quantitative analysis, outlining the synchronic struc-
ture and apparent-time robustness of the formal variation observed in the data. The
results are discussed in Section 4.6. Finally, Section 4.7 is the conclusion.
 Chapter 4.  The construction i don’t know 71

4.2 Previous research on i don’t know

i don’t know has received a fair amount of scholarly attention. This section re-
views first qualitative studies concerned with the construction’s pragmatic func-
tions in discourse before reviewing quantitative studies concerned with patterns of
variation and change in its use. There follows at the end of this section an outline of
how the present study extends the scope of previous research by including multiple
contextual factors in the quantification of all tokens of the variable in a socially
stratified corpus of interview data.

4.2.1 Qualitative studies: Discourse-pragmatic functions of i don’t know21

Discourse and conversation analysts argue that the construction i don’t know
has little importance as a verbal representation of a cognitive mental state, i.e., lack
of knowledge. Instead, analysts who have studied i don’t know in its sequential
and interactional context of occurrence concur in highlighting the construction’s
significance as a pragmatic device that performs multiple interpersonal and textual
functions in discourse.
Potter (2004: 212) describes i don’t know as a “stake inoculation” device: it
serves to avert potential contradictions from interlocutors and to protect speak-
ers’ positive face wants by denying the relevance of immediately following and/or
preceding utterances and reducing the risk that these are interpreted unfavour-
ably (see also Drew 1992). To exemplify this function of i don’t know, Potter
draws on Martin Bashir’s 1995 Panorama interview with Princess Diana. Potter
argues that when the late Princess precedes and follows her assessment of why she
had given consent to Andrew Morton’s biography with ‘dunno,’ the construction
serves to manage a potentially unsympathetic inference of her as “a spurned and
vindictive ex-wife” (2004: 215). Wooffitt’s (2005: 125) analysis supports a descrip-
tion of i don’t know as an interpersonal face-saving device which “attend[s] to
sensitive or delicate matters generated in interaction.” Wooffitt observes that when
youth are asked in interviews whether they belong to a particular subculture (e.g.
punks), they frequently preface their turns with i don’t know in attempts to dis-
tance themselves from subcultural categories and resist self-identification in these
terms. One of the most comprehensive discussions of i don’t know’s face-saving
potential in everyday conversation is provided by Tsui (1991). She demonstrates
that depending on the construction’s placement in discourse, it may function to

21. Because the functional repertoire of i don’t know is illustrated in detail in Section 4.4, no
examples are provided here of the functions identified for it in the literature.
72 The Structure of Discourse-Pragmatic Variation

qualify preceding or upcoming propositions; to soften the effect of disaligning or

impolite remarks; or to withhold sequentially and interactionally dispreferred ac-
tions. Tsui thus draws attention to the fact that the functionality of i don’t know
cannot be reduced to a single pragmatic meaning.
This view is supported by Beach & Metzger’s (1997) analysis of casual con-
versations and courtroom interactions. They argue that in second pair parts of
question-answer adjacency pairs i don’t know is strategically employed by in-
teractants to achieve any of the following actions: to mark upcoming assessments
as uncertain; to avoid requests for action by providing responsive yet noncom-
mittal replies; to dismiss others’ contributions in attempts to complete actions
in progress or terminate problematic topics. They thus draw attention to the fact
that i don’t know functions both in the interpersonal domain to express speaker
stance and in the textual domain to structure discourse. The construction’s textual
functions have also been noted elsewhere. Ford & Thompson (1996: 169–170)
suggest that claims of insufficient knowledge serve as topic-closing moves when
speakers are reluctant to pursue a topic that is not of prime concern or interest to
them. Hutchby (2002) argues that in counselling sessions children strategically
use i don’t know to avoid answering unwelcome questions and resist discussing
problematic topics. According to Hutchby, it is the construction’s potential to be
produced in response to virtually any question, including requests for elaboration
following declarations of insufficient knowledge, which makes it such a powerful
interactional tool. Beyond topic-control, i don’t know can also function in the
textual domain as a turn-yielding and turn-initiating device (Östman 1981: 27;
Schegloff 1996: 61–62).
Importantly, i don’t know can operate concurrently in the interpersonal
and textual domains. Schegloff ’s (1996: 61–62) reference to turn-initial tokens of
i don’t know as “prefatory epistemic disclaimer[s]” highlights these tokens’ mul-
tifunctionality: they initiate a turn while at the same time hedging its content. This
interpretation is confirmed by Weatherall’s (2011) analysis of turn-initial tokens
of i don’t know that constitute grammatically independent turn-constructional
units. With the appropriate prosodic encoding, these tokens can signal epistemicity
and turn continuation simultaneously. Weatherall also successfully demonstrates
that the functionality of i don’t know is affected by the construction’s prosodic
encoding as well as by whether the next action follows a pause or contains an epis-
temic phrase. Diani’s (2004) attempt to establish how the functionality of i don’t
know is affected by its co-occurrence with other discourse-pragmatic features is
only moderately successful. Not only is her analysis not consistently concerned
with i don’t know as a pragmatic device, but her discussion is not firmly fo-
cused on the functionality of i don’t know (see, for example, Diani’s [2004: 170]
 Chapter 4.  The construction i don’t know 73

­assessment that “in collocation with I don’t know, you know seems to be used as
an invitation to acknowledge a new piece of information”).
The studies outlined above have analysed the functional versatility of i don’t
know in different situational contexts, thereby demonstrating the construction’s
extensive functional range in discourse. However, because these studies have
largely focused their analyses on the use of i don’t know in specific sequential
contexts, notably second pair parts of (question-answer) adjacency pairs, we can-
not be sure how comprehensive currently available descriptions of i don’t know
are, or which of the functions introduced above are the most common in discourse.
Moreover, while most studies have emphasised that the pragmatic meaning of a
particular instance of i don’t know depends largely on its sequential context, few
studies have successfully described the effect of the ambient linguistic context on
the functions performed by i don’t know. The current study will address some
of these limitations.

4.2.2 Quantitative studies: Variation and change in the use of i don’t know

As pointed out at the beginning of this chapter, quantitative studies of i don’t

know are still in short supply. This shortage is regrettable, for the few extant stud-
ies demonstrate that quantitative analyses of i don’t know’s functional versatility,
syntactic positioning and formal encoding have the potential to enhance our un-
derstanding of the extra- and intra-linguistic mechanisms which create discourse-
pragmatic variation.
Aijmer’s (2009) analysis of i don’t know in a corpus of spoken learner English
and a native speaker control corpus reveals the following differences between ad-
vanced learner and native speaker usage of i don’t know: learners prefer to use i
don’t know in turn-final position as a speech-management signal (e.g. to curtail
topics); native speakers prefer to use it in turn-initial position as a face-saving
device (e.g. to avoid answering questions directly). Aijmer (2009: 166) attributes
learners’ frequent use of i don’t know for textual functions to the fact that they do
not have at their disposal the same range of linguistic resources as native speakers
to overcome planning difficulties in spoken interaction.
Similar acquisition-based differences in the use of i don’t know are reported
in Baumgarten & House (2010). Their analysis reveals that in English as a lingua
franca interactions i don’t know tends to be used mainly in simple-clause and ma-
trix-complement constructions as an expression of insufficient knowledge; when
used in extra-sentential position, it tends to function as a non-conventionalised­
signal of on-line planning difficulties. By contrast, in their recorded native speaker
interactions, i don’t know is rarely used in clausal constructions and occurs
74 The Structure of Discourse-Pragmatic Variation

mostly in extra-sentential positions with conventionalised interpersonal and tex-

tual functions such as stance-marking and turn-completion. Baumgarten & House
(2010: 1197) attribute the differential distribution and use of i don’t know in
learner and native discourses to inter-variety differences in the functional diver-
sification of i don’t know which result from grammaticalisation processes only
affecting native varieties of English.
The grammaticalisation of i don’t know and its effects on the construction’s
phonetic encoding, functional versatility and constituent structure are discussed
in two studies drawing on the same data of American English: Bybee & Scheibman
(1999) and Scheibman (2000). Depending on the realisation of the vowel in don’t
as either full, [o], or reduced, [ǝ], these scholars classified the 37 tokens of i don’t
know in their data of naturally occurring conversations into two categories: full
vowel variants and reduced vowel variants.22 Distributional analyses showed that
an overwhelming majority of tokens were produced with a reduced vowel (78%,
N = 29). Further, the results revealed a strong correlation between phonetic reali-
sation and discourse function: referential uses were realised with full variants as
well as reduced variants (N = 7 and N = 12 respectively); tokens functioning as a
mitigation device were overwhelmingly realised with reduced variants (N = 9 out of
10); and all turn-yielding tokens of the variable were realised with a reduced vowel
in don’t (N = 8). Scheibman (2000: 120) summarises the distribution of phonetic
variants according to function as follows:
[a]ll variants of don’t in [i don’t know] convey the expression’s lexical mean-
ing of ‘not knowing’, but, with one exception, only reduced vowel forms occur
in contexts of the collocation’s interactive, face-saving functions. Moreover, only
reduced vowel variants participate in what is arguably the most conventionalized,
or ritualized, use of the expression – that of signalling a speaker change.

These results, then, reveal a distribution of variants along discourse-functional

factors: full variants are associated with referential meanings and reduced variants
with pragmatic meanings.
According to Bybee & Scheibman (1999), the underlying cause of the func-
tional split between variants is high frequency of use and repetition. With increased
use in interaction, i don’t know has become stored and processed as a single unit
(chunking) and has gradually weakened its association with its composite parts

22. Bybee & Scheibman (1999: 579–580) initially also differentiated between full vowel vari-
ants that were produced with an initial stop or an initial reduced consonant (flap), and between
reduced vowel variants that were produced with a reduced initial consonant or no initial conso-
nant (see also Scheibman 2000: 107–109). They disregarded the consonantal differences in the
quantification of the data.
 Chapter 4.  The construction i don’t know 75

(I, don’t, know) (autonomy) (see also Bybee 2003: 617–618, 2006: 725–726). The
loss of compositionality and analysability has triggered morpho-phonological and
semantic-pragmatic changes which obscure the original internal structure and
meaning of i don’t know. The association of the non-fused and non-reduced
variant with referential uses on the one hand and of the fused and reduced vari-
ant with pragmatic uses on the other hand, then, is a result of the variable’s gram-
maticalisation, i.e., “the process by which a frequently used sequence of words or
morphemes becomes automated as a single processing unit” (Bybee 2003: 603).
The studies outlined above demonstrate the structured nature of observed
variation patterns in the usage and distribution of i don’t know, and highlight
the importance of function in conditioning the variation. Function accounts for
the construction’s differential frequency and positional distribution in learner and
native speaker interactions, and operates as a constraint on its formal variation in
American English conversation. The limited size of Bybee & Scheibman’s dataset
(37 tokens from six speakers aged 27–52, two male and four female) precludes
investigation of the social embedding of variation in the form of i don’t know
and of the robustness of the functional split between its variants. Grant’s (2010)
analysis of the functional distribution of full and reduced variants of i don’t know
across age, sex, region and conversation type in the spoken part of the British
National Corpus (henceforth BNC) (one million words) fails to provide this in-
formation because it suffers from serious methodological flaws. Firstly, as Grant
(2010: 2290) readily admits, it is not known on what basis the transcribers of the
BNC distinguished between full and reduced variants of i don’t know or how
rigorously this distinction was adhered to in the transcription process. Secondly,
and perhaps even more worryingly, Grant reports raw numbers of the variable
and its variants instead of following standard variationist procedure, i.e., indexing
frequency scores of the variable as proportions of total word counts and report-
ing the proportional frequency of variants out of the variable, as dictated by the
principle of accountability (see Chapter 2.3.2). In the absence of reliable transcrip-
tions and accountable quantifications, Grant’s (2010) study can offer no insights
into the formal variability of i don’t know in contemporary British English.23 In
contrast to Grant (2010), the following study of i don’t know is based on faith-
ful transcriptions and accountable statistical analyses, thus affording reliable new
insights into the variable’s distribution in discourse.

23. The same criticisms apply to Grant’s (2010) analysis of i don’t know in the Wellington
Spoken Corpus. However, this part of the study does not investigate form variation in the con-
struction’s use and is therefore not of concern here.
76 The Structure of Discourse-Pragmatic Variation

4.2.3 Aims and contribution of this chapter

As mentioned at the beginning, this chapter investigates the high degree of varia-
tion in the realisation of i don’t know in a sizeable and socially stratified corpus of
a variety of English spoken in north-east England. It differs from previous analyses
of i don’t know by testing the effect on variation of social factors, conducting an
apparent-time analysis of i don’t know variation, and studying the distribution
of both non-localised and localised variants.
Inclusion of social factors alongside linguistic factors as independent variables
in the analysis allows us to model the simultaneous effect of multiple contex-
tual factors (age, sex, function, syntax) on the formal variation of i don’t know.
Further, apparent-time analysis of a large number of tokens of the variable makes
it possible to establish the robustness of variation in the data, and to identify the
existence and trajectory of any ongoing changes to variants’ intra- and extra-lin-
guistic conditioning. Finally, investigation of a variety which features both non-
localised and localised variants of i don’t know (see further Section 4.3.2) enables
us to determine whether constraints on variant choice are identical across variants
with differential geographical distribution. The current project thus has a much
broader scope than previous studies which focused exclusively on the distribution
of non-localised variants, left unaccounted for the role of extra-linguistic factors
in the distribution of i don’t know variants, and did not probe the stability of
form-meaning correlations in apparent time (see Section 4.2.2). The broad scope
of the present analysis contributes new insights into the grammar underpinning
the formal variation of i don’t know and thereby illuminates the sociolinguistic
embedding of discourse-pragmatic variation and change.
In order to integrate function as an independent variable in the quantitative
analysis and test previous claims about its impact on the formal variation of i don’t
know, this study must necessarily establish the inventory of functions performed
by i don’t know in the BwE corpus. The qualitative data analysis presented in
Section 4.4 represents a development beyond the studies reviewed in Section 4.2.1
because it investigates all tokens of the variable in their range of utterance- and
turn-positions, fully acknowledges the multifunctionality of individual tokens of
the variable, and describes the effect of co-occurring linguistic material on the
functionality of i don’t know. The analysis thus yields a comprehensive func-
tional taxonomy of i don’t know which, albeit representative only of interview
data, makes an important contribution to current descriptions of the construc-
tion’s functional repertoire. Also, by quantifying the results of the qualitative data
analysis, the study establishes the relative frequency and importance of the various
functions performed by i don’t know in the interview data.
 Chapter 4.  The construction i don’t know 77

Prior to presenting the results of the qualitative and quantitative analyses,

the next section sets out the procedures for circumscribing the variable context,
allocating tokens of i don’t know to variant categories, and coding the data for
extra- and intra-linguistic conditioning factors.

4.3 The variable context and data coding

4.3.1 The variable and the envelope of variation

The variable under investigation in this chapter is the construction i don’t know,
as defined in Chapter 2.3.2. To yield principled and accountable results, every
instance of a construction derived from the following linear string of components
was extracted from the data: (I) + (present tense negative do) + (know). This
yielded tokens in different syntactic configurations: unbound tokens of i don’t
know without overt complementation (2a); tokens with a dependent wh-word
(2b); and bound tokens with phrasal or clausal complements in the form of noun
phrases (2c), adverbial phrases (2d), prepositional phrases (2e), finite conditional
clauses with if or whether (2f), or finite wh-interrogative clauses (2g).24 This study
departs from the practice in previous studies (e.g. Aijmer 2009) and includes in
the variable context all tokens of i don’t know in all syntactic configurations. This
decision was motivated by the preliminary observation that all instances of i don’t
know in the BwE corpus are variable in form and function.
(2) a. I dinnae ken, Mary. I’m no sure.
Americans just, I dunno, they have this thing about Scotland.
b. And I dunno why, but people in Berwick have a stigma.
c. But I feel sort of intimidated wi Muslims, cos I divn’t knaa their religion.
d. I dunno much about it.
e. Drunk? I dunno about drunk now. What would drunk be?

24. When i don’t know signals pragmatic meanings, we are not in fact dealing with a matrix-
complement construction as is the case when it functions referentially. Instead, we are dealing
with a mono-clausal construction that contains an epistemic frame (see further Section 4.6). For
ease of reference, however, the labels ‘matrix clause’ and ‘complement clause’ are used throughout
this chapter to identify the construction i don’t know and the clause over which it has scope,
irrespective of their actual syntactic hierarchy or status in specific contexts of use (see also Boye
& Harder 2007; van Bogaert 2010).
78 The Structure of Discourse-Pragmatic Variation

f. And I don’t know if they have a Romany connection.

I don’t know whether it’s very much of an age thing and that.
g. I divn’t knaa what else I would call the rain.

Following standard variationist procedure, tokens of the variable whose form

could not be determined and tokens which occurred in quoted speech were ex-
cluded from the database. A total of 600 tokens of the variable i don’t know were
retained for in-depth analysis. Each of these tokens was coded for its form and a
series of extra- and intra-linguistic factors which were hypothesised to have an
effect on the construction’s morpho-phonology. The following sections detail the
coding decisions.

4.3.2 The dependent variable: Variants of i don’t know in BwE

The extracts from the data given in (1) in Section 4.1 illustrate that the variable
i don’t know manifests a high degree of formal variation in BwE. The variation af-
fects the realisation of negative periphrastic do, the morpheme boundary between
negative do and the cognition verb, as well as the choice of the negator and the
lexical item signalling a cognitive state. Although allocation of tokens to discrete
categories necessarily masks the full details of variation, close auditory analysis of
all tokens of i don’t know in the BwE corpus indicates that, for quantitative pur-
poses, tokens can be meaningfully divided into five variant categories. This section
introduces the five formal variants identified in the data, and broadly groups them
into localised and non-localised variants based on their geographical dispersion
within England and the English-speaking world.
Tokens in the first three categories of variants differ from each other with re-
gard to the morpheme boundary between don’t and know (audible vs. non-audible)
and/or the vowel quality in don’t (full vs. reduced). The first category contains to-
kens such as [dɔnʔnɔː] or [dʊnʔˈnʊ]: the vowel in don’t is realised as [ɔ] or [ʊ] and
a conspicuous morpheme boundary occurs between the nasals of don’t and know,
usually in the form of [ʔ]. Due to their relative lack of reduction, these variants
are labelled full variants and are orthographically represented as I don’t know. The
second category contains tokens such as [dʊˈnɔː]: the first vowel is produced with
lip-rounding but in contrast to the full variants there is no morpheme boundary
between the nasals of don’t and know. Because they share properties with full vari-
ants (lip-rounded vowel) as well as reduced variants (inaudible morpheme bound-
ary), these variants are labelled semi-reduced variants and are orthographically
represented as I dono. The third category contains tokens such as [dǝnɔ] or [dǝnʊ]:
 Chapter 4.  The construction i don’t know 79

there is no audible morpheme boundary between the nasals of don’t and know, and
the vowel in don’t is reduced to [ə]. Because of their high degree of articulatory
reduction, these variants are labelled reduced variants and are orthographically
represented as I dunno.
The full variant I don’t know is the standard variant and as such is non-
localisable­. The semi-reduced variant I dono represents a stage in the development
from non-reduced to reduced realisations (don’t know > dono > dunno) which is
likely to occur in any variety in which i don’t know is affected by gradual fusion
and attrition due to frequency of use and repetition (see further Section 4.6). The
vowel quality in the first syllable of dono will, however, differ across varieties. The
reduced variant I dunno is cited in the Oxford English Dictionary (henceforth OED)
(2010) without a geographical label. Although the precise nature of the variation
is not always detailed, the co-occurrence of full and reduced variants of i don’t
know has also been reported for other varieties in the UK and the US (Aijmer
2009; Baumgarten & House 2010; Bybee & Scheibman 1999; Grant 2010; Östman
1981; Scheibman 2000). Since it can be assumed that they are widely distributed
throughout England and the English-speaking world, the variants I don’t know,
I dono and I dunno are categorised as non-localised variants.
Tokens in the fourth and fifth categories of variants differ from those above
in that negative periphrastic do is realised as divn’t or dinnae (see Chapter 3.3)
and the cognitive verb may in some instances take a form other than know. The
category of variants orthographically represented as I divn’t knaa contains tokens
such as [dɪvn̩nɜ̃ː] or [tɪfn̩ˈnɐ]: negative periphrastic do contains a kit-vowel and
some degree of friction. Also, with this category of tokens, the lexical item know
is usually, but not always, replaced with knaa [nɐ]. The category of variants ortho-
graphically represented as I dinnae ken contains tokens such as [dəni̞x̃ en]: peri-
phrastic do is negated with the negative clitic -nae, yielding the negative auxiliary
dinnae [dənɪ] or [dənə]. Also, with this category of tokens, know is usually, but not
always, replaced with ken [ken].
As detailed in Chapter 3.3.2, the forms divn’t and dinnae are not used through-
out all of England or the English-speaking world. Divn’t is strongly associated
with the north(-east) of England; the same is true of knaa (Griffiths 2005).
Dinnae is strongly linked with Scotland, although its use has been recorded in
Northumberland; ken is a shibboleth of Scottish speech (Grant & Murison 1931).
Because their distribution in England is largely restricted to the north(-east), the
variants I divn’t knaa and I dinnae ken are categorised here as localised variants.
Table 4.1 summarises the BwE inventory of variants for i don’t know, with
an indication of their geographical distribution in England.
80 The Structure of Discourse-Pragmatic Variation

Table 4.1  Inventory of variants of the construction i don’t know in BwE

non-localised variants localised variants
I don’t know I divn’t knaa
I dono I dinnae ken
I dunno

4.3.3 Independent variables: Data coding

In addition to coding each occurrence of i don’t know for its variant form, every
token of the variable was coded for the operation of extra- and intra-linguistic
constraints hypothesised to have an effect on variant selection.
Previous research has demonstrated that the grammaticalisation processes
leading to the synchronic co-existence of discourse-pragmatic variants are so-
cially embedded, and that social factors, including sex, age, socio-economic class,
education and locality, are strongly implicated in the synchronic distribution of
formal variants in discourse-pragmatics (see, for example, Buchstaller & D’Arcy
2009; Cheshire 2007; Ferrara 1997; Ito & Tagliamonte 2003; Macaulay 2002b;
Palander-Collin 1999; Pichler & Levey 2011; Tagliamonte & Hudson 1999).
Moreover, the results reported in Chapter 3 have shown that in productive con-
structions the occurrence of negative auxiliaries formed with -nae is strongly
constrained by age. Assuming, for now, that the formal variation illustrated in
(1) and introduced in more detail in Section 4.3.2 reflects the layering effect as-
sociated with grammaticalisation and that conditioning factors remain constant
across productive and formulaic constructions, we may hypothesise that social
factors are implicated in the formal variation of i don’t know. To test this hypoth-
esis and probe the social distribution of variants, each token of the variable was
coded for speaker sex and age. Inclusion of the independent variable age in the
coding protocol also allows us to evoke the apparent-time construct with a view
to identifying any ongoing changes to the conditioning of variants, and locating
the leaders of any such changes. All tokens were further coded for individual
speaker. This allows us to control for individual speaker effects on the extra- and
intra-linguistic distribution of variants in the multivariate analysis of the data
(see Chapter 2.3.3).
Bybee & Scheibman (1999) and Scheibman (2000) have demonstrated that the
variation between full and reduced variants of i don’t know in American English
conversational data is conditioned by discourse-functional constraints. Similar
conditioning effects arising from grammaticalisation have been reported for other
discourse-pragmatic variables. For example, Lindemann & Mauranen (2001) and
Stenström (1998) found that phonetically full forms of just and because correlated
 Chapter 4.  The construction i don’t know 81

with older layers of meaning (‘exactly’ for just; grammatical subordination for
because), while phonetically reduced forms correlated with newer, grammatical-
ised layers of meaning (mitigation for just; continuation signalling for because)
(see also Andersen 2001: 213; Drager 2011 on lexical vs. discourse like). To test
whether function is also implicated in the formal variation of i don’t know in
BwE and whether it affects non-localised and localised variants alike, all tokens
of the variable were coded for their function in the data. Viewing any emergent
form-function correlations in apparent time will enable us to assess whether
they are conventionalising and strengthening over time, as predicted by Bybee
(2006: 725–726). The results of the qualitative data analysis and the coding schema
for discourse-functional constraints are outlined in Section 4.4.

4.4 Qualitative analysis of i don’t know

Applying the methods outlined in Chapter 2.5, this section identifies the functions
performed by i don’t know in the BwE interview data. To demonstrate the valid-
ity of the analysis, each function is illustrated with one or more examples from
the data. The examples are presented in their sequential and interactional con-
text of occurrence; typographical means are employed to replicate accompanying
paralinguistic and prosodic features (see page xvii for a key to the transcription
conventions). These features as well as lexical material in the ambient context,
notably co-occurring discourse-pragmatic features, adverbs and filled pauses, are
discussed below when they make a systematic contribution to data interpretation.
Recurrent co-occurrence patterns are listed in Appendices 1–3 which summarise
the results of the qualitative analysis. Where pertinent to the analysis, the larger
discourse context and the situational context in which i don’t know occurs are
attended to in the discussion of the data, and specific reference is made to any
CA theories implicated in the interpretation of the data. Through providing a
detailed description and illustration of the functionality of i don’t know in the
BwE interview data, the analysis presented here yields a detailed coding protocol
for the factor group functionality (see Table 4.2 in Section 4.4.4), suitable for
replication and elaboration in future studies of the variable.
Because tokens in different syntactic configurations cover different functional
spectra, the functionality of unbound tokens, bound tokens with a phrasal or
clausal complement, and tokens with a dependent wh-word is discussed sepa-
rately. I will start with unbound tokens since these are numerically the most fre-
quent and functionally the most versatile. The discussion of other tokens will be
less detailed, relying heavily on the insights gained from the preceding analysis of
unbound tokens.
82 The Structure of Discourse-Pragmatic Variation

4.4.1 Functions performed by unbound i don’t know

Unbound tokens of i don’t know constitute structurally and intonationally inde-

pendent units in discourse which can occur in a range of clause or utterance posi-
tions. As a result of their syntactic freedom and positional mobility, they perform
a wide range of interpersonal and textual functions related to expressing speaker
attitudes and coordinating speaker-hearer relations as well as creating structure
and coherence in discourse.
In the interpersonal domain, i don’t know functions subjectively to signal
speakers’ doubts regarding the accuracy of their propositions. This use is illustrated
in Extract (3) from the data, where Keith tells the interviewer HP and his interview
partner Adam about a confrontation he had during a night out in Carlisle. Keith
signals that his assessment of the origin of his interlocutors may not be reliable by
retrospectively qualifying it with I dunno. The tentative effect of I dunno is pro-
duced and reinforced by its fall-rise intonation contour and its co-occurrence with
other hedging devices (probably, or something like that). The qualification fulfils
an important face-saving function because it allows Keith to avoid committing
himself to his proposition.
(3) Keith: When I was in the in the toilet, someone well a a lad said
HP: mhm
Keith: = “What do you want a cigarette?” And I said “No.” And then
HP: @
Keith: = his pal came in. They were I-Irish. They were [probably] from up
HP: [yeah]
Keith: = the road. They were probably [gipsies] or something like that.
HP: [mhm]
Keith: = I dunno? And they were (.) they said “O:h. Where are you
HP: [mhm]
Adam: [@]
Keith: = from.” You know, straight away.

i don’t know does not always occur outside the main constituents of an utter-
ance, nor does it always have scope over entire propositions. This is illustrated in
(4), where two teenage informants had been asked their opinion of the Labour
party’s regional government proposals for the north-east. Rebecca argues in fa-
vour of a devolved government and implies that whereas the present govern-
ment is remote from Berwick and oblivious to its current affairs, a regional
government in nearby Durham would be better informed about the town. The
end of Rebecca’s turn contains several linguistic features that are indicative of
her difficulty in formulating a coherent justification for her support of regional
 Chapter 4.  The construction i don’t know 83

devolution: a false start (and it’s just), prosodic lengthening of sounds (mo:re,
s:ense), unfilled pauses, and instances of i don’t know as turn-holding and
epistemic repair devices. Near the end of the extract, we find another instance of
I dunno. This token occurs internal to the syntactic structure between obligatory
constituents of the clause: NP I dunno VP. By bringing not the entire proposition
but merely the immediately following lexical items in its scope, I dunno signals
that the following expression only loosely communicates Rebecca’s thoughts
and/or that she is unsure of her choice of wording. The attenuating effect of I
dunno is achieved prosodically through its rising intonation contour and the
slow production of the items in its scope.
(4) Alicia: I think it would make a difference. [I think] we’d be included a
HP: [mhm]
Alicia: = lot more [in things.]
HP: [yeah]
Rebecca: [Yeah.] (h) Because like (.) likes of the people down
in London, they don’t know >what’s going on in Berwick
[and things.]< (h) (..) And (.) they’d have to travel forever to get
HP: [mhm]
Rebecca: = here. (h) [Whereas if] <they’re just at Durham,> (..) although
HP: [mhm]
Rebecca: = they’re not gonna be here, (.) they’re closer! And it’s just
HP: mhm
Rebecca: = (.) I dunno? It’ll be mo:re (..) <@ I dunno really? It would be
better. @> [They] just (..) I dunno? >have more s:ense< of
HP: [mhm]
Rebecca: = what’s going on (.) really, if they were in the north-east.

In the interpersonal domain i don’t know can also function intersubjectively.

It mitigates interactional conflict by reducing the anticipated unwelcome effect
of disparaging remarks, contentious propositions or disagreements. This effect is
implicit in Examples (3) and (4) above and is illustrated more clearly in the ex-
amples in (5) below. In (5a), Luke blames children’s drug abuse on their parents’
negligence. By prefacing his turn with I divn’t knaa and terminating it in final
rising pitch, Luke marks his assertion to appear uncertain and unreliable (see also
his use of the hedges I think and just). If challenged, this allows him to distance
himself from this potentially controversial assessment and to avert open conflict.
In (5b), Lori disaligns herself from her father’s positive evaluation of the Isle of
Skye. Her diverging view is prefaced and delayed by I dunno which works towards
accomplishing the disagreement in a mitigated way and to redress its potentially
face-threatening effect (Pomerantz 1984: 64–77) (see also Tsui 1991: 610–612).
84 The Structure of Discourse-Pragmatic Variation

The fact that Lori starts her modified disagreement in a higher than usual pitch
reinforces this effect.
(5) a. Luke: For the kids that are on drugs I blame the parents me. @
HP: Why.
Luke: I divn’t knaa? I think they’re just (?) Aye. They’re no looking
after their kids properly, or they just (.) just letting them get
away wi it?
b. Godfrey: ºYeah,º it was it’s a really nice place.
HP: [ºYeah.º]
Lori: [I dunno,] because I got
Godfrey: == I had a great holiday up there.
HP: Yeah?
Lori: I got bored and sunburnt.

Besides using i don’t know to attenuate assessments, speakers sometimes claim

insufficient knowledge in order to avoid making assessments (see also Beach &
Metzger 1997: 575; Tsui 1991: 609–610). This is particularly true in situations
such as (6) where provision of an assessment may lead to interactional conflict.
In this example, Leah and Shannon discuss the fact that some Berwickers dif-
ferentiate between the north and south parts of the town and regard the north
side as the ‘posh’ side. Shannon, who lives on the south side of the town, is getting
increasingly upset as the conversation proceeds on this issue (see the increase
in volume and speech rate in the third line of the extract). When she asks Leah
why her sister, Judith, disdains the south side of the town, Leah withholds the
requested assessment. She uses don’t know to avoid attending to an issue that
seems highly sensitive to her friend.
(6) Shannon: ºWhat’s wrong with (.) not north?º @
Leah: Judith doesn’t like (.) not north
Shannon: <WHAT’S WRONG [WITH NOT NORTH?>]
HP: [No. But] some people, you know, make
the difference between north and south.
Leah: Mhm.
Shannon: See, I [>never ever<] WHY.
Leah: [See, Judith does.]
Don’t know.
Shannon: But she lives in ORD!
Leah: <£ Yeah, I know. But that’s posh Ord. £>
 Chapter 4.  The construction i don’t know 85

In addition to the functions i don’t know performs in the interpersonal do-

main, it fulfils a range of functions relating to repair-marking, topic-development­
and turn-exchange in the textual domain. As shown in (7), i don’t know can
be used to fill the gap between the repaired and the repairing segments in self-
initiated turn-repairs “in which an emerging utterance is stopped in some way
and is then aborted, recast, continued or redone” (Fox et al. 1996: 190). In (7a),
Daniel recasts his utterance after a cut-off; in (7b), Charlene abandons her initial
utterance. In both examples, the transition between the interrupted or aban-
doned utterance (I s-, I think it’s getting more) and the beginning of the rephrased
or recast utterance (I keep saying …, It’s only an hour …) is bridged with I dunno.
In (7b), I dunno is surrounded by filled and unfilled pauses and uttered with
moderately higher but lowering pitch. This introduces an element of hesitation
to the repair sequence.25
(7) a. Daniel: No, well, (.) I s- I dunno, I keep saying, (.) if if it’s so good
where you came from, why don’t you go back? That’s what
I say, you know?
b. Charlene: I think it’s getting mo:re (..) e:m °I dunno.° (..) It’s only an
hour on the train to [Newcastle] or Edinburgh now, so I
HP: [mhm]
Charlene: = think (.) a few people sort of work [(h) (.)] you know,
HP: [mhm]
Charlene: = Edinburgh or Newcastle or (.) and they prefer to live (.)
either Alnwick or (.) Berwick or somewhere similar than,
you know, in the city.

In the textual domain, i don’t know can also be strategically employed by speak-
ers to affect the topical development of interactions. This usage of i don’t know
is particularly pertinent to interview situations. Because the topical agenda is
generally set by the interviewer, successful topic-development requires interview-
ees’ collaboration and access to proffered topics (see Geluykens 1993; Schegloff
2007: 169–180). What happens when these conditions are not met is illustrated
in Extract (8) where HP, the interviewer, asks Gabriel, the interviewee, about age-
based differences in non-standard language use. Gabriel replies to HP’s follow-up
(Why do you think) with the minimal response token dunno. This token serves

25. By virtue of bridging a gap in discourse, repair tokens of i don’t know also aid speakers
in securing their hold on the floor. What distinguishes the tokens in (7) from the turn-holding
tokens discussed in (12) below is their syntactic context of occurrence, i.e., the former’s occur-
rence between repaired and repairing elements of utterances.
86 The Structure of Discourse-Pragmatic Variation

two functions. On the one hand, it attends to HP’s face needs as an interviewer by
providing a response to her question, albeit a minimal one. On the other hand, it
serves to decline the proffered topic by disavowing access to it (see also Pomerantz
1984: 57–58; Schegloff 2007: 169–180). Although Gabriel’s dunno does not have
a marked descending intonation contour, it is strongly closure-implicative, not
least because it cannot be challenged (e.g. with ‘why not?’). This is attested by
HP’s subsequent behaviour. HP does not probe Gabriel’s minimal response but
acknowledges the complexity of her question (It’s hard to say) and almost imme-
diately moves on to the next item on the interview agenda.
(8) HP: Would you say that younger people, older people use more
non-standard grammar than younger ones?
Gabriel: Yeah.
HP: Yeah? Why do you think.
Gabriel: Dunno
HP: It’s hard to say.
Gabriel: Mhm.
HP: == Yeah. (..) Ok. (h) And do you think there’s a difference
between girls and boys, or males and females?

As well as to decline topics, i don’t know serves to curtail topics (see also Beach
& Metzger 1997: 571–575; Ford & Thompson 1996: 169–170). Example (9) illus-
trates this usage. Gregory and Julia initially embrace the topic proffered by HP’s
question regarding characteristic features of BwE. However, following a brief and
unsuccessful attempt to provide a satisfactory response, Gregory curtails the topic
by declaring insufficient knowledge to expand on the issue. Gregory’s desire to
bring the topic sequence to a close is reinforced by the following just can, with just
indicating “a speaker’s assumption that nothing more can or need be said about a
proposition” (Wauchope 1993: 182).
(9) HP: What gives it away as a Berwick accent.
(..)
Gregory: Couldn’t really nail it down like, but
Julia: No. You cannae, but it’s just
(..)
Gregory: You can tell when someone’s (.) frae Berwick. You can,
Julia: You can.
Gregory: = don’t know just can.
HP: mhm
 Chapter 4.  The construction i don’t know 87

Finally, as far as topic-development functions are concerned, i don’t know is

used by interviewees to discard successful interruptions by their interview part-
ners (see also Beach & Metzger 1997: 571–575). This function is illustrated in (10)
where Adam provides information about football in Berwick. When Adam uses a
lengthened filled pause (e::m) mid-utterance, Keith successfully disrupts his hold
on the floor to put forward a verification-seeking tag question. Adam’s response,
I divn’t knaa, attends to Keith’s face wants by providing the required second pair
part of the adjacency pair. Yet at the same time, this response disclaims the knowl-
edge required to respond to the question more fully (see (8) above). Talk on the
issue raised by Keith cannot therefore be sustained and Adam is in a position to
resume his account of football in Berwick, with but marking the return to the
original topic (Schiffrin 1987: 177). The slight prosodic rise on knaa adds a tone
of tentativeness to Adam’s minimal response. It makes his reluctance to engage
with Keith’s question sound less abrupt and hostile, and thus presumably more
acceptable to Keith.
(10) Adam: Well, Berwick Rangers they s::- (..) play in the Scottish league
HP: mhm
Adam: (h) (h) but then you have like $ [e::m (.h)]
Keith: [They] can play in either or
though, can’t they?
Adam: I divn’t knaa? But
Keith: == They used to be (?).
Adam: You have like (.) (h) teams like em Highfields and (.) Spittal
HP: mhm
Adam: = and that. [And] they play wi like (.) in a league called the Nor-
HP: [mhm]
Adam: = thern Alliance.

However, the most frequent textual uses of i don’t know in the BwE corpus do
not relate to repair-marking or topic-development but to turn-exchange. In com-
bination with other linguistic material, the construction regularly serves to affect
and prevent turn-transfer between interviewees. The extracts from the data in
(11) serve to illustrate the use of i don’t know as a turn-taking device. In (11a),
Godfrey launches his turn at the very moment Lori’s turn reaches a complex tran-
sition relevance place (henceforth CTRP), i.e., a place where speaker change can
occur because the current speaker’s turn has reached syntactic, pragmatic and
intonational completion (Ford & Thompson 1996). Godfrey’s turn-initial instance
of I dunno neither contributes to the propositional content of his turn, nor com-
municates an affective stance (reflected in the lack of noticeable pitch movement).
Instead, it allows Godfrey to secure his hold on the floor while he is still in the
88 The Structure of Discourse-Pragmatic Variation

process of planning his contribution. This interpretation is supported by the fact

that Lori temporarily manages to regain the floor, that I dunno is produced with
increased speech rate, and that other turn elements are present which are sugges-
tive of Godfrey’s lack of planning: a false start; a vague expression (it all depends);
syllable lengthening on the continuation signal because. In (11b), Luke’s turn-initial
use of I dunno is not motivated by a competition for the floor (he is the sole in-
terviewee), an attempt to stall for time (apart from th-, the beginning of his turn
does not contain any elements suggestive of planning or hesitation), nor a desire
to mitigate potential unwelcome effects of his utterance (he overtly expresses his
dislike of Geordies). I dunno here functions as a take-off for further talk. It avoids
impairing the projectability of Luke’s turn, the start of which overlaps with the
closure of HP’s preceding turn (see Sacks et al. 1974: 718–721). In fact, the length-
ening on the final syllable of dunno might be motivated by Luke’s attempt to delay
the actual start of his proposition until HP’s turn reaches a CTRP.
(11) a. HP: Would you say that either old or young people are lazier
or sloppier with their grammar?
Lori: Younger. @
HP: Yeah?
Lori: Younger but then again (I mean) old people, (h) they’ve
been taught like, and miss words out.
Godfrey: == <I dunno,> it it all depends >becaus::e<,
Lori: == Things that they’re saying.
Godfrey: >aulder people. (..) likes of (…) my father’s generation.<
HP: mhm
Godfrey: = They was all (.) constructed in the army and all that
<and they had to change their language.> To be under-
HP: yeah
Godfrey: = stood.
b. Luke: I despise <@ Geordies. @>
HP: Why. Wh-what is [wrong with them.]
Luke: [I dunno::,] every time I go down to New-
castle th- all the Geordies are always really cheeky. You know,
they’ll say °“ah you fucking”° and gie you abuse all the time,
[and] all the rest of it, and then I hate to be associated with
HP: [yeah]
Luke: = them. I just, I hate them. @
HP: yeah
 Chapter 4.  The construction i don’t know 89

In turn-medial position, when it does not occur between elements of a repair se-
quence, i don’t know often functions to secure speakers’ hold on the floor. This
function is illustrated in the examples in (12). In (12a), I dunno occurs between
syntactically complete turn-constructional units (henceforth TCUs), i.e., the build-
ing blocks (clauses, phrases, words) which combine to constitute a turn or action
(Sacks et al. 1974: 702–703). I dunno is produced without noticeable pitch move-
ment, thus projecting that there is more to come and working towards extending
the turn past its possible completion point. The fact that I dunno and the following
lexical elements are uttered in faster tempo than the surrounding discourse indi-
cates that I dunno also works towards preventing other interactants from taking the
floor and enabling Rebecca to build a multi-unit turn. The risk of losing one’s hold
on the floor is highest when speakers hesitate in their construction of an extended
or forthcoming turn, as is the case in (12b). Keith’s use of and following the TCU
And that was with her signals his intention to extend his turn past a CTRP, for and
regularly marks speaker continuation (Schiffrin 1987: 141–150). The fact that the
following I dunno is uttered without noticeable pitch movement and is followed
by a clause containing obviously rule out an epistemic interpretation of this token.
Rather, I dunno, together with the preceding filled pause eh, secures Keith’s hold
on the floor by signalling his communicative presence while he is preparing the
continuation of his turn.
(12) a. Rebecca: They think nothing of it? [It’s just] a normal thing?
HP: mhm
Alicia: [yeah]
Rebecca: = [<I dunno,] I think that’s just> because it’s such large
HP: [mhm]
Rebecca: = widespread country.
b. Keith: And then eh (..) my girlfriend at the time, we broke up and
that [was who: (.)] I was gonna go away with and (.) I’d never
HP: [mhm]
Keith: = rea- I do- never <I’d only been away once before.>
HP: mhm
Keith: = >And that was< with her. [And] eh (..) I dunno (.) just like
HP: [mhm]
Keith: = obviously I wanted to go Australia [and that. I wanted] to go
HP: [mhm mhm]
Keith: = away, like, for a year.
90 The Structure of Discourse-Pragmatic Variation

Apart from turn-taking and turn-holding tokens, the corpus also contains tokens
of i don’t know that function as turn-yielding cues. In (13), Rebecca answers
HP’s question about the positive aspects of life in Berwick upon Tweed posed
shortly before Rebecca’s interview partner, Alicia, rejoined the interview following
a brief comfort break. Although Rebecca talks in response to HP’s question for
no less than 57 seconds (not all reproduced below), her answer fails to directly
address it. After Alicia’s return to the interview room and HP’s informing her
about the current topic of conversation, Rebecca immediately yields the floor
to Alicia. This speaker shift is initiated in multiple ways: (i) the use of the turn-
relinquishing and pre-closing discourse feature so (Schegloff & Sacks 1973: 306;
Schiffrin 1987: 218); (ii) the use of I dunno as an appeal to Alicia to take over, with
the falling intonation on dunno indicating a CTRP; (iii) Rebecca’s question (what
do you think?), which as the first pair part of an adjacency pair requires the provi-
sion of a relevant second pair part, i.e., an answer; and (iv) the contrastive stress
on I in I dunno and on you in the following question. The combination of these
features constitutes a rather forceful turn-yielder. However, the fact that Alicia
begins her response before the completion of Rebecca’s turn-yielding question
suggests that the use of so and I dunno alone might have been quite sufficient in
affecting turn-transfer.
(13) Rebecca: I mean i- likes of (.) like, everyone says <“oh if you go to
Newcastle there’s loads of things to do.” But like what,> what
would you do, if you lived in Newcastle? People
HP: mhm
Alicia: mhm
Rebecca: = there’ll probably say all the time “Oh, there’s nothing to do,
there’s [nothing to do.”]
HP: [mhm mhm] mhm ((to Alicia)) We’re talking about the
good things in [Berwick.] Yeah.
Alicia: [Yeah.]
Rebecca: [@]
ºSo I dunno. ((to Alicia)) What [do you think.º]
Alicia: [I like (.)] the good things. I
like the friendliness of the people. Everyone gets on.

With some turn-final instances of i don’t know, such as those in (14), it is not
entirely clear whether they function as turn-exchange or topic-development de-
vices. When Adam’s contribution in (14a) (I don’t even think you have to go as
far as the border) reaches a CTRP, it is not immediately followed by further talk
or recipient uptake. It is only after an unfilled pause of 2.3 seconds that Adam
 Chapter 4.  The construction i don’t know 91

resumes talk with and, a signal of speaker continuation (Schiffrin 1987: 150), fol-
lowed by I divn’t knaa. The descending intonation contour on I divn’t knaa signals
that the construction functions to re-occasion turn-completion. It is, however,
not clear whether Adam’s intention is to bring the topic sequence to a close or
to yield the turn to Keith. The ambiguity is reflected in the ensuing non-talk: HP
fails to provide the next topic-proffer and Keith does not immediately take up the
floor. In (14b), Leah aborts her turn before it reaches a CTRP (cos you never). The
following I dunno thus functions to signal turn-completion, with the high-rising
intonation signalling an appeal for co-operation. Whether the appeal is directed
at Leah’s interview partner to provide an assessment or to HP to proffer the next
topic is unclear. While the speakers’ intention behind using i don’t know in (14)
may be ambiguous between turn-yielding and topic-closing, the effect is clearly
to close speaker turns.
(14) a. HP: So would you say the border is like a linguistic divide?
Keith: E::h (.h)
Adam: A::h (.) I don’t even think you have to go as far as the border.
(..) And I divn’t knaa. @
(…)
Keith: Um (..) Aye w-w-w-we’re we’re (..) we’re definitely different.
b. Leah: (.h) >Has no effect on Berwick cos you never,< (.) I dunno?

The examples above have shown that some tokens that perform a mitigating
function also signal epistemicity and vice versa, and that some tokens which
function as repair devices or turn-taking/holding devices simultaneously con-
vey a strong element of hesitation. i don’t know can also be multifunctional in
the sense of a single token operating simultaneously across the communicative
domains, i.e., concurrently performing interpersonal and textual functions. In
(15), for example, the turn-initial token of i don’t know is produced with rising
pitch. It functions to launch Barbara’s turn (similar to the tokens in (11) above) as
well as to reduce her commitment to the following assessment regarding causes
for generational differences in non-standard language use (see also her use of
maybe, just and e:h).
(15) ((Barbara has just asserted that older people use more non-standard grammar
than younger people.))
HP: Why do you think that is.
Barbara: I dunno? Maybe just just just e:h education at the schools.
92 The Structure of Discourse-Pragmatic Variation

In (16), i don’t know occurs turn-medially and functions as an epistemic or

mitigating repair and turn-holding device. The token in (16a), by nature of its po-
sitioning between an abandoned and a recast utterance, bridges the gap between
the repaired and repairing element. At the same time, though, it hedges Luke’s
commitment to his upcoming description of the Newcastle dialect, as evident from
the rising contour on I dunno. The token in (16b) serves to link two syntactically
complete utterances while simultaneously working towards toning down Albert’s
negative evaluation of a third party’s opinion.
(16) a. Luke: But with like a proper Geordie from Newcastle they talk really,
(h) like like a proper Geordie, they’re (.) mo:re I dunno? Their
words are more (.) shorter. Frae like a Berwick person.
HP: Yeah?
b. ((Albert just told HP about a man who, after emigrating to New Zealand,
described New Zealand as the best place to live in the world.))
Albert: There was nothing that he (.) eh he described (h) in the place
he lived in New Zealand <£ that was any different frae here £>,
[you know.] I don’t know I I I found that a little bit weird, you
HP: [yeah yeah]
Albert: = know.

Finally, the data contain multifunctional turn-final tokens of i don’t know. In

(17), for example, Leah’s I dunno is produced with a very moderate final fall and a
decrease in volume. It thus serves to retrospectively qualify her description of her
accent as well as to close her turn.
(17) HP: What accent would you say you had and do you like it?
(.)
Leah: Em. It’s a mixture of probably Scottish and Geordie. But °I dunno°.

The preceding discussion has demonstrated the functional versatility of unbound

i don’t know in the BwE interview data. To conclude this section, the extracts
in (18) illustrate referential uses of i don’t know which convey speakers’ lack of
knowledge. Although they do not constitute a discourse-pragmatic feature, these
tokens have been included in the variable context in order to test whether prag-
matic and referential uses of i don’t know favour different realisations.
(18) a. HP: And would you rather have a different accent or dialect?
Ryan: I don’t know? Never really thought about it.
b. Jane: Well, I was a telephonist for years and a lot of people thought,
you know frae further down the country thought I was Welsh.
HP: Why?
Jane: I dunno? We divn’t knaa the connection?
 Chapter 4.  The construction i don’t know 93

The distinction between referential and pragmatic meanings of i don’t know is

necessarily fluid and fuzzy, not least because pragmatic uses are not completely de-
void of referential meanings (see Hopper’s [1991] notion of persistence). Following
previous claims that prosodic features help disambiguate referential from prag-
matic meanings (see, for example, Dehé & Wichmann 2010; Hirschberg & Litman
1993), these features were identified as typifying referential rather than pragmatic
uses of i don’t know: primary stress on know; lack of perceptible variation in
speech rate, loudness and pitch range. (The rising intonation contour on know
in the examples in (18) is moderate and serves to signal speakers’ willingness to
elaborate on their initial declaration of insufficient knowledge.) Moreover, in the
present dataset referential tokens do not generally co-occur with other discourse-
pragmatic features, filled or unfilled pauses, and when they occur as second pair
parts of question-answer adjacency pairs, they are always followed by an account
for the lack of knowledge, as in (18a) above (Never really thought about it), and
never by the provision of the requested assessment, which is typical of epistemic
uses, as in (15) (Maybe just just just e:h education at the schools).
Appendix 1 summarises the functional categories identified for unbound
i don’t know in the BwE data, including short descriptions of their use, charac-
teristic positions and co-occurring linguistic features.

4.4.2 Functions performed by bound i don’t know with phrasal

or clausal complements

In contrast to the unbound tokens discussed above, bound tokens of i don’t know
are followed by a phrasal or clausal complement with which they are prosodically
integrated to form one tone unit. Due to their structural and prosodic integration,
these tokens lack the positional mobility and functional versatility characteristic of
unbound tokens of i don’t know. Their usage is limited to signalling insufficient
knowledge and epistemicity.26
The use of bound i don’t know to signal its referential meaning of insufficient
knowledge is exemplified in (19). In (19a) and (19b), i don’t know communicates
the speakers’ lack of familiarity with the referent in the phrasal complement. In
both instances, i don’t know is accented and the complement carries a falling

26. The strong association between bound i don’t know and epistemicity confirms the validity
of the labels attributed to i don’t know in the literature: “low modality item” (He 1993, quoted
in Kärkkäinen 2003: 24), “epistemic item/marker/phrase” or “subjective marker of epistemic
stance” (Kärkkäinen 2003: 25, 37), “formulaic stance marker” or “stance formula” (Thompson
2002: 125, 139) or “epistemic clause” (Thompson & Hopper 2001: 38).
94 The Structure of Discourse-Pragmatic Variation

intonation contour. These prosodic features, together with the emphasis on the
negative polarity item anybody in the complement in (19b), signal the speakers’
commitment to their assertions and rule out an epistemic reading of their utter-
ances. Prosodic features also suggest a referential reading of i don’t know in (19c).
Almost all elements of Joseph’s utterance, including i don’t know and its clausal
complement, are emphasised and produced in staccato articulation. What Joseph
communicates here is not his uncertainty regarding the defining characteristics of
a true Berwicker, but his cognitive inability to provide such characteristics.
(19) a. Daniel: <I’ve nothing against the blacks! I think the [blacks] are alright
HP: [mhm]
Daniel: = really. But I feel sort of intimidated wi Muslims. Cos
HP: mhm
Daniel: = I divn’t knaa their religion.
b. HP: And do they sometimes when they come here, do they take on
a Spittal accent? [or]
Joseph: [+ (h)] Well, very little, [very little.] No.
HP: [very little]
<@ It’s too hard. @>
Joseph: I-it i- I don’t know of anybody that’s come s-, a stranger
coming in (h) and after a few years (started to speak Spittal).
c. ((Joseph has already indicated that he finds it difficult to provide defining
criteria of a Berwicker.))
HP: Do you have to be born or bred here?
Joseph: I cannae I >don’t know what makes< a Berwicker.

While bound tokens with phrasal complements most commonly function referen-
tially, as in (19a) and (19b) above, it would be misleading to assume that such to-
kens never function epistemically. In specific contexts, phrasal complements after
negative know correspond semantically to wh-clauses (e.g. ‘I don’t know the reason
for that.’ = ‘I don’t know what the reason for that is.’) (Quirk et al. 1985: 1052–1053).
The potential for such tokens to express a non-referential meaning is illustrated
in (20) where Helen uses divn’t knaa to signal her uncertainty regarding Scottish
pensioners’ exemption from paying TV licence.
(20) ((Talking about the financial benefits received by pensioners in Scotland.))
Helen: Well, [they] get free buses for a start. Divn’t knaa
HP: [mhm] mhm mhm
Jane: mhm
Helen: = about the TV licence?
 Chapter 4.  The construction i don’t know 95

With regard to constructions with overt wh-interrogative complements as in (19c),

Quirk et al. (1985: 1051) argue that these “leave a gap of unknown information,
represented by the wh-word.” This is not to say, though, that i don’t know fol-
lowed by a wh-interrogative clause only ever has referential meaning. In (21), for
example, the wh-word does not signal what Glenn lacks knowledge of. Rather, it
signals what it is he is uncertain about.
(21) HP: So what would you say is the local football derby. Berwick Rangers
playing who.
Glenn: E:h (.) Divn’t knaa who it would be? <There isn’t one, is there.>

As predicted by Akatsuka (1985: 635–636), who places constructions containing

negative know towards the irrealis pole of the epistemic scale, the degree of uncer-
tainty expressed by i don’t know is generally very high, especially if the comple-
ment is uttered in continuously rising pitch, as in (21) above, or if the utterance
containing the variable is softly spoken, as in (22) below.
(22) Luke: If I if I won loads of money now, I’d still live in Berwick like.
HP: Yeah?
Luke: In Britain anyway. Aye, definitely. ºI divn’t knaa if I could live in a
cityº.

In (20)–(22) above, the epistemic scope of i don’t know is limited to its comple-
ment. This is not always the case. In certain contexts, such as those in (23) below,
the epistemic scope of i don’t know can extend over the larger surrounding
discourse. In (23a), Keith makes a guess at the transmission date of a television
programme he had seen recently (probably over a year). This guess is hedged not
just with the modal adverb probably but also by the preceding utterance contain-
ing i don’t know in which Keith raises his uncertainty regarding the time of the
broadcast (I divn’t knaa how long ago it was). Similarly, in (23b), Janet’s epistemic
stance relates not only to her being funny but, together with the rising intonation
on the following but-clause, it also modifies her viewpoint expressed in this clause,
i.e., that linguistic variation is socially conditioned.
(23) a. Keith: There’s been a thing on the telly recen- well. I’m saying recent-
ly, it was (.) I divn’t knaa how long ago it was. Probably over a
year. It was to do wi Eyemouth.
b. ((Talking about the causes of inter-speaker linguistic variation.))
Janet: I d- I’m no <I dunno whether I’m being> (..) s- funny here or
no, but it’s more of a social thing?
96 The Structure of Discourse-Pragmatic Variation

The potential of bound i don’t know to have global scope is restricted to the
following contexts: where the construction’s complement explicitly raises the
question answered in the following utterance, as in (23a); where it contains in-
formation that does not constitute the main argument of a speaker’s turn, as
in (23b); or where it constitutes what Brown & Levinson (1987: 168–171) call
a “relevance hedge,” as in (24). In this example, i don’t know per se has local
scope over the following conditional clause (if you’re interested). Yet the utterance
containing i don’t know has more global scope. It constitutes a parenthetical
discourse unit which comments on the rest of the turn and signals Daniel’s un-
certainty as to whether pictures of historic Berwick are in fact relevant to HP
and the interview context. The subsidiary status of I dunno if you’re interested is
indicated by the heightened tempo with which it is spoken (see Couper-Kuhlen
& Selting 1996: 28).
(24) Daniel: I’ve got em (..) old pictures of Berwick eh (..) I’ll get a a picture of
Berwick <I dunno if you’re interested> of the old where it looks
like in the old days.

For the sake of completeness, the examples in (25) are included here to illustrate
that syntactic structures containing i don’t know can also serve textual func-
tions in discourse, similar to those identified for unbound tokens. In (25a), Cody’s
assertion that he lacks the knowledge necessary to discuss the effects of Scottish
devolution on the border serves to decline the proffered topic. In (25b), Shannon’s
assessment that she is uncertain whether the Scottish parliament has devolved
powers other than the ones already discussed works towards terminating the topic
under discussion. Finally, in (25c), Natalie’s uncertainty regarding the geographical
distribution of the dialect word doylem has a turn-yielding effect. In all examples,
it is the positioning of the utterances containing i don’t know that triggers their
textual effect.
(25) a. ((Following HP’s question whether the significance of the Scottish-
English border has changed in the aftermath of Scottish devolution.))
Glenn: <@ Can’t remember. @>
Cody: Neither do I. ºI dunno much about it.º
b. ((At the end of a lengthy discussion about the devolved powers of the
Scottish parliament.))
HP: Any other reasons why you would rather be governed by
Scotland?
Shannon: I dunno anything else that they do?
 Chapter 4.  The construction i don’t know 97

c. ((Discussing the dialect word ‘doylem’ for standard English ‘idiot’.))

Natalie: Aye, but you occasionally Betty comes in, she says “Oh,
you’re such a doylem, man.” (.) Don’t you? Doylem’s
Charlene: yeah
Natalie: = a it’s I dunno if it’s a a Berwick word.
Charlene: ºI don’t think it is.º

Appendix 2 summarises the functional categories identified in the BwE data for
bound i don’t know and those identified for utterances containing i don’t know.
Note that the latter are not included in Section 4.5 as factors in the factor group
function since the analysis is concerned with the construction i don’t know per
se, not with utterances containing it. Depending on their meaning, tokens of bound
i don’t know such as those in (25) were therefore recoded as referential or epis-
temic tokens for the quantitative analysis. As with unbound tokens, the boundary
between referential and pragmatic uses of bound i don’t know is not sharp but
fluid. One of the main cues differentiating the two meanings is whether i don’t
know is accented and carries full stress.

4.4.3 Functions performed by i don’t know with dependent wh-words

Finally, tokens of i don’t know which are followed by a dependent wh-word con-
stitute a syntactic, pragmatic and intonational unit. The data contain no tokens of
i don’t know wh-word which function referentially. Epistemicity is their prime
function in the BwE corpus. All tokens are encoded with high pitch and they signal
either local epistemicity, as in (26a), or global epistemicity, as in (26b).
(26) a. Leah: <It’s like, @ they think you’re all psycho. @ I don’t know why. Cos
most of them probably haven’t even been to Berwick.
b. HP: Yeah, how does it how is it [portrayed.]
Albert: [+ It al-] it always seems to come
across very well. Em. I don’t know why? Maybe cos cos
HP: yeah
Albert: = you live here and you relate to it, you know.

Like utterances containing unbound i don’t know, the whole unit of i don’t
know wh-word can fulfil textual functions such as topic-decline (27a) and turn-
closure (27b).
(27) a. HP: Would you say the way you speak is more Scots or more Geordie?
Adam: Me, personally, I would say more Scots.
HP: Yeah because of your (..)
Adam: Just (.) I divn’t knaa why. @
98 The Structure of Discourse-Pragmatic Variation

b. Adam: Having said that, when (.) like the World Cup was on and that.
(h) I was still supporting Ireland.
HP: [mhm]
Keith: [Oh!] (?)
Adam: == [[@]]
HP: == [[mhm]] mhm
Adam: But no. (..) Wouldn’t support Wales. And divn’t knaa why.
HP: mhm

Appendix 3 summarises the functional categories identified for i don’t know

wh-word. Because the interest of this investigation is in the construction i don’t
know per se rather than the utterances containing it, tokens such as those in (27)
were coded as belonging to the epistemic category.

4.4.4 Summary of functions

The preceding sections have provided detailed overviews and illustrations of

the broad spectrum of uses i don’t know is put to in the BwE interview data.
Unbound i don’t know has been shown to perform a wide range of functions
beyond signalling insufficient knowledge. In the interpersonal domain, it serves
to signal speakers’ epistemic stance vis-à-vis their propositions and word choice,
and works towards mitigating interactional conflict by attenuating the illocution-
ary force of speech acts and by allowing speakers to disengage from controversial
topics. It thus frequently serves a face-saving function in discourse. In the textual
domain, unbound i don’t know contributes to the smooth development and over-
all coherence of discourse. Apart from serving as a repair device that bridges the
gap between repaired and repairing elements, unbound i don’t know functions
as a topic-development device which allows speakers to exert some control over
the topical development of the interview. Also, i don’t know plays an important
role in turn-exchange, serving to launch, hold, yield and close turns. Lastly, some
instances of unbound i don’t know function across communicative domains as
textual devices with epistemic and mitigating effects. These uses illustrate well the
intrinsic multifunctionality of discourse-pragmatic features, thus providing empir-
ical evidence for the view expressed in Chapter 1.2: that Fraser’s (1990) typological
distinction between ‘discourse markers’ and ‘pragmatic particles’ is untenable (see
also Chapter 5.4). The functional range of bound i don’t know and i don’t know
wh-word is far more limited than that of unbound tokens of i don’t know. In the
current dataset, these tokens are used exclusively to express lack of knowledge and
to signal epistemicity.
 Chapter 4.  The construction i don’t know 99

Because the preceding analysis was based on data from one speech event (in-
terviews) and one dialect (BwE), the functional inventory outlined above cannot
claim to be comprehensive of all uses of i don’t know across all text types and
all varieties of English. Discourse-pragmatic features regularly perform different
functions across text types (e.g. Lam 2009) and varieties (e.g. Hoffmann et al.
forthc.). Future studies will modify and elaborate the above account to incorpo-
rate any alternative or additional functions which i don’t know may perform
in other speech events or varieties. Yet despite the aforementioned limitations,
the analysis presented above constitutes an important contribution to current
descriptions of i don’t know. It has incorporated in its remit all tokens of i don’t
know in all sequential environments and syntactic configurations, thus provid-
ing the first full account of the construction’s functionality in a corpus of British
English interview data.
The main aim of the qualitative data analysis was to develop a coding schema
which details the factors typifying the factor group function in the BwE interview
data and enables us to include function as an independent variable in the quantita-
tive data analysis. Table 4.2 outlines the coding schema which arose from the quali-
tative analysis outlined in Sections 4.4.1 to 4.4.3. Following this coding protocol,
each token of the variable i don’t know in the data was allocated to the functional
category which best described its use. As pointed out previously, the individual cat-
egories are not discretely distinct from each other. They should be conceptualised
as a network of meanings which are related by virtue of retaining nuances of the
construction’s source meaning even in their most bleached functions.

Table 4.2  Coding schema for i don’t know functions

unbound bound wh-word
ideational referential marker of referential marker of
mode insufficient knowledge insufficient knowledge
interpersonal epistemic marker epistemic marker epistemic marker
mode mitigation device
textual mode repair marker
topic-development
device (decline, curtail,
close, discard)
turn-exchange device
(take, hold, yield, close)
interpers.-text. epistemic repair marker
mode epistemic/mitigating
turn-exchange device
100 The Structure of Discourse-Pragmatic Variation

4.5 Quantitative analysis of i don’t know

Having identified the variants of i don’t know, the factor groups that might af-
fect their distribution as well as the factors that constitute these factor groups, this
section presents the results of the quantitative analysis into the formal, functional
and social variability of i don’t know usage in the BwE data. An outline of the
overall distribution of variants in the corpus, across individuals and across social
cohorts is followed by a short description of how the variable and its variants are
distributed across functional categories and syntactic configurations. These distri-
butions determine how the data are configured for the mixed-effects multiple re-
gression analyses. They model the simultaneous effect of the independent variables
on the observed patterns of formal variability in the data whilst accounting for
individual levels of variance. Together with detailed investigation of the variable’s
and its variants’ frequencies across individual speakers, close scrutiny of the results
that emerge from the apparent-time regression analyses will reveal the organisa-
tion, robustness and stability of the variable grammar underlying the distribution
of variants. This will afford important new insights into the distribution of non-
localised and localised variants of i don’t know, the nature and strength of any
form-function correlations in the data as well as the sociolinguistic mechanisms
of discourse-pragmatic variation and change more generally.

4.5.1 Distributional analysis

The dataset available for analysis consists of 600 tokens of i don’t know which,
following close auditory analysis, were categorised into five formal variants (see
Section 4.3.2 above). The frequency of these variants is given in Figure 4.1. I dunno
is proportionally dominant. All the other variants combined occur less often than
I dunno. I divn’t knaa is the second and I don’t know the third most frequent variant
in the data, constituting roughly a fifth and a sixth of all tokens respectively. I dono
is comparatively infrequent, accounting for less than ten per cent of all occurrences
of the variable. Even though it is no less functionally versatile than other variants,
I dinnae ken is only negligibly instantiated in the data. Due to its paucity, this vari-
ant is excluded from further analysis.
Initial data runs reveal that the remaining variants are not evenly distributed
across the 35 speakers in the sample who use the variable i don’t know. Firstly,
one speaker categorically uses the variant I dunno (N = 5). Following standard vari-
ationist practice, data from this speaker are removed from the ensuing analysis.
Secondly, while all of the 34 variable speakers employ non-localised variants of
i don’t know, albeit with differential frequencies, only nineteen of these speakers
 Chapter 4.  The construction i don’t know 101

I dono I dinnae ken

9% 1%
(N = 55) (N = 6)

I don’t know
17%
(N = 101)

I dunno
52%
(N = 312)

I divn’t knaa
21%
(N = 126)

Figure 4.1  Overall distribution of the variants of i don’t know

employ the localised variant I divn’t knaa. To avoid skewing the results and con-
cealing important constraints on observed variation patterns, only speakers who
employ I divn’t knaa are included in the quantification of this variant across in-
dependent variables.27 Thirdly, the social cohorts represented in the corpus data
differ in the frequency with which they use non-localised and localised variants
of i don’t know. The black bars along the y-axis in Figure 4.2 show that I dunno
is the most frequent variant in most of the social cohorts listed along the x-axis.
Older and younger males’ lower than average rate of I dunno (black bars) is coun-
terbalanced by their proportionally higher rates of I don’t know (dark grey bars)
and I divn’t knaa (light grey bars) respectively. Speakers from the middle age group,
in particular males, seem to boast the highest proportion of I dono (white bars).
However, some of these distributions must be interpreted cautiously. As shown
in Section 4.5.2 which explores in more detail the precise nature and statistical
significance of social and linguistic constraints on variant choice, the overall as-
sociation in Figure 4.2 of I dono with middle (male) speakers, of I don’t know with

27. As a result, the percentages provided for I divn’t knaa in Figures 4.2 and 4.3 below are derived
from consideration of a subsample of the i don’t know token numbers given next to the factors
listed along the x-axis.
 80
70%
67% (86)
70 (58)

60 52%
50% 49%
(59)
(23) (32)
44% 43%
50 42% 41% (53) (15) don’t know
(30) (29)
dono
40
30% 29% dunno
28%
(20) (6) 26% divn’t knaa
23% (20)

Percentage
30 (10)
(26) 19%
102 The Structure of Discourse-Pragmatic Variation

(23)
20 13% 12%
10% 9% (6)
7% (8)
(7) (5)
(8) 3% 4% 4% 3%
10
(4) (5) (2) (4)
0
old (N = 71) middle (N = 113) young (N = 120) old (N = 46) middle (N = 66) young (N = 123)
male female

Figure 4.2  Distribution of variants of i don’t know across speaker sex and age
(Figures in round brackets indicate raw token numbers. The results for I divn’t knaa are derived from data produced
by divn’t knaa-users only [N = 305].)
 Chapter 4.  The construction i don’t know 103

middle female ­speakers and of I dunno with young female speakers is created by
some individuals’ disproportionate use of the variable or these variants. Even so,
Figure 4.2 provides some preliminary evidence that the variation between variants
of i don’t know is constrained by social factors.
The distributional analysis in Figure 4.2 is based on a reduced dataset of 539
tokens. It does not include tokens of I dinnae ken (N = 6), tokens produced by an
invariable speaker (N = 5), tokens not coded for syntactic configuration and/or
function due to insufficient context (inaudible turn continuation, interruption,
truncation) (N = 20), and tokens followed by a dependent wh-word (N = 30).28
As shown in Table 4.3, tokens of i don’t know with a dependent wh-word are
liminal, accounting for only five per cent of all tokens of the variable in the data.
Because of these tokens’ marginal occurrence, socially skewed distribution (young
speakers produced three quarters of tokens) and functional invariability (all tokens
signal epistemicity), the quantitative analysis focuses on unbound and bound to-
kens of i don’t know. These are more frequently instantiated in the corpus, more
evenly distributed across social cohorts, and more functionally versatile.29

Table 4.3  Distribution of i don’t know across syntactic configurations (not including
tokens of I dinnae ken, tokens produced by an invariable speaker, and tokens not coded
for syntactic configuration and/or function)
% N
unbound 65.4 372
bound 29.3 167
wh-word   5.3   30
569

The analysis in Section 4.4 revealed that bound tokens of i don’t know perform
a smaller range of functions than unbound tokens which, in addition to signalling
insufficient knowledge and epistemicity, also function to structure and control dis-
course. The percentages for the functional categories along the x-axis in Figure 4.3
reveal that interpersonal uses account for the vast majority of instances of bound

28. Inclusion of the latter sets of tokens in the distributional analysis slightly alters some percent-
ages but does not affect the overall pattern of variant distribution outlined in Figure 4.2. While
Goldvarb has a slash-function which removes tokens not coded for one particular factor group
from consideration in this factor group only, Rbrul does not offer this option. Consequently,
tokens not coded for one factor group must be removed from consideration in all factor groups.
29. All of the 30 tokens of i don’t know with a dependent wh-word function epistemically. 43%
of them were realised as I dunno and 23% as I don’t know, suggesting that these tokens exhibit
similar form-function correlations as bound and unbound tokens (see Figure 4.3).
 64% 65%
70 62% 61% (153)
(18) 60%
(34) (33) (6)
60
50%
(9) 46%
50 (69)
38% 37%
37%
33% 33% (10) (28) don’t know
40 (55)
(10) (5) dono
dunno
30 divn’t knaa

Percentage
19%
17% 16% (28)
20 (3) (24)
11%
104 The Structure of Discourse-Pragmatic Variation

11% 11%
9% 9%
(6) (3) 7% (6)
(5) (5) 6% 5%
10 (2) (15) (12)
0%
(0)
0
referential interpersonal interp.-text. textual referential interpersonal
(15%, N = 55) (7%, N = 28) (15%, N = 54) (63%, N = 235) (11%, N = 18) (89%, N = 149)
UNBOUND TOKENS (N = 372) BOUND TOKENS (N = 167)

Figure 4.3  Distribution of variants of i don’t know across syntactic contexts and functional categories
(Figures in round brackets indicate raw token numbers. The results for I divn’t knaa are derived from data produced
by divn’t knaa-users [N = 305].)
 Chapter 4.  The construction i don’t know 105

i don’t know (on the right). This stands in stark contrast to unbound tokens (on
the left) which exhibit a propensity for textual uses. Neither unbound nor bound
tokens manifest a high concentration of referential uses.30 These overall functional
distributions are fairly consistently repeated across social cohorts, as shown in
Tables 4.4 and 4.5. It is only the relative frequency of unbound non-textual tokens
which differentiates social cohorts. Speakers from the oldest age group and female
speakers from the middle age group have comparatively high rates of referential
tokens; male speakers from the middle age group have higher rates than other
cohorts of interpersonal tokens; and speakers from the youngest age group have
higher rates of interpersonal-textual tokens than they do of referential or inter-
personal tokens. These patterns are in line with previous reports of age- and sex-
differentiated functional uses of discourse-pragmatic variables (e.g. Erman 1992;
Stubbe & Holmes 1995).

Table 4.4  Frequency in per cent of unbound i don’t know across functional categories
by age and sex (Figures in brackets indicate raw token numbers.)
old middle young
male female male female male female
referential 30 (11) 29 (8)   5 (4) 26 (11)   8 (8) 14 (13)
interpersonal   5 (2)   7 (2) 13 (9)   5 (2)   6 (6)   8 (7)
interp.-text. 16 (6) 14 (4)   5 (4) 14 (6) 20 (19) 16 (15)
textual 49 (18) 50 (14) 77 (57) 55 (23) 66 (65) 62 (58)
TOTAL N 37 28 74 42 98 93

Table 4.5  Frequency in per cent of bound i don’t know across functional categories
by age and sex (Figures in brackets indicate raw token numbers.)
old middle young
male female male female male female
referential 12 (4)   6 (1)   8 (3) 12 (3)   5 (1) 20 (6)
interpersonal 88 (30) 94 (17) 92 (36) 87 (21) 96 (21) 80 (24)
TOTAL N 34 18 39 24 22 30

30. Of the unbound textual tokens, the majority signal turn-exchange. Topic-development is the
second-most frequent textual function, followed by repair marking. Of the bound interpersonal
tokens, the vast majority have local epistemic scope.
106 The Structure of Discourse-Pragmatic Variation

Returning to Figure 4.3, the bars along the y-axis indicate the frequency of formal
variants across the functional categories and syntactic contexts listed along the
x-axis.31 The differential frequency of full and reduced variants across function in
both unbound and bound contexts is strongly reminiscent of the results reported
in Bybee & Scheibman (1999) and Scheibman (2000): I don’t know (dark grey bars)
is tightly linked to referential uses; I dunno (black bars) is consistently associated
with pragmatic uses. In contrast to I don’t know and I dunno, the semi-reduced
variant I dono (white bars) and the localised variant I divn’t knaa (light grey bars)
are differently distributed across unbound and bound contexts. In the former, they
occur with similar frequency across functional categories; in the latter, I dono is
used exclusively to serve interpersonal functions while I divn’t knaa shows a slight
propensity to signal insufficient knowledge. Taken together, the patterns in Figure
4.3 demonstrate that in addition to the extra-linguistic variables discussed earlier,
function and, to a lesser degree, syntax are implicated in the occurrence of different
variants of i don’t know in BwE.
Prior to subjecting these observations to statistical testing, it is necessary to
establish the degree of inter-speaker variation in the frequency of use of i don’t
know. Figures 4.4 and 4.5 plot the frequency of referential and pragmatic uses of i
don’t know across individual speakers in the sample. Individuals are arranged by
age on the x-axis, from youngest on the left to oldest on the right. The bullets and
squares along the y-axis show the normalised frequency per 10,000 words with
which each of the 18 male speakers in the sample (black bullets) and each of the
18 female speakers in the sample (grey squares) use i don’t know for referential
or pragmatic uses. (Pragmatic uses include all tokens that perform interpersonal,
textual or a combination of these functions.) The closer a bullet or square is lo-
cated to the x-axis, the lower a speaker’s frequency of use of i don’t know; the
more distant it is from the x-axis, the higher a speaker’s frequency of use of i
don’t know. Figure 4.4 shows that, with the exception of three female speakers
who use referential i don’t know with a rate of 13–15 tokens per 10,000 words,
the vast majority of speakers use referential i don’t know with a frequency of
0–4 tokens per 10,000 words. This is particularly true of male speakers. The linear

31. Within the broad functional categories displayed in Figure 4.3, variants are similarly dis-
tributed across the fine-grained functions identified for i don’t know in Section 4.4. This is
not to imply that the detailed functional analysis in Section 4.4 was unneeded. Previous studies
(e.g. Plug 2010) as well as the analysis of i don’t think in Chapter 5 demonstrate that micro-
level functional distinctions may be key to interpreting formal variation patterns in discourse-
pragmatics. While this turns out not to be the case for i don’t know in BwE, the validity of
the results reported in this chapter cannot be ascertained unless this possibility is ruled out at
the outset.
 Chapter 4.  The construction i don’t know 107

trend-lines for males and females further indicate that despite individual levels
of variance, the average frequency of use of referential i don’t know is fairly
stable in apparent time. This is not the case with pragmatic i don’t know. The
linear trend-lines in Figure 4.5 indicate a general tendency for rates of pragmatic
i don’t know to rise in apparent time, both amongst males and females. While
the vast majority of speakers, especially those in the middle and old age groups,
use pragmatic i don’t know with rates lower than 30 tokens per 10,000 words,
some young male speakers far exceed this rate. Consequently, individual rates
of use of pragmatic i don’t know are highly variable, ranging from 0 up to 69
tokens per 10,000 words. To avoid skewing the results in Section 4.5.2 below, it
is necessary to account for this high degree of inter-speaker variability in the
multivariate analyses.

15
Frequency/10,000 words

10 male
female
Linear (male)

5 Linear (female)

0
15 25 35 45 55 65 75 85
speaker age

Figure 4.4  Normalised frequencies of referential uses of i don’t know across individuals

80
70
Frequency/10,000 words

60
50 male
female
40
Linear (male)
30 Linear (female)
20
10
0
15 25 35 45 55 65 75 85
speaker age

Figure 4.5  Normalised frequencies of pragmatic uses of i don’t know across individuals
108 The Structure of Discourse-Pragmatic Variation

4.5.2 Multivariate analysis

In this section, mixed-effects multiple regression analyses are conducted with

Rbrul (see Chapter 2.3.3 for details on the technicalities and practicalities of
Rbrul-based analyses). They will establish the relative contribution and com-
bined impact of the independent variables discussed above to the occurrence of
the variants of i don’t know while consistently accounting for the high levels
of individual variance in i don’t know usage shown in Figures 4.4 and 4.5. For
the more frequently instantiated variants in the corpus – I don’t know, I dunno,
I divn’t knaa – independent multivariate analyses are carried out for each of the
three age groups represented in the corpus. Identical apparent-time analyses for
these variants and close inspection of the direction of effect, significance and
relative contribution of predictors enable us to assess the stability of the vari-
ants’ underlying grammars across age groups and to compare the organisation of
variability across variants. Because I dono does not occur at sufficient frequency
thresholds for viable apparent-time statistical modelling (see footnote 32 below),
its distribution is examined across all age groups combined. While this proce-
dure does not afford an apparent-time window on the variant’s sociolinguistic
conditioning, it still affords insights into the underlying grammar of I dono in
the sample as a whole. Adverse effects associated with badly distributed data are
mitigated by collapsing the three pragmatic factors identified for unbound tokens
(interpersonal, interpersonal-textual, textual) into one factor pragmatic which
together with the factor referential constitutes the factor group function. Also,
unbound and bound tokens are included in the same runs, albeit differentiated as
two factors constituting the factor group syntax. While these configurations may
obscure the differential patterning of I don’t know and I dono in non-referential
uses across unbound and bound contexts (see Figure 4.3), they maximise cell
sizes and thus improve the overall reliability of the results (Guy 1980). In addi-
tion, application of mixed-effects modelling and inclusion of individual speaker
as a random effect in the regressions reduces the risk of providing inaccurate
significance estimates of social factors in the unbalanced dataset under scrutiny
here. Readers unfamiliar with the output of multivariate analyses are referred
to Chapter 2.3.3, which provides a detailed explanation of how to interpret the
quantitative results shown in Tables 4.6 to 4.9.
Table 4.6 displays the results of three independent multivariate analyses of
the contribution of sex, function and syntax to the choice of I don’t know for
the three age groups in the corpus. The range values indicate that in all age
groups it is function that makes the most important contribution to the occur-
rence of this variant. The associated factor values confirm that I don’t know is
strongly favoured for referential uses and strongly disfavoured for pragmatic


Table 4.6  Contribution of external and internal factors to the probability of I don’t know
(with I dunno, I dono and I divn’t knaa as non-application values)
I don’t know
OLD MIDDLE YOUNG
input prob. 0.324 0.303 0.156
total N 117 179 243
deviance 102.519 108.352 119.997
factor % N log odds factor % N log odds factor % N log odds
weights weights weights
sex p = 0.0393 not significant p = 0.0397
female .232 13.0 46 −1.197 [.672] 30.3   66 −[0.718] .739 18.7 123 −1.043
male .768 42.3 71 −1.197 [.328]   7.1 113 [−0.718] .261   3.3 120 −1.043
range .536 .478
function p = 2.16e-05 p = 1.74e-07 p = 1.93e-08
referential .806 66.7 24 −1.424 .837 66.7   21 −1.639 .853 46.4   28 −1.756
pragmatic .194 21.5 93 −1.424 .163   8.9 158 −1.639 .147   6.5 215 −1.756
range .612 .674 .706
syntax not significant p = 0.0191 not significant
unbound [.542] 33.8 65 −[0.167] .342 12.1 116 −0.656 [.541] 11.5 191 −[0.163]
bound [.458] 26.9 52 [−0.167] .658 22.2   63 −0.656 [.459]   9.6   52 [−0.163]
range .316
speaker random st. dv 1.418 random st. dv 0.93 random st. dv 1.187
Chapter 4.  The construction i don’t know 109
110 The Structure of Discourse-Pragmatic Variation

uses.32 These form-function correlations exhibit remarkable apparent-time sta-

bility, as indicated by the consistency across age groups in statistical significance
and constraint hierarchies. In contrast to the relative stability in the functional
conditioning of I don’t know, its social conditioning is less stable. The input prob-
abilities given at the top of the table demonstrate a slight decrease in the overall
frequency of use of I don’t know in the middle age group which is accelerated in
the youngest age group. Moreover, the sex effect for I don’t know reverses across
age. While use of the variant is strongly favoured by male speakers in the oldest
age cohort, in the youngest cohort its use is strongly favoured by female speak-
ers. In the middle cohort, the contribution of sex to the occurrence of I don’t
know does not reach statistical significance despite much higher factor weights
for females than males. (This is due to individual levels of variance and token
imbalance among female speakers in this age group.) In contrast to speaker sex,
which exerts a significant effect to the occurrence of I don’t know in the old and
young but not the middle cohort, syntax exerts a significant effect to the occur-
rence of I don’t know in the middle but not in the old and young cohorts. Among
the latter cohorts, I don’t know is weakly preferred in unbound contexts; among
the former cohort, it is favoured in bound contexts. Considering the consistently
strong association of I don’t know with referential uses, the syntactic distribu-
tion of I don’t know across age suggests that speakers from the middle age group
are more likely than speakers from other age groups to specify through overt
complementation what it is they lack knowledge of.
The variable grammar underlying the occurrence of I dunno is remarkably
constant in apparent time. The results of the three independent multivariate analy-
ses for this variant in Table 4.7 show that both internal constraints operate in
the same direction across all age groups: I dunno is consistently associated with
pragmatic uses and unbound contexts. Parallel to I don’t know, function is the
most explanatory factor group for I dunno by far, as revealed by the high range val-
ues. The consistently high factor values for pragmatic uses confirm the functional
asymmetry between I don’t know and I dunno noted earlier: while the former is
largely reserved for referential uses, the latter is largely reserved for pragmatic uses.
The second strongest constraint on the occurrence of I dunno is syntax. Although
this effect only reaches statistical significance in the middle and young age groups,
unbound contexts consistently have higher rates of I dunno than bound contexts.

32. According to general statistical laws, the threshold for obtaining reliable results is 30 tokens
per cell. However, Guy (1980) points out that acceptable levels of accuracy can be obtained with
numbers in excess of ten tokens. Consequently, we can be reasonably confident that the results
for the factor referential, which contains between 21 and 28 tokens in each age group, are not
greatly skewed by random perturbations in the data.


Table 4.7  Contribution of external and internal factors to the probability of I dunno
(with I don’t know, I dono and I divn’t knaa as non-application values)
I dunno
OLD MIDDLE YOUNG
input prob. 0.174 0.173 0.308
total N 117 179 243
deviance 118.927 215.584 251.617
factor % N log odds factor % N log odds factor % N log odds
weights weights weights
sex not significant not significant not significant
female [.538] 50.0 46 −[0.153] [.534] 48.5   66 − [0.135] [.605] 69.9 123 −[0.425]
male [.462] 40.8 71 [−0.153] [.466] 52.2 113 [−0.135] [.395] 44.2 120 [−0.425]
range
function p = 0.000402 p = 3.7e-07 p = 5.5e-08
referential .201   8.3 24 −1.381 .152   4.8   21 −1.718 .189 21.4   28 −1.459
pragmatic .799 53.8 93 −1.381 .848 57.0 158 −1.718 .811 61.9 215 −1.459
range .598 .696 .622
syntax not significant p = 0.0191 p = 0.0247
unbound [.562] 46.2 65 −[0.248] .619 57.8 116 −0.484 .614 59.2 191 −0.463
bound [.438] 42.3 52 [−0.248] .381 38.1   63 −0.484 .386 50.0   52 −0.463
range .238 .228
speaker random st. dv 1.655 random st. dv 0 random st. dv 1.483
Chapter 4.  The construction i don’t know
111
112 The Structure of Discourse-Pragmatic Variation

Speaker sex does not make a significant contribution to the occurrence of I dunno
in any of the age groups. In the old age group I dunno is weakly associated with
female speakers; in the middle age group it is weakly associated with male speakers.
(The interaction effect in the middle age group is caused by the fact that females
in this age group never use I dunno referentially.) The appearance of a strong sex
effect in the youngest age group (see the divergent factor values) is created by
within-group variation. As indicated by the lack of statistical significance, however,
there is in fact little evidence of a real sex effect in this age group. Lastly, the input
probabilities at the top of Table 4.7 demonstrate a sudden increase in the use of
I dunno in the youngest age group which, considering the strong association of
I dunno with pragmatic uses, may be linked to the apparent-time rise of pragmatic
uses of i don’t know noted in Figure 4.5.
Moving on to Table 4.8 which shows the contribution of contextual factors
to the probability of the non-localised semi-reduced variant I dono, the results
confirm that this variant occurs far less often in the data than either of the non-
localised variants discussed above (see the low input probability given at the top
of the table). Also, I dono is affected differently by function than I don’t know and
I dunno. Unlike the use of full and reduced variants, the use of the semi-reduced
variant is not correlated with a functional category but occurs with roughly equal
probability across referential and pragmatic uses. However, this distribution pat-
tern arises from viewing the data in the aggregate. As shown earlier in Figure 4.3,
I dono is never used referentially in bound contexts and has slightly higher rates of
pragmatic compared to referential uses in unbound contexts (9% vs. 6%). External
factors make no significant contribution to the occurrence of I dono either, despite
relatively divergent factor weights especially for age. The appearance of social ef-
fects is produced by individual speaker variation. While age, sex and function
have no or little effect on the use of I dono when individual variance is considered,
syntax makes a significant contribution to the occurrence of this variant. Unlike
the reduced variant I dunno, the semi-reduced variant I dono is favoured in bound
contexts. Cross-tabulations with age (not shown here) demonstrate that this ef-
fect as well as that of function and speaker sex remain relatively constant across
age groups, suggesting that I dono resembles the other non-localised variants in
exhibiting relatively stable variation patterns.
Finally, Table 4.9 displays the results of three independent multivariate analy-
ses for the localised variant I divn’t knaa which, as outlined earlier, is only used
by a subsample of speakers in the data. Unlike with any other variant of i don’t
know in the BwE data, speaker sex has the strongest effect on the occurrence of
I divn’t knaa. However, the direction of effect is inconsistent in apparent time. In
the oldest age group, the variant is strongly favoured by females. This correlation
weakens in the middle age group to the extent that the female preference for I
 Chapter 4.  The construction i don’t know 113

Table 4.8  Contribution of external and internal factors to the probability of I dono
(with I don’t know, I dunno and I divn’t knaa as non-application values)
I dono
input prob. 0.06
total N 539
deviance 298.66
factor % N log odds
weights
age not significant
old [.450]   7.7 117 [−0.202]
middle [.695] 19.0 179 −[0.824]
young [.349]   3.7 243 [−0.621]
range
sex not significant
female [.429]   6.0 235 [−0.284]
male [.571] 12.5 304 −[0.284]
range
function not significant
referential [.478]   6.8   73 [−0.087]
pragmatic [.522] 10.1 466 −[0.087]
range
syntax p = 0.0357
unbound .412   6.5 372 −0.356
bound .588 16.8 167 −0.356
range .176
speaker random st. dv 1.097

divn’t knaa is no longer significant. In the youngest age group, the sex effect is re-
versed and regains significance. Here it is males, not females, who exhibit a strong
favouring effect for I divn’t knaa. When the data are reconfigured to combine age
and sex into one social factor group with six factors (old male, old female, middle
male, middle female, young male, young female), analyses not shown here reveal
that the favouring effect is stronger among young males than it is among older
females (factor weights of .858 vs. .606). This result may account for the regular
increase across age groups in the overall probability of I divn’t knaa. The input
probabilities show that the likelihood for I divn’t knaa to occur in the BwE data
more than doubles between the oldest and youngest generation. In stark contrast to
the social constraints on the choice of I divn’t knaa, the internal constraints do not
generally fluctuate across age groups nor do they ever reach statistical significance.
I divn’t knaa shares with I dono a weak preference to occur in bound rather than
Table 4.9  Contribution of external and internal factors to the probability of I divn’t knaa
(with I don’t know, I dunno and I dono as non-application values)
I divn’t knaa
OLD MIDDLE YOUNG
input prob. 0.22 0.357 0.473
total N 88 92 125
deviance 79.56 104.926 151.036
factor % N log odds factor % N log odds factor %  N log odds
weights weights weights
sex p = 0.0171 not significant p = 0.0147
female .739 42.9 35 −1.042 [.523] 28.6 21 −[0.093] .296 25.6   39 −0.997
male .261   9.4 53 −1.042 [.477] 28.2 71 [−0.093] .731 67.4   86 −0.997
114 The Structure of Discourse-Pragmatic Variation

range .478 .435

function not significant not significant not significant
referential [.523] 29.4 17 −[0.093] [.505] 42.9  7 −[0.020] [.580] 50.0   16 −[0.321]
pragmatic [.477] 21.1 71 [−0.093] [.495] 27.1 85 [−0.020] [.420] 55.0 109 [−0.321]
range
syntax not significant not significant not significant
unbound [.411] 19.6 51 [−0.358] [.401] 25.0 72 [−0.402] [.409] 53.6   97 [−0.369]
bound [.589] 27.0 37 −[0.358] [.599] 40.0 20 −[0.402] [.591] 57.1   28 −[0.369]
range
speaker random st. dv 0.625 random st. dv 0.948 random st. dv 0.451
 Chapter 4.  The construction i don’t know 115

unbound contexts as well as a tendency to occur with roughly equal probability

across pragmatic and referential uses. While the low token numbers constituting
the factor referential (see in particular the middle age group) warn that the results
for function must be interpreted with caution (see footnote 32 above), an identical
multivariate regression analysis of the data from all age groups combined confirms
their robustness and reliability by producing factor weights of [.540] for referential
and [.460] for pragmatic tokens. Yet, as reflected in the antithesis between the
percentages and the factor weights for function in the youngest age group, the
marginal preference for referential uses is not shared by all speakers in the young
age cohort. Hidden cross-tabulations with gender reveal that young males exhibit
a very slight propensity for pragmatic uses of I divn’t knaa.

4.5.3 Summary of results

Identical analyses for I don’t know, I dunno and I divn’t knaa across the three age
groups represented in the data and a corresponding analysis of I dono with all
age groups combined make it possible to compare the mechanisms of variability
across variants. The comparisons yield the following key insights into the formal
variability of i don’t know in BwE:
– Different variants are exploited to signal different meanings: The most fre-
quent non-localised variants in the data, I don’t know and I dunno, occupy
asymmetrical functional niches in the system of i don’t know variants. The
former is strongly associated with referential meanings; the latter is strongly
associated with pragmatic meanings. In stark contrast, the marginally occur-
ring non-localised­ variant I dono and the localised variant I divn’t knaa are
multi-purpose variants which are used in the BwE data to express insufficient
knowledge just as much as to signal pragmatic meanings.
– The occurrence of non-localised and localised variants of i don’t know is condi-
tioned by different parameters: At least one internal factor always has primacy
over social factors in constraining the occurrence of the non-localised variants,
I don’t know, I dono and I dunno. By contrast, external factors have primacy in
conditioning the occurrence of the localised variant I divn’t knaa.
– Internal and external conditioning factors do not operate in unison: The internal
conditioning of variant choice, in particular by function, exhibits remarkable
apparent-time stability for all variants analysed above. Conversely, the social
conditioning of some variants, notably of I don’t know and I divn’t knaa by
speaker sex, is subject to rigorous change in the short time-span covered by
the BwE corpus.
116 The Structure of Discourse-Pragmatic Variation

Inspection of social and individual patterns of variant distribution, inclusion of

individual speaker as a random effect in the multivariate analyses, and tabulation
of the frequency of i don’t know across and within social cohorts yield another
key insight into the distribution of i don’t know in BwE:
– The frequency, usage and form of i don’t know exhibit a fair amount of inter-
group and inter-speaker variability: Social cohorts and individual speakers vary
in the frequency with which they use the construction i don’t know, the
functional uses to which they put it as well as their preference for individual
variants.

4.6 Discussion

This chapter set out to account for the high degree of variation in the realisation of
i don’t know in BwE, a peripheral variety of British English which features both
non-localised and localised variants of the targeted variable. Systematic qualita-
tive and quantitative data analyses in Sections 4.4 and 4.5 have provided a de-
tailed overview of the construction’s functionality in the BwE interview data and
established the effect of extra- and intra-linguistic predictors on variant selection.
This section discusses the key findings of the preceding analyses, situates them
vis-á-vis previous studies of i don’t know in other contexts and varieties, and
explores the diachronic and social mechanisms underlying observed patterns of
synchronic variability.
The in-depth qualitative data analysis in Section 4.4 served to establish the
functions performed by i don’t know in the BwE data with the aim of making
possible inclusion of function as a factor group in the quantitative data analysis. By
way of establishing the factors constituting the factor group function, the qualita-
tive analysis has enhanced previous descriptions of the functional repertoire of
i don’t know in several important ways. Firstly, by including in its scope all
tokens of i don’t know in the data irrespective of their positioning within ut-
terances, turns or sequences, this study has provided a comprehensive account
of the complete functional inventory of i don’t know in the current dataset. It
thus advances previous accounts which are largely based on subsets of tokens
and yield potentially incomplete or biased descriptions of the construction’s usage
in social interaction. Secondly, by differentiating the syntactic configurations in
which i don’t know occurs, this analysis has established that unbound tokens of
i don’t know are more functionally versatile than bound tokens or tokens with a
dependent wh-word whose usage is limited to signalling insufficient knowledge
and epistemicity. It thus confirms that the functionality of discourse-pragmatic
features is intricately linked to their syntactic integration and positional mobility
 Chapter 4.  The construction i don’t know 117

(see Brinton 2006). Thirdly, by establishing the frequency with which different
functions of i don’t know occur in the data, this analysis has provided numerical
evidence for observations in the literature about the construction’s great impor-
tance in qualifying propositions and structuring discourse and its negligible role
in signalling a cognitive state. However, the overall predominance of textual over
interpersonal uses revealed in Figure 4.2 may not generalise beyond the current
data. Unlike in everyday casual conversations, in interviews speakers may be more
concerned with exerting some control over the course and development of the
interaction than with protecting their own and others’ face needs. The preponder-
ance in the current data of textual tokens of i don’t know may therefore reflect
the specific constraints of the interview situation rather than more widespread
functional preferences in the use of i don’t know. Notwithstanding this caveat,
the functional data analysis in Section 4.4 provides important new insights into
the usage of i don’t know in spoken interaction.
The quantitative data analysis in Section 4.5 served to establish the constraints
operating on the formal variation in the data. To this end, multivariate analyses
were run which tested the simultaneous effect of multiple contextual factors (age,
sex, function, syntax) on the choice of formal variants of i don’t know. Where
feasible, identical analyses were run for each of the three age groups represented
in the corpus in order to establish the apparent-time stability of any constraints
on variant choice. One of the overarching findings to emerge from these analy-
ses is the complementary functional distribution of I don’t know and I dunno,
which confirms the results reported in Bybee & Scheibman (1999) and Scheibman
(2000): the full variant is strongly favoured for referential uses; the reduced vari-
ant is strongly favoured for pragmatic uses. Beyond confirming that functional
distribution patterns previously only documented for American English are also
present in a peripheral variety of British English, the current analysis contributes
important new insights into the robustness and stability of these form-function
correlations (as well as others discussed below). Cross-tabulations (not shown
here) of function with other predictors vindicate the regularities shown in Tables
4.6 and 4.7. The effect of function is consistent across unbound and bound contexts
and although older male speakers are more likely than other speakers to use I don’t
know for pragmatic uses, the functional split between I don’t know and I dunno
is uniformly present in the speech of both sexes and virtually all individuals.33
The results, then, demonstrate that the functional split between I don’t know and

33. Where individuals appear not to participate in the form-function split outlined above, this
is due to the fact that the emergence of any such patterns is impeded by these individuals’ in-
frequent use of the variable overall or by their infrequent use of the variable for one of the two
broad meanings (referential vs. pragmatic).
118 The Structure of Discourse-Pragmatic Variation

I dunno is firmly entrenched in the speaker sample and that it remains unaffected
by inter-speaker and inter-group frequency fluctuations in the use of the variable
or its non-localised variants.
Bybee & Scheibman (1999) and Scheibman (2000) posit that the functional
split between full and reduced variants of i don’t know arises from the con-
struction’s gradual grammaticalisation and associated changes to its constituent
structure. Due to high frequency of use and repetition, i don’t know has become
conventionalised as a single storage and processing unit (chunking) and weakened
its association with its composite parts (autonomy). Loss of compositionality and
analysability has triggered morpho-phonological and semantic-pragmatic changes,
i.e., vowel reduction of don’t and its fusion with know as well as (inter-)subjectifi-
cation of meaning. A grammaticalisation view of i don’t know is supported by
Thompson (2002) who argues that i don’t know, together with other frequent
subject-predicate constructions, has been reanalysed from a complement-taking
matrix clause signalling insufficient knowledge (see, for example, Extract (19c)
in Section 4.4.2) to a comment clause signalling an epistemic stance towards the
content of the accompanying finite clause (see, for example, Extracts (21) and
(22) in Section 4.4.2) (see also Kearns 2007; Keevallik 2006). According to these
scholars, then, synchronic variation in the form, function and syntactic status of
i don’t know reflects the co-existence of newer (i.e. phonetically reduced, non-
compositional and decategorialised) uses of i don’t know alongside older (i.e.
phonetically full, compositionally transparent and non-decategorialised) uses of
i don’t know (see Hopper’s [1991] principle of layering).
Apparent-time comparisons of the constraint hierarchies for functional fac-
tors in the current dataset do not furnish any evidence of grammaticalisation in
progress (see in particular Tables 4.6 and 4.7). Also, observed changes in the sig-
nificance and direction of effects of syntactic factors are inconsistent with any in-
ferences of change in progress. Yet despite the absence of any statistical imprints of
ongoing change in the linguistic conditioning of variants, the data do not invalidate
hypotheses that the stable synchronic distribution of i don’t know in the BwE
data is the product of grammaticalisation processes. Several observations from the
BwE data support a grammaticalisation scenario.
Firstly, as is the case in Bybee & Scheibman’s (1999) and Thompson’s (2002)
data, i don’t know is highly variable in BwE in terms of transparency, analysabil-
ity and syntactic status. The data contain tokens of i don’t know which constitute
fully compositional, referential matrix or independent clauses (see, for example,
the extracts in (19c) and (18) in Section 4.4.2) as well as tokens which constitute
non-compositional, conventionalised comment clauses or parentheticals (see, for
example, the extracts in (21, 22) and (4, 7, 11, 12) in Section 4.4). As a compositional
 Chapter 4.  The construction i don’t know 119

word sequence which is put together from three elements (I, don’t, know), i don’t
know retains its morpho-syntactic analysability. In the BwE data, this is reflected
in the construction’s regular modification through intervening adverbs such as re-
ally or even. Conversely, as a non-compositional formula i don’t know manifests
several structural indices of unithood and freezing in BwE: its constituent structure
is rendered opaque through phonetic reduction and fusion; and its modification
through interpolation is virtually non-existent (0.4% with non-compositional uses
compared to 12% with compositional uses). These patterns are indicative of the
construction’s decreased analysability and increased autonomy, thus supporting
Bybee & Scheibman’s (1999) view that i don’t know, when used non-referentially,
tends to be accessed from memory as a single unit or chunk.
Secondly, the synchronic co-existence in the data of full and reduced variants
of i don’t know with complementary functions and differential morpho-syntactic
properties is compatible with current hypotheses about the unidirectionality and
regularity of the changes constituting grammaticalisation. These hypotheses pre-
dict developments along the following clines:
(28) phonologically long > phonologically short  (see Bybee 2003: 615–617)
(29) truth-conditional/content/non-subjective > non-truth-conditional/
procedural/(inter-)subjective  (Traugott & Dasher 2002: 40)
(30) major category > minor category  (Hopper & Traugott 2003: 107)

The OED (2010) records the first attestation of the form dunno as dating from 1842.
The first example of dunno in the OED that clearly conveys pragmatic rather than
referential meanings dates from 1938 (“He’ll begin to say to himself: ‘Well, I dunno.
P’raps I’m wrong.’”). These data suggest that the grammaticalisation changes creat-
ing the synchronic variability of i don’t know described above are not a recent
development but have been operative for some time and have begun prior to the
time-span covered by the BwE corpus. The synchronic stability of the form-mean-
ing correlations affecting I don’t know and I dunno in the BwE data may thus reflect
the fact that the variants have become fossilised in their current distribution.
Thirdly, even though the linguistic conditioning of all variants in the BwE data
is stable in apparent time and provides no evidence of ongoing grammaticalisa-
tion, the existence and distribution in the data of the variant I dono furnishes
cogent evidence of synchronic gradience in the use of i don’t know. As detailed
in Section 4.3.2, I dono shares the vowel quality in don’t of full variants and the
lack of a morpheme boundary between don’t and know of reduced variants. Its
form is therefore more opaque than that of I don’t know and less opaque than that
of I dunno. In terms of function, I dono differs from I don’t know and I dunno in
120 The Structure of Discourse-Pragmatic Variation

that it does not occupy a functional niche but is used with roughly equal prob-
ability to signal compositional and non-compositional meanings. Thus, by the very
nature of its formal and functional properties in the data, I dono seems to occupy
an intermediate stage in the formal and semantic-pragmatic development from
I don’t know to I dunno. Embracing Traugott & Trousdale’s (2010: 40) view that
“synchronic gradience is often the result of grammaticalisation” and that gram-
maticalisation typically involves a series of small-step changes, the synchronic
variation and gradience in the formal encoding and functional distribution of
non-localised variants of i don’t know can be argued to reflect: the gradual-
ness of i don’t know’s fusion, attrition and desemanticisation; and the gradual
establishment of form-function correlations over time. The relative infrequency
of I dono in the BwE data (see Figure 4.1) does not undermine the validity of this
argumentation since “not every position on a cline is likely to be equally elaborated
at any particular stage” (Hopper & Traugott 2003: 109).
Like non-localised variants, the localised variant I divn’t knaa has been affected
by the semantic-pragmatic meaning changes associated with the grammaticalisa-
tion of i don’t know. Unlike I don’t know and I dunno, however, I divn’t knaa has
not at present specialised to occupy a functional niche in the system of i don’t
know variants. Rather, it parallels I dono in being a multi-purpose variant which
occurs with roughly equal probability across referential and pragmatic uses. This
effect and that of syntax, which is also insignificant, is robust across age and sex,
confirming the apparent-time stability of linguistic constraints on variant choice
noted earlier. (Paucity of tokens precludes investigation of the variant’s robust
linguistic conditioning across individuals.) Yet despite the lack of linguistic con-
straints, the occurrence of I divn’t knaa in BwE is far from random. An important
finding to emerge from the inclusion in the variable context of non-localised and
localised variants is that their occurrence is conditioned by different parameters.
As shown in Tables 4.6 and 4.7, function is the most important conditioning factor
for the occurrence of the most frequent non-localised variants in the data, I don’t
know and I dunno; social factors play only a secondary role in conditioning their
occurrence. Conversely, the localised variant I divn’t knaa does not carry a func-
tional load; as shown in Table 4.9, its occurrence is conditioned by social factors.
Thus, while frequent non-localised variants display function-specific patterning,
the localised variant analysed here parallels the behaviour of localised phonological
variants such as those investigated by Llamas (2007b) and Milroy et al. (1994) in
displaying systematic variation across social factors. I divn’t knaa is thus a prime
candidate for marking social differentiation and indexing social group member-
ship. Close inspection of variant distribution across individuals reveals that the
speakers who exhibit the strongest propensity for using the variant I divn’t knaa are
 Chapter 4.  The construction i don’t know 121

the three young males with the highest rate of use of pragmatic i don’t know in
the sample (see the top left corner of Figure 4.5 above). This pattern suggests that
these young males cultivate the frequency and formal variability of i don’t know
for its pragmatic and socially symbolic functions. The results also bolster evidence
provided by recent studies of general extenders which have shown that the social
indexicality of discourse-pragmatic features is not limited to their frequency and
functionality but extends to their form (Cheshire 2007; Pichler & Levey 2011;
Tagliamonte & Denis 2010).
A final key insight offered by careful consideration of both external and in-
ternal factors concerns their independence in discourse-pragmatic variation and
change. As discussed above, the contribution of internal factors to variant choice
is remarkably stable in apparent time. Yet the data exhibit strong evidence of insta-
bility in the social conditioning of variant selection. The analyses in Section 4.5.2
have shown that the variant I dunno almost doubles in frequency in the youngest
age group, possibly as a result of the apparent-time increase in pragmatic uses of
i don’t know with which this variant is closely associated (see Figure 4.5 above).
Other variants undergo dramatic changes in terms of age as well as sex effects. I
don’t know, in addition to suddenly dipping in frequency in the youngest age group,
loses its association with male speakers in the oldest age group to become strongly
associated with female speakers in the youngest age group. I divn’t knaa, in addi-
tion to exhibiting an incremental increase in frequency across age groups, loses
its association with female speakers in the oldest age group to become strongly
associated with male speakers in the youngest age group. These patterns reveal that
social changes in variant choice occur despite the stable linguistic conditioning
of variants discussed earlier (see also Pichler & Levey 2011). They thus demon-
strate the methodological importance of including external factors in studies of
language change and grammaticalisation, for exclusive focus on internal factors
may yield inaccurate conclusions about the stability of variant selection (see also
Labov 1982: 76).
In sum, the formal variability of i don’t know in BwE is characterised by
external instability and internal stability. Despite high degrees of inter-group and
inter-speaker variation in the frequency, function and form of i don’t know, all
speakers in the sample, irrespective of age and sex, use individual variants for the
same broad meanings. This reduces the cognitive load on utterance interpretation
and promotes communicative efficiency. There might then be an interactional
motivation for the stability in the functional conditioning of variant choice.
122 The Structure of Discourse-Pragmatic Variation

4.7 Conclusion

The preceding analysis of the construction i don’t know has shown that the
distribution of its formal variants in BwE is not random but systematically con-
strained by age, sex, function and/or syntax. The results confirm the functional
split between I don’t know and I dunno previously reported for American English,
and provide evidence for claims in the literature that the synchronic distribution of
variants is an outcome of grammaticalisation (Baumgarten & House 2010; Bybee
& Scheibman 1999; Scheibman 2000). Previously undocumented findings include
the differential conditioning of non-localised and localised variants, the role of
social factors in constraining patterns of formal variation in the use of i don’t
know as well as the stability of internal and the instability of external predictors
on variant selection.
Yet despite conducting an apparent-time analysis of the data configured to test
the simultaneous effect on variant choice of internal as well as external predictors,
this chapter has failed to illuminate the developmental trajectory of the variants
of i don’t know or the social forces driving the grammaticalisation of i don’t
know. This is due to the internal stability of linguistic variation in the dataset and
the relatively shallow time-depth of the BwE corpus. Comparable diachronic and
longitudinal data are required to: verify the grammaticalisation scenario postulated
in Section 4.6; explore whether the sex-differentiated patterns in the usage and
distribution of the variable and its variants noted in Sections 4.5.1 and 4.5.2 are
indicative of male- or female-led innovations; and establish whether young males
are in the process of assigning a specialised meaning to I divn’t knaa, as suggested
by the patterns discussed at the end of Section 4.5.2.
 chapter 5

The construction i don’t think*

5.1 Introduction

The construction i don’t think has been studied quite extensively as a proto-
typical example of transferred negation or negative raising (Fillmore 1963; Fischer
1998; Horn 1978, 2001: 308–330; Nuyts 1990; Tovena 2001), whereby the negative
particle adjacent to the predicate in the complement clause is transferred, or raised,
to the predicate in the matrix clause, as illustrated in (1).34
(1) a. I think that he will not come.
b. I do not think that he will come. (Horn 2001: 315)

Although most of these studies have addressed the alleged synonymy between
sentence (1a) and weak readings of sentence (1b) (see further Section 5.2.1), lit-
tle is known about the interactional importance of i don’t think except that it
functions as an epistemic device that serves in casual conversation to hedge utter-
ances and to mitigate interactional conflict (Bublitz 1992; Simon-Vandenbergen
1998). This chapter explores the functionality of i don’t think in the BwE cor-
pus with a view to establishing whether function, together with speaker sex, age
and complement type, affects the construction’s formal variability in the data. The
variable form of i don’t think is illustrated in the extracts from the corpus in
(2). It is depicted here and throughout by means of variation in orthography, with
each orthographic form (I don’t think, I doØ think, I divn’t think, I dinnae think)
representing a different variant whose phonetic, phonological and morphemic
properties are described in detail in Section 5.3.2.
(2) a. HP: When you’re married.
Ryan: I don’t think I’ve got one for when you’re married. I doØ think there is one sort of
like.

* An earlier, less advanced analysis of the data investigated here was presented in Pichler (2009).

34. Although pragmatic uses of i don’t think constitute epistemic phrases modifying their fol-
lowing host clauses (see further Section 5.6), the labels matrix clause and complement clause are
employed throughout this chapter to differentiate the syntactic components of i don’t think-p
constructions (see further Chapter 4, footnote 24).
124 The Structure of Discourse-Pragmatic Variation

b. HP: Would you call it the bog?

Alicia: >Not really.<
HP: No.
Alicia: [(h) I think that doesn’t] sound right, does it? [I’m going
Rebecca: [I don’t think I would use it.] [I’ve heard
Alicia: = to the bog.]
Rebecca: = people] say it, but I doØ think I would.
c. HP: If you could, would you change where you came from?
Godfrey: [I don’t] think so.
Lori: [No.]
No.
Godfrey: I divn’t think so.
d. HP Do you think that ol- eh younger people also talk differently
than older people?
Evelyn: ºNo, I dinnae think so. No.º

Bybee & Scheibman’s (1999) analysis of i don’t think in a corpus of American

English conversational data revealed that the construction is not consistently
phonetically reduced even though it constitutes a compositionally unanalys-
able unit which functions exclusively in the interpersonal domain of discourse
to hedge or mitigate utterances and is never used referentially to express a lack
of cognitive activity. Systematic analysis of the construction’s variable form and
function in the socially stratified BwE corpus uncovers similar patterns, and
demonstrates that the lack of consistent reduction of i don’t think (I don’t
think > I doØ think) is, at least in part, interactionally motivated to signal fine-
grained nuances of pragmatic meaning. Moreover, the analysis shows that non-
localised and localised variants are conditioned by different parameters, with
the former carrying a functional and the latter a social load. Close analysis of
the construction’s distribution and usage in the data also reveals that, despite its
lack of consistent attrition, i don’t think constitutes a formulaic discourse unit
in BwE. The analysis presented here thus addresses the broader research aim of
this project set out in Chapter 1 which is to establish the mechanisms underly-
ing discourse-pragmatic variation and change. Attainment of this objective is
warranted in large part by conducting an analysis of i don’t think which is
comparable to that of i don’t know in the previous chapter.
For ease of cross-variable comparison, the organisation of this chapter parallels
that of Chapter 4. Section 5.2 reviews previous studies of i don’t think which
inform the current analysis, and sets out in more detail the aims which guide the
analysis. Section 5.3 circumscribes the variable context, outlines the coding deci-
sions, and details the variants of i don’t think identified in the BwE corpus. This
 Chapter 5.  The construction i don’t think 125

is followed by Section 5.4 which introduces the range of functions performed by

i don’t think in the present dataset and thus isolates the factors constituting the
factor group function. The effect of function and other predictors on the distribu-
tion of formal variants of i don’t think is examined in Section 5.5 which presents
the results of the quantitative analysis. The results are discussed in Section 5.6.
Finally, Section 5.7 is the conclusion.

5.2 Previous research on i don’t think

This section reviews the recent literature on i don’t think. It discusses first stud-
ies which provide insights into the semantic-pragmatic uses of i don’t think in
spoken discourse. It then reviews previous studies of don’t-reduction, including
the only quantitative study carried out to date which has focused specifically on
the formal variability of i don’t think. The section ends with an outline of how
the present study illuminates the conditioning of form variation in the use of
i don’t think by including non-localised and localised variants in the envelope
of variation, isolating fine-grained pragmatic meanings of the construction’s use,
and including multiple contextual factors in the quantitative analysis of a socially
stratified corpus of interview data.

5.2.1 Qualitative studies: Discourse-pragmatic functions of i don’t think35

Accounts of the construction’s semantic-pragmatic meanings have been largely

concerned with exploring the differential readings of negative raising with i don’t
think, and with invalidating claims about the alleged synonymy between weak
readings of i don’t think and corresponding structures with the negative clitic
in the complement clause.
Scholars investigating i don’t think as a phenomenon of transferred nega-
tion distinguish two readings for sentences of the type ‘I don’t think p,’ as in (1b)
above. In what is commonly referred to as their strong reading, these sentences are
construed as negative propositions: transferred negation does not occur and the
negative takes scope over think (Simon-Vandenbergen 1998: 313–315). Speakers
use i don’t think in this way to correct or deny a suggestion made in the preced-
ing discourse that they hold the opinion expressed in the following complement

35. Section 5.4 describes in detail the functional repertoire of i don’t think. Therefore, no
examples are provided here of the functions identified for it in the literature.
126 The Structure of Discourse-Pragmatic Variation

(Nuyts 1990: 572). Simon-Vandenbergen (1998) refers to this reading as the con-

struction’s propositional or literal reading. In what is commonly referred to as
their weak reading, sentences of the type ‘I don’t think p’ are construed as nega-
tive modal expressions: the negative does not apply to think but, through negative
transfer, applies to the predicate in the following complement clause (Simon-
Vandenbergen 1998: 313–315). The transfer of the negative is possible because in
the weak reading the clause boundary between the matrix clause and complement
clause is transparent to the scope of negation (Bublitz 1992: 557–558).36 This read-
ing of (1b) is generally paraphrased as ‘I think not-p.’ Because it constitutes an
epistemic modality marker that signals speakers’ reduced commitment to their
propositions (Bublitz 1992), Simon-Vandenbergen (1998) refers to this reading
as the construction’s non-propositional or modal reading.
Nuyts (1990) and Simon-Vandenbergen (1998) posit that the scalar distinc-
tion between the propositional and modal readings of i don’t think is largely
affected by the construction’s ambient linguistic context. Tonic stress on the nega-
tive, an intonational break between the negative and think, and the construction’s
co-occurrence with lexical items that strengthen the illocutionary force of an ut-
terance (e.g. at all, really) combine to trigger a strong or propositional reading of
i don’t think. By contrast, the co-occurrence of i don’t think with weakening
elements (e.g. the hedges well, you know, a bit, sort of), neutral prosody, i.e., lack
of tonic stress on any of its elements, or tonic prominence on think when the
construction is followed by so trigger a weak or modal reading of i don’t think.
What both readings seem to share, though, is that they refer to Persson’s (1993)
stative meaning of think in the sense of ‘believe/find,’ i.e., a cognitive state resulting
from a cognitive process, rather than the verb’s dynamic meaning in the sense of
‘cogitate/conceptualise’ (Simon-Vandenbergen 1998: 313–314; see also Bybee &
Scheibman 1999: 587–588).
Scholars addressing the meaning relation between the weak reading of trans-
ferred negatives, (1b), and their corresponding structures with the negative in the
complement clause, (1a), have argued that the two do not stand in total synonymy.
The negative force of ‘I don’t think p’ is said to be weaker than that of ‘I think
not-p’ (Givón 1993: 201; Horn 1978: 131–136, 2001: 315; Tovena 2001: 333). This
suggestion is supported by Bublitz’s (1992) detailed exploration of the interactional

36. Transferred negation with weak reading is restricted to the semantic classes of predicates that
denote opinion, perception, probability, intention/volition and judgment/weak obligation (Horn
1978: 187–208). With predicates that fall into other semantic classes, the clause boundary is also
the boundary to the scope of negation, and the transfer of the negative from the complement
to the matrix clause results in a change of propositional content (for further details, see Bublitz
1992; Tovena 2001).
 Chapter 5.  The construction i don’t think 127

motivation for negative raising with first person singular pronouns and opinion
predicates in the London-Lund Corpus (Svartvik 1990). Bublitz argues that trans-
ferred negatives increase the tentativeness of positive polarity epistemic markers
through: (i) the distance of the negative from the proposition that it negates (see
also Horn 1978: 132–133, 2001: 315); (ii) the type of predicate that is negated;
and (iii) the utterance-initial positioning of the negative. By raising the negative
from the complement clause where it semantically belongs to the preceding matrix
clause, the negative becomes dislocated from the constituent that it negates. This
weakens the force of the negative. Further, the transferred negative weakens the
force of the proposition by reinforcing the epistemic modality already inherent
in opinion predicates such as think (Urmson 1952). Lastly, the dislocation of the
negative from near-turn-final positioning, which is typical of new information,
to utterance-initial positioning adjacent to I marks the following proposition as
more subjective, and softens the novelty and the potentially antagonistic effect of
disaligning turns. Bublitz’s (1992: 560) account of i don’t think thus highlights
the construction’s interactional import as “an expression of epistemic modality,
involvement and politeness [which is] used by the speaker to increase the degree of
qualification and tentativeness of the underlying proposition” (see also Kärkkäinen
2003: 20, 36; Thompson 2002).
Based on her analysis of casual conversations and political genres, Simon-
Vandenbergen (1998) provides an overview of i don’t think’s most frequent con-
texts of occurrence. They include: disagreement sequences; contexts which require
polite rejections or refusals; contexts where speakers aim to distinguish their own
point of view from that of others; and contexts where speakers strive to signal from
the start their attitude towards an upcoming proposition. In addition to perform-
ing interpersonal functions in these contexts, i don’t think has been argued to
have a textual function related to the theme/rheme structure of discourse: trans-
ferred negatives are preferred to non-transferred negatives because they avoid focal
prominence on the negation (Horn 1978; Nuyts 1990; Simon-Vandenbergen 1998).
The review of the literature demonstrates that i don’t think is employed by
speakers to qualify their commitment to their propositions or to forcefully deny
suggestions that they entertain particular opinions. The distinction between these
readings is largely affected by prosody. Without marked tonic prominence the
construction tends to have an attenuating effect, whereas with tonic accent on one
of its elements it tends to signal its propositional reading. There is reason to sus-
pect, however, that currently available descriptions of the functionality of i don’t
think may be inadequate and incomplete. While it is conceivable that transferred
negation with i don’t think is motivated by concerns for textual coherence, as
outlined immediately above, it seems equally plausible that this strategy may be
motivated by concerns about face and attempts to reduce the negative force of
128 The Structure of Discourse-Pragmatic Variation

interactionally dispreferred turns. Consequently, a straightforward classification

of such uses as operating in the textual domain seems debatable. This is not to
deny the possibility that i don’t think may perform textual functions. Its positive
counterpart, I think, performs a number of routine organisational tasks in addition
to signalling various degrees of epistemic modality (see, for example, Kärkkäinen
2003). It is therefore plausible that the pragmatic usage of i don’t think may not
be restricted to the interpersonal domain, as suggested in previous studies. Similar
to I think, i don’t think may also play a discourse- and information-structuring
role, albeit in negative contexts. Comparison with I think reveals another potential
weakness of current functional accounts of i don’t think. As demonstrated above,
Bublitz (1992) argues that transferred negation increases the attenuating effect of
I think. Yet I think functions not just as a hedge to signal reduced epistemicity; it
also functions as a booster to strengthen speakers’ commitment to their proposi-
tions (see, for example, Holmes 1990, 1995: 92–95). It is therefore plausible that,
depending on their prosodic encoding, transferred negatives might reinforce not
just the attenuating but also the strengthening effect of I think. The current study
explores the various possibilities raised above.

5.2.2 Quantitative studies: Variation and change in the use of i don’t think

Even rarer than qualitative studies are quantitative studies of i don’t think, es-
pecially those exploring variation in the construction’s morpho-phonology or
phonetics. This shortage is deplorable. Together with detailed analysis of negative
periphrastic do variation in other constructions, systematic quantitative analysis
of the formal variation of i don’t think enhances current understanding of the
nature and correlation of grammaticalisation processes which give rise to struc-
tured patterns of synchronic language variation.
Previous quantifications of the distribution of i don’t think in spoken
corpora have generally been concerned with comparing the relative frequency
of i don’t think with that of ‘I think not-p’-constructions and with providing
interactionally-based explanations for the marked preference for the former (see
further Bublitz 1992; Simon-Vandenbergen 1998). To date, only one study has
examined the formal and functional variation of i don’t think. This study was
conducted by Bybee & Scheibman (1999) on a small corpus of American English
conversational data and was motivated by their contention that non-referential
modal uses of i don’t think, as described in Section 5.2.1 above, constitute a
grammaticalised discourse unit. With increased frequency in discourse, i don’t
think has lost its internal structure, become automated as a processing unit
(chunking) and weakened its association with its composite parts (I, don’t, think)
 Chapter 5.  The construction i don’t think 129

(autonomy). As a result of these changes, i don’t think has generalised its mean-
ing and developed functions in the interpersonal domain of discourse. In Bybee &
Scheibman’s data, all tokens of i don’t think were used non-referentially to sig-
nal an epistemic stance or to soften disagreements. In order to establish whether
the construction’s desemanticised status is reflected in its realisation as a reduced
form, as was the case in their data with i don’t know, Bybee & Scheibman coded
the 19 tokens of i don’t think in their corpus for their phonetic form: full vowel
variants with [o] and reduced vowel variants with [ǝ] in don’t (see further foot-
note 22 in Chapter 4.2.2). The fact that the majority of i don’t think tokens
were non-reduced showed that the grammaticalised status of i don’t think as a
formulaic construction with interpersonal functions is not congruent with pho-
netic reduction:
though the meaning of [i don’t think] is compositionally unanalyzable in con-
versation – indicating a more grammaticized unit – there is no consistent formal
reduction concomitant with this functional shift […] (i.e., we find in the data both
full and reduced variants of don’t in [i don’t think]).
 (Bybee & Scheibman 1999: 588)

These findings contrast sharply with the results obtained by Bybee & Scheibman
for i don’t know which suggested that formal reduction and semantic-prag-
matic change operate in unison (see also Chapter 4 of this book). However,
the findings for i don’t think correspond to Brinton’s (2005: 293, 2006: 308)
observation that attrition may occur when discourse-pragmatic variables un-
dergo grammaticalisation but that it is not a necessary concomitant of semantic-
pragmatic change.
Patterns of variation in the realisation of don’t have also been explored in a
small number of studies of so-called ‘secondary contraction’ in Yorkshire varieties
of British English (Petyt 1978; Richards 2008; Whisker 2007; Whisker-Taylor in
prep.). These studies demonstrate that don’t-reduction in northern British English
primarily affects the nasal and, to some extent, the plosive of the clitic particle in
don’t, yielding variants such as [doːt], [dõt], [dot], [dõʔ] or [doʔ]. They also reveal
that the reduction of the nasal is conditioned by social and stylistic factors: the
frequency of reduction varies across the varieties investigated; where it occurs, it
is generally more frequent among younger speakers, among speakers from the
lower socio-economic classes and in informal contexts. However, these studies
have not isolated the construction i don’t think from productive constructions
containing negative periphrastic do. Consequently, they provide little insight into
the specific nature and conditioning of variability in the form of i don’t think
in British English.
130 The Structure of Discourse-Pragmatic Variation

The lack of consistent reduction of pragmatic tokens of i don’t think in

Bybee & Scheibman’s (1999) American English data suggests that the variation
between full and reduced variants of i don’t think is not conditioned by broad
referential vs. pragmatic meaning distinctions. Because their study was concerned
with constituent structure rather than the structure of discourse-pragmatic vari-
ation, Bybee & Scheibman made no attempt to uncover the impact of other fac-
tors on the alternation between full and reduced variants of i don’t think. Even
if they had attempted to test claims made elsewhere regarding for example the
importance of social factors in conditioning don’t-reduction, these would have
been foiled by the limited size of their dataset (N = 19). The socially stratified BwE
corpus contains a far larger number of i don’t think tokens and thus makes
possible testing these claims.

5.2.3 Aims and contribution of this chapter

The present investigation into form variation in the use of i don’t think in the
260,000-word corpus of BwE develops previous quantitative and qualitative stud-
ies of this construction by: testing the contribution of both internal and external
factors to observed patterns of form variation in the data; including both non-
localised and localised variants in the envelope of variation; and investigating in
finer detail the range of pragmatic functions performed by i don’t think across
all domains of discourse.
Examination in this study of the joint effect of intra- and extra-linguistic pre-
dictors (function, complement type, sex, age) on formal variation patterns in the
data represents a development beyond Bybee & Scheibman’s (1999) single fac-
tor analysis which allows us to fully uncover the organisation of variability in
the realisation of i don’t think in BwE. Consideration of social factors in the
quantitative analysis also makes it possible to assess whether any apparent-time
changes in the frequency of reduced variants of i don’t think in the BwE corpus
parallel those reported for secondary reduction in Yorkshire English (Richards
2008; Whisker 2007; Whisker-Taylor in prep.). Further, examination of the socio-
linguistic distribution of variants with different geographical spreads (see further
Section 5.3.2) broadens the scope of Bybee & Scheibman’s (1999) analysis which
focused on non-localised variants, and enables us to determine whether the dif-
ferential patterning of non-localised and localised variants established for i don’t
know in Chapter 4 extends to variants of i don’t think. Finally, the thorough
functional analysis of i don’t think in Section 5.4 makes it possible to establish
in Section 5.5 whether observed patterns of variation in the construction’s form are
conditioned by fine-grained pragmatic differences in its use. Also, close inspection
 Chapter 5.  The construction i don’t think 131

of the construction’s semantic-pragmatic meanings in all its sequential contexts

of occurrence and across multiple discourse domains yields a more detailed and
accurate description of the construction’s functional repertoire than are currently
available in the literature. The broad scope of the analysis thus affords important
new insights into the constraints on the formal variation of i don’t think and
contributes to advancing our current understanding of the structure of discourse-
pragmatic variation.
Because the results of any variationist analysis are determined in large part by
the methodological procedures adopted for circumscribing the variable context,
identifying the set of alternative variants and coding the data for contextual con-
straints, it is important to detail these procedures in Section 5.3.

5.3 The variable context and data coding

5.3.1 The variable and the envelope of variation

The variable selected for analysis in this chapter is the construction labelled i don’t
think, defined as any first person singular subject-negative-opinion predicate con-
struction in the present tense (see Chapter 2.3.2).37 Inclusion in the variable context
of the range of opinion predicates instantiated in the corpus (think, suppose, believe,
feel) enables us to establish the lexical formulaicity and internal, ­morphological

37. Van Bogaert (2010) notes that in the BNC subject(-negative)-opinion predicate construc-
tions manifest a high degree of variation in terms of TAM (tense-aspect-modality) specifica-
tions which does not bar them from functioning as epistemic markers. The existence of tense
and modality variation in the present dataset does not affect the circumscription of the variable
context set out above. Although subject-negative-opinion predicate constructions in the past
tense or with modal verbs can function pragmatically in the BwE data (e.g. I didn’t think they did
but mum and Amber thought so. – I wouldn’t think it was fashionable.), they are not included in
the envelope of variation. As shown in Chapter 3, past tense periphrastic do and modal would
are virtually categorically negated with non-localised negative particles, i.e., they are not subject
to the same degree of variation as present tense negative do. Furthermore, second and third
person subject-negative-opinion predicate constructions (e.g. Do you not think Geordie’s a lot
more common than Scottish? – They don’t really think, like, the people who are intelligent have a
life.) were not included in the envelope of variation either. Constructions with second person
subjects function pragmatically only in questions (Thompson & Mulac 1991a: 243), a context
which is far less variable in BwE than declaratives (see Chapter 3). Constructions with third
person subjects do not generally function non-referentially (Urmson 1952) (for exceptions, see
Kearns 2007).
132 The Structure of Discourse-Pragmatic Variation

fixation of negative epistemic marking with opinion predicates.38 With the variable
thus defined, the initial stage in the creation of the data file involved extraction
from the corpus of every instance of a construction derived from the linear string
of components exemplified in (3): (I) + (present tense negative do) + (opinion
predicate). Unless otherwise stated, the label i don’t think is used as the umbrella
term for the diversity of constructions included in the variable context. Individual
opinion predicates are differentiated only in the context of the quantitative analysis
in Section 5.5 and in the discussion of the results in Section 5.6.
(3) a. I doØ think we’ve ever had another name for cat.
b. I don’t suppose I would be able to alter their opinion.

The extraction process described above yielded instances where think, suppose,
believe and feel denote ‘have a sensation’ (4a), ‘accept the claim that’ (4b), ‘regard as
having a certain quality’ (4c), or where they refer to a cognitive process (4d). When
they denote these meanings, opinion predicates are followed by non-finite phrasal
complements, and the matrix clauses of which they are a part do not constitute
examples of transferred negation. They are thus not subject to the same functional
versatility as tokens that denote ‘believe, find’ and are followed by a finite clausal
complement or the pro-form so (see the examples in (3) above and (5) below).
These tokens were therefore removed from the variable context of analysis.
(4) a. I don’t feel a Borderer.
b. I don’t believe it.
c. I dinnae think of it that way.
d. Na. Don’t really think about the border.
I don’t think like that so much now.

38. The semantic-pragmatic meanings of these predicates are addressed in the course of this
chapter. Note that they have elsewhere been termed “modal verbs” (Coates 1987), “weak assertive
predicates” (Hooper 1975), “epistemic verbs” (Kearns 2007), “psychological verbs” (Leech 1983),
“private verbs” (Stubbs 1986), “parenthetical verbs” (Urmson 1952) or “mental predicates” (van
Bogaert 2010, 2011). Positive polarity constructions of these predicates with I have been referred
to in the literature as “(first person) epistemic parentheticals” (Brinton 1996: 211–212; Thompson
& Mulac 1991a, b), “epistemic phrases” (Kärkkäinen 2003), “comment reduced parenthetical
clauses” (Kaltenböck 2005), “comment clauses” or “content/parenthetical disjuncts” (Quirk et al.
1985: 1112). I refrain from using these terms because they imply that the predicates listed above
(think, believe, suppose, feel) are always used to qualify an assertion. As shown in Section 5.4,
however, this is not necessarily the case. ‘Opinion predicates’ is therefore used here as a neutral
label. In contrast to the labels listed above, it also has the advantage of more precisely delimiting
the range of predicates included in the analysis.
 Chapter 5.  The construction i don’t think 133

In the BwE data, tokens of i don’t think that qualify as matrix clause con-
structions with transferred negation are followed either by complement
clauses introduced with that (5a), by complement clauses introduced with a
zero-complementiser­ (5b), or by the pro-form so (5c). The pro-form presup-
poses across sentence boundaries the element to which it is structurally related
(Halliday & Hasan 1976: 130–141; Quirk et al. 1985: 880–881). In other words,
i don’t think so substitutes for a complement clause and denies or modifies
a preceding assertion by self or other without repeating it. It thus contributes
to the cohesion of discourse.39 Unlike i don’t think and its clausal comple-
ments, i don’t think and so are always intonationally integrated with each
other. Tokens followed by clausal complements and tokens followed by so alike
are included in the analysis.
(5) a. Now they would still be from Berwick because I don’t think that you talk
about people being from Tweedmouth or Spittal.
b. Because certainly Alan Beith, our MP, I don’t think Ø he has much power
with the Labour government in London.
c. I divn’t think so. They’re all just grouped together.

In addition to containing tokens of i don’t think which occur before the clause
that they are construed with in terms of negation, as in (5) above, the data con-
tain tokens which occur after the clause over which they have scope, as shown
in (6) below. These tokens are referred to here as post-positioned or post-posed
tokens of the variable. Unlike pre-positioned tokens, which form an intonation
unit with the following complement, post-positioned tokens always constitute a
separate intonation unit and occur only if the associated clause is independently
negated. Both pre- and post-positioned tokens of i don’t think are included in
the analysis.
(6) He wouldn’t call him fither to his face, I divn’t think.

After tokens whose form could not be determined had been removed, a total of 270
tokens of the variable i don’t think remained in the database for in-depth analy-
sis. To uncover the constraints on their distribution, these tokens were coded for
their form, as described in Section 5.3.2, as well as a series of potential contextual
constraints, as described in Section 5.3.3.

39. Its more formal equivalents I don’t think that and I think not do not occur in the data.
134 The Structure of Discourse-Pragmatic Variation

5.3.2 The dependent variable: Variants of i don’t think in BwE

The main focus of this analysis is on uncovering the constraints on the vari-
able morpho-phonology of i don’t think illustrated in the extracts in (2) in
Section 5.1. As was the case with i don’t know in Chapter 4, the variation mainly
affects the realisation of negative periphrastic do and the choice of the negative
clitic. Notwithstanding the drawbacks of allocating tokens to discrete categories
(see Chapter 4.3.2), auditory analyses established that the 270 tokens of i don’t
think in the database could be meaningfully divided into four discrete variant
categories. This section introduces these categories and assigns them the status of
non-localised or localised variants depending on their geographical dispersion.
Tokens in the first two categories of variants differ from each other with regard
to the realisation of the negative clitic particle -n’t (full realisation vs. reduction/
deletion­) and optionally the vowel quality of do (full vs. nasalised/reduced). The
first category contains tokens such as [donʔθɪŋk] or [dʊnθɪŋk]: the nasal of the
negative clitic particle -n’t is fully realised, and the final plosive of the clitic is
either realised as [ʔ] or omitted. Due to their relative lack of phonological reduc-
tion, these variants are labelled full variants and are orthographically represented
as I don’t think. The second category contains tokens such as [doʔθɪŋk], [dʊθɪŋk]
or [dəθɪŋk]: the negative clitic particle -n’t is either reduced to [ʔ] or deleted alto-
gether, and do is optionally produced with vocalic nasality and/or reduced to [ə];
very occasionally, the initial fricative of think is realised as [f] or [ɦ] or is omitted
altogether, resulting in forms like [doʔĩŋk] or [dəʔɪŋk]. Because of their phonologi-
cal reduction, the variants with nasal deletion are labelled reduced variants and are
orthographically represented as I doØ think.
The full variant I don’t think is the standard variant and as such is non-
localisable­. The reduced variant I doØ think is reminiscent of secondary contrac-
tion reported for Yorkshire varieties of British English (see Section 5.2.1). Petyt
(1978) argues that secondary contraction, i.e., the reduction or deletion of the
clitic negative particle, is regionally restricted within Britain. However, this the-
sis is based on informal observations and is not confirmed by empirical studies
of -n’t-reduction across varieties of English. In her survey of negation strategies
across UK varieties of English, Anderwald (2002: 66) reports secondary reduction
for the Midlands. Bybee & Scheibman’s (1999) American English data also contain
tokens of i don’t think without a nasal in don’t. In light of these findings and
the fact that it was the erosion of the negative particle that led to the introduction
of do-support in the first instance (Labov 1994), we may assume that the reduc-
tion of don’t in i don’t think is more widespread than implied in Petyt (1978)
(see also Richards 2008: 124). Phonologically full and reduced variants of i don’t
think are therefore categorised as non-localised variants.
 Chapter 5.  The construction i don’t think 135

Tokens in the third and fourth categories of variants differ from those above
in that negative periphrastic do is realised as divn’t or dinnae (see Chapter 3.3).
The category of variants orthographically represented as I divn’t think contains
tokens such as [dɪvn̩θɪ̈ŋk]: negative periphrastic do contains a kit-vowel and
some degree of friction. The category of variants orthographically represented as
I dinnae think contains tokens such as [dɐ̃nɪ̃θɪ̃ŋk]: periphrastic do is negated with
the negative clitic -nae, yielding the negative auxiliary dinnae [dənɪ] or [dənə]
or [dɐnɪ]. As detailed in Chapter 3.3.2, the forms divn’t and dinnae are not used
throughout all of England or the English-speaking world. Divn’t is strongly asso-
ciated with the north(-east) of England. Dinnae is strongly linked with Scotland,
although its use has been recorded in Northumberland. Because their distribution
in England is largely restricted to the north(-east) of England, the variants I divn’t
think and I dinnae think are categorised as localised variants.
Table 5.1 summarises the BwE inventory of variants for i don’t think, with
an indication of their geographical distribution in England.

Table 5.1  Inventory of variants of the construction i don’t think in BwE

non-localised variants localised variants
I don’t think I divn’t think
I doØ think I dinnae think

5.3.3 Independent variables: Data coding

In addition to coding each occurrence of i don’t think for its variant form, a
coding schema was devised for those extra- and intra-linguistic constraints which
were hypothesised to affect the formal variation in the data.
If Bybee & Scheibman’s (1999) premise concerning the grammaticalised status
of i don’t think in their American English data applies more generally to BwE,
we may assume that social factors impact on the formal variation introduced in
Section 5.3.2 above. Grammaticalisation processes are generally socially embed-
ded, and the synchronic distribution of grammaticalised variants tends to be so-
cially conditioned (see Chapter 4.3.3 for relevant references). The hypothesis that
social factors impact on the formal variation patterns in the data is supported by
studies of northern English dialects which have demonstrated the effect of speaker
sex and age on the frequency of don’t-reduction (see Section 5.2.2 above) as well as
the results reported in Chapters 3 and 4 which have shown the operation of social
factors on the occurrence of negative auxiliaries with -nae in productive construc-
tions and on the occurrence of divn’t in formulaic constructions. In order to probe
the social distribution of variants and control for the effect of individual speaker
136 The Structure of Discourse-Pragmatic Variation

on formal distribution patterns in the data, each token of the variable was coded
for speaker sex, age and individual speaker.
Furthermore, it is possible that complement type constrains the distribution of
full and reduced variants of i don’t think. Simon-Vandenbergen (1998) argues
that when i don’t think occurs in reply to a polar question and is followed by
the pro-form so, the only new information provided is the modality and negation
encoded in i don’t think. The pro-form replaces a preceding proposition and as
a result its role is cohesive rather than referential (see Section 5.3.1). We may there-
fore hypothesise that because it is the negative in i don’t think so that introduces
the new referential information, i don’t think is less likely to be reduced when
it occurs with the pro-form than when it occurs with a clausal complement which
introduces new referential information. To test this hypothesis, all tokens of the
variable were coded for complement type: finite clause vs. pro-form.
Finally, even though there is no close correlation between formal reduction
and pragmatic meaning in Bybee & Scheibman’s (1999) set of i don’t think to-
kens, there is reason to believe that function may be implicated in the formal
variation of i don’t think. Kärkkäinen’s (2003: 121–129) analysis of naturally-
occurring American English data has shown that when I think does not signal an
epistemic stance but instead does some routinised organisational work such as
acting as a frame or marking boundaries in discourse, it is generally realised in
reduced form. Plug’s (2010) analysis of a set of formulaic constructions in spoken
Dutch has revealed that these constructions have different phonetic realisations
depending on their particular pragmatic functions and contexts of occurrence.
These findings furnish hypotheses that macro- and/or micro-level intra-pragmatic
distinctions may account for formal variation patterns in the use of i don’t think
in the BwE corpus. To test these hypotheses, each token of i don’t think in the
data was subjected to a painstaking functional analysis aimed at establishing its
specific pragmatic function in its specific context of use. Section 5.4 details the
results of the qualitative data analysis and introduces the coding schema for the
factor group function.

5.4 Qualitative analysis of i don’t think

The multiple functions performed by i don’t think in the BwE interview data
were isolated by following the procedures described in Chapter 2.5 and success-
fully applied in Chapter 4.4. To make the analysis transparent and replicable, each
function is illustrated with examples from the data and described with reference
to its interactional, sequential and ambient linguistic context (see page xvii for a
 Chapter 5.  The construction i don’t think 137

key to the conventions used for replicating prosodic features). Because many of the
functions detailed below have not previously been documented in the literature
and because fine-grained functional differentiations in the use of i don’t think
are crucial for explaining the formal variation patterns described in Section 5.5,
the analysis is presented in minute detail. The detailed coding protocol to emerge
from it is summarised in Table 5.2 at the end of this section.

5.4.1 Functions performed by i don’t think

The labels attributed to i don’t think in the literature include “epistemic modi-
fication” (Halliday 1985: 333–340), “epistemic marker/clause” (Bublitz 1992;
Kärkkäinen 2003: 20, 36; Thompson & Hopper 2001: 25), and “epistemic formu-
laic fragment” (Thompson 2002). They highlight the construction’s interactional
importance in the interpersonal domain to signal speakers’ assessment of the reli-
ability of their propositions. The usage of i don’t think to hedge utterances is
highly persuasive in the present data and is illustrated in the extracts in (7). By
prefacing their assessments with i don’t think, (7a) and (7b), and by choosing
to replace or follow the negative response particle no with i don’t think, (7c)
and (7d), the speakers in (7) signal their uncertainty vis-à-vis the validity of the
assessments and denials which they provide in response to the preceding opinion
questions. The construction’s attenuating effect in the examples in (7) is produced
and/or reinforced by its prosodic encoding with lowered volume, reduced tempo,
high pitch or a combination of these. The above examples, in particular (7a) with
its marked emphasis on I, also demonstrate that i don’t think is a marker of
subjective epistemic modality, i.e., “a conclusion drawn by the speaker from [their]
own knowledge of the state of the world at the time of speaking” (Watts 1984: 131).
Examples (7c) and (7d) further illustrate that i don’t think so constitutes a gram-
maticalised response particle.
(7) a. HP: What about males and females. Are there differences?
Leah: (.h)
Shannon: ºI don’t think there are [really.]º
Leah: [Not] really. Not [(eh)]
Shannon: º[I don’t] think it’s
male and female.º
b. HP: Are you only temporarily skint or can you be skint and be in
serious debt?
Jane: >ºI divn’t think you would say skint for serious debts.º< I
wouldn’t say skint for serious debts.
138 The Structure of Discourse-Pragmatic Variation

c. HP: Is jougle a bit derogatory?

(…)
Charlene: Um, I doØ think so.
d. HP: Do you sometimes go broader Berwick than you would
usually speak?
(..)
Rebecca: No, I doØ think so

Beyond modalising (near-)turn-initial assessments produced in response to polar

or alternative questions, i don’t think has the potential for signalling low degrees
of certainty when it occurs towards the end of multi-unit turns, as in (8a), or when
it occurs in extended opinion sequences involving one or more speakers, as in (8b).
The tokens in (8) are produced without prosodic variation, thereby communicat-
ing a lower degree of uncertainty than those produced with contrastive prosody
in (7) above.
(8) a. HP: And do you like the way you speak?
Elizabeth: I (.) (h) never sort of gave it much thought, really, to be
honest with you. (h) Only like em sort of like because em
(.) with me going away up to university now, [and] up in
HP: [mhm]
Elizabeth: = Edinburgh, and like (.) em when I’m speaking to
HP: mhm
Elizabeth: = people, they (.) they like em (.) sometimes they say to
me, “E? Sometimes you sound really Scottish and [other
HP: [mhm]
Elizabeth: = times] you sound really Geordie.” You know? And
HP: uh-huh
Elizabeth: = it’s just like (.) but (.) like I’m not like sort of (.) I do I
don’t mind, I doØ think, cos I’ve never thought about it
that much.
b. Elsie: They’ve got a lovely maternity hospital. [But] if anything if
Ronald: [There’s a (?)]
Elsie: there’s anything em (.) +
Ronald: Slightest complication, [they’re gone.]
Elsie: [They’re] they’re they’re whipped
away, [[aren’t they?]] You know?
HP: mhm mhm
Ronald: [[mhm]]
I don’t think there’s a big percentage of first babies ever
born in Berwick.
 Chapter 5.  The construction i don’t think 139

Although the literature describes i don’t think solely in terms of reduced epis-
temicity, the BwE data contain tokens of i don’t think which serve to signal
personal commitment and conviction. Immediately prior to the dissent-turn
sequence reproduced in (9), Matthew’s view that it is possible to recognise the
Berwick accent on the television had been challenged by his interview partner
Gerald. As Kotthoff (1993: 213) points out, “giving up a position that has already
been argued for can […] be face-threatening [to the speaker’s own face], be-
cause it could be interpreted as submissiveness.” While the conditional clause
in Matthew’s first turn in (9) slightly qualifies his earlier view, several linguistic
features signal his commitment to his original position: stress on don’t, tonic stress
on the negative polarity items in the complement (any, at all), the utterance-final
descending intonation contour, and the emphasis in the first turn on any difficulty
at all through reduced tempo.
(9) Matthew: If they were speaking eh if they were speaking broad enough, I
don’t think there’s >any [difficulty at all.<]
Gerald: [Th-they’d have to be speaking] very
broad.
HP: Mhm.
Matthew: I don’t think there’s any difficulty in spotting them at all.

Emphatic tokens of i don’t think can also occur in second pair parts of question-
answer adjacency pairs. Extract (10) forms part of a discussion about the north-
east referendum of November 2004, in which an overwhelming majority voted
against the setting up of a regional assembly in the north-east. After Alfred had
explained to HP why he had voted against the assembly, HP asks if he did not think
a regional assembly would have benefited Berwick. Alfred forcefully disputes the
proposition implicit in HP’s question. The strength of his denial is produced by
the lexical and prosodic encoding of the first TCU: the use and slow delivery of
i don’t think, the stress on don’t think, and the utterance-final falling intonation
contour. The animated tone and divergent content of the following TCUs, which
provide an alternative assessment to HP’s, add further weight to the incongruity
between HP’s and Alfred’s turns.
(10) HP: But would you (.) do you think it would have made it any bet-
ter if they [had the:]
Alfred: [>I don’t think] so<, to be [very honest wi you.] We
HP: [if you were governed]
Alfred: = could hae been worse off! We could hae ended up being
HP: Yeah?
Alfred: = worse off!
140 The Structure of Discourse-Pragmatic Variation

In turn-initial thematic position, emphatic i don’t think points backwards and

forwards, signalling that the upcoming turn will bring a new or different perspec-
tive to the previous turn. This is not the case with post-positioned boosters which
always only point backwards in discourse, as illustrated in (11):
(11) HP: So, would you eh would you rather be governed by Scotland
than by England?
Theodore: Oh, it wouldn’t make any difference, pet, I don’t think.

Whether subjective tokens of i don’t think function as hedges or boosters is de-

termined primarily by the ambient linguistic context. As a hedge, i don’t think
is generally produced with reduced volume, high pitch and lack of tonic stress;
where stress occurs, it is placed either on I to emphasise the subjectivity of the
following assessment, or on think when the variable is followed by the pro-form
so. As a booster, i don’t think is generally produced with emphatic stress on don’t
(think); decreased speech rate, increased volume and falling or final intonation
contours also tend to signal high degrees of certainty and commitment. Moreover,
hedges generally co-occur with low modality items (e.g. sort of, maybe, would,
may) in the complement clause, while boosters tend to co-occur with assertive-
ness markers (e.g. really, at all). These co-occurrences intensify the degree of epis-
temic commitment expressed (Fetzer 2011: 265) and create macro-modalities of
(un-)certainty (Simon-Vandenbergen 2000: 56).
The boosting function of i don’t think must not be confused with the con-
struction’s literal meaning (see Section 5.2.1). In (9) to (11) above, i don’t think
does not serve to deny the suggestion that the speakers hold the opinion expressed
in the complement: ‘I do not believe that p’ (propositional reading). Instead, it
serves to forcefully deny the proposition expressed in the complement: ‘p is not
the case’ (modal reading). My reading of these tokens as modal rather than propo-
sitional tokens is guided by the construction’s prosodic encoding and sequential
context of occurrence as well as a set of criteria laid down to disambiguate modifier
uses of first person opinion predicates from their main verb uses (Kearns 2007: 483,
496): the fact that i don’t think can be omitted from these utterances without
changing their propositional content (‘There’s no difficulty in spotting them at
all.’, ‘No.’, ‘It wouldn’t make any difference at all, pet.’); the fact that i don’t think
can be paraphrased with epistemic adverbials (‘There’s absolutely no difficulty in
spotting them at all.’, ‘Certainly not.’, ‘It wouldn’t make any difference at all, pet, no
way.’); and the fact that i don’t think is not transparent to tag questions (*‘I don’t
think there’s any difficulty at all, do I?’).
In the interpersonal domain i don’t think functions not only subjectively
to modify propositions but also intersubjectively to redress potential face-threats
 Chapter 5.  The construction i don’t think 141

and to signal affiliation and involvement. In (12), Luke responds to HP’s ques-
tion regarding his potential desire to speak another accent than Berwick. The
low-volume production of i don’t think-p and the use of the irrealis would
combine to mitigate what would otherwise be an overt and potentially offensive
negation of Luke’s desire to speak Cockney or Scouse. The strategic use of these
features thus serves to protect both the speaker’s and his hearers’ face needs
by reducing Luke’s commitment to his proposition and forestalling his hearers’
potential antagonism.
(12) Luke: Aye if I spo- if I was Manchester spoken I’d be alright. But (.) I
wouldn’t like to sound like a Cockney or [(h)] a liver- (.) like you
HP: [Yeah.]
Luke: = know Liverpool, like a Scouser. °Divn’t think I’d like to
HP: Yeah.
Luke: = sound like them.° @

What makes i don’t think such a useful interactional device is the fact that it
combines several techniques commonly employed to weaken the negative effect
of disagreements (see Pomerantz 1984: 64–77; Schegloff 2007: 64–73): it defers the
proposition that is being negated by pushing it further into the turn while simulta-
neously marking it as uncertain (see also Bublitz 1992: 572; Simon-Vandenbergen
1998: 321). As illustrated in (13), this makes unemphatic tokens of i don’t think
particularly well-suited for softening the presentation of disaligning second pair
parts of assessment sequences. In (13a), Glenn responds uncertainly to HP’s ques-
tion (see the slow tempo), indicating that he might accommodate to the speech of
non-Berwickers a little bit. Cody disaffiliates from Glenn’s response when he says
I don’t think I would. While the stress on I in the complement clearly indicates a
divergent opinion, the slowly articulated, unaccented construction i don’t think
mitigates the potentially face-damaging effect of Cody’s disaligning proposition
and adds a tone of tentativeness to it. In (13b), Charlene softens her disagreement
with HP’s suggestion that she might have a telephone voice by delaying it with a
follow-up question (Have I?) and prefacing it with i don’t think. The mitigating
effect of the turn sequence is heightened by its high-pitched encoding, demonstrat-
ing that speakers often draw on several linguistic devices to soften the unwelcome
interactional effect of disaligning turns.
(13) a. HP: And do you think (.) for instance if you talked to a Scottish
person that you would go a bit more Scottish and a bit more
Geordie when speaking to a Geordie person?
Cody: U::m?
142 The Structure of Discourse-Pragmatic Variation

Glenn: >A little bit.<

HP: °Yeah?°
Cody: I do- >I doØ think I would<, cos (..) I wouldn’t go more Scot-
tish cos (.) you know (.) we sound a little bit Scottish as it is
[so,] and I wouldn’t really go towards the Geordie
HP: [mhm] (.) mhm
Glenn: [mhm]
Cody: = side either cos, (..) I’d probably s:peak the same actually.
HP: °mhm°
b. Charlene: I would just stay the same. I use the phone a lot at work,
[so we] sort of (..)
HP: [mhm]
You have your [you’ve] a telephone voice. Yeah.
Charlene: [But I]
I do- have I? I doØ think I’ve got a telephone voice.

The face-saving effect of i don’t think is also evident in the examples in (14)
where the construction serves to mitigate or avoid the bluntness and abruptness
of negative assessments made in response to opinion questions (see also Simon-
Vandenbergen 1998: 321). In (14a), Godfrey’s turn consists of two TCUs: an em-
phatic negative response particle which rejects outright the proposition inherent
in HP’s question regarding the trendiness of Berwick upon Tweed; and I don’t
think so somehow which functions to attenuate the bluntness and abruptness of
the preceding flat denial no. i don’t think thus works towards maintaining a
friendly and non-hostile atmosphere in the interview situation. In (14b), the ut-
terances following Barbara’s initial I doØ think so indicate that she is committed
to her negative stance towards a devolved government in the north-east. Although
the predicate in i don’t think does not actually refer to a cognitive activity (see
Section 5.2.1), its association with a mental process implies that Barbara is not
rejecting HP’s preceding proposition outright but has given it some thought before
denying its validity. The denial is thereby not only made less blunt but also more
acceptable. Unlike hedging uses of i don’t think which are usually produced
with prosodic variation, attenuating uses of i don’t think like those in (14) are
consistently produced with marked falling intonation contours and without vari-
ation in tempo or volume.
(14) a. HP: Do you think Berwick is a fashionable place to be?
@
Godfrey: <@ No::. @ I don’t think so somehow. @>
 Chapter 5.  The construction i don’t think 143

b. HP: But do you think you might be better off, because right
now you’re quite far away from London. But then if you
were governed by Newcastle, do you think you might be
better off?
Barbara: [I doØ think so.] Th-they make a mess of everything. They
Paula: [No, no, no.]
Barbara: = cannae do anything right.
Paula: No.

Presumably because of their association with contradictions, disagreements and

rejections, negatives are often described as dispreferred and unwelcome moves in
discourse (Givón 1993: 188, 193; Pomerantz 1984). However, in certain sequential
contexts, negatives are interactionally preferred moves signalling confirmation and
agreement, acceptance and support, or affiliation and sympathy (Heinemann 2005;
Jefferson 2002). The negative polarity construction i don’t think is a case in point.
Following negatively framed assessments or verification questions, it can serve as an
intersubjective affiliation marker and positive politeness device, as shown in (15).
After discussing Berwick’s peripheral status vis-à-vis England and Scotland preced-
ing the extract in (15a), Jane comes to the conclusion that Berwickers are unwanted
mongrels. The final high rise on the following you know requests corroboration,
reassurance or validation of this assessment from Jane’s interview partner. Helen’s
assertive-sounding reaction No. I divn’t think so (see the lower than usual pitch and
falling intonation contour) provides the requested confirmation and signals Helen’s
affiliation with Jane’s assessment as well as her involvement in the interaction (see
also Bublitz 1992). In (15b), HP requests some corroboration from Ronald of her
inference from what he had said about gendered language differences prior to the
extract given below. By mirroring the polarity of HP’s negatively framed question
and producing don’t (and the following no) with moderately raised pitch, Ronald’s
response turn implements a tentative agreement and confirmation of HP’s infer-
ence. In contrast to those contexts where i don’t think is used in response to
positively framed utterances to implement disagreement, as in (13) above, affili-
ative instances of i don’t think are not usually followed by an elaboration, as is
characteristic of preferred response types more generally (Ford 2001).
(15) a. Jane: Aye we’re mongrels. I divn’t think er- anybody really wants wa.
You know?
HP: Yeah.
Helen: No. I divn’t think so.
b. HP: So you wouldn’t say that men are broader than women?
Ronald: No, I I don’t think so. No.
144 The Structure of Discourse-Pragmatic Variation

Contra reports in the literature, the use of i don’t think is not motivated solely
by speakers’ desire to be tentative or assertive, or their motivation to create and
maintain positive rapport with co-conversationalists. While at all times convey-
ing speakers’ subjective stance towards their propositions, strategically placed in-
stances of i don’t think may also function in the textual domain. Similar to some
of the discourse-pragmatic features discussed in Redeker (1991) and Schiffrin
(1980, 1987), i don’t think can act as a frame which serves to mark a range of
transitions and boundaries in discourse. The examples given below illustrate that it
is in fact the construction’s association with (inter-)subjectivity that makes it such
a useful tool for structuring and organising information and texts.
In (16), Theodore introduces Scottish nationalism into the discussion about
Berwick’s status vis-à-vis England and Scotland. Following a longish stretch of
discourse in which he relates to HP Scottish nationalists’ past attempts to bring
Berwick back under Scottish control, Theodore provides his own point of view
on the subject: But I doØ think it’ll ever happen like, you know. The connective
but preceding I doØ think signals that the upcoming discourse unit stands in
contrast to the preceding unit (Schiffrin 1987: 164–177). Drawing on its inher-
ent subjectivity, i don’t think (in neutral prosody) cues the co-participants in
the interaction as to the kind of contrast and transition that is underway: that
between the provision of factual information and the provision of a subjective
assessment of these facts.
(16) Theodore: You see, at one time th- (.) they reckon it they didn’t want it to
go back to England. (h) You’ll know, you have heard of Wendy
Wood? She’s the Scottish nationalist, and after the war
HP: mhm
Theodore: = she come down with the pipers. (.) And they’d (.) kilts
HP: mhm
Theodore: = on, and they marched across the new bridge, you know the
[Royal Tweed Bridge], [half way] across. And she took a chalk
Guy: [mhm]
HP: [mhm]
Theodore: = and she marked it. They [wanted] Home Rule for [Ber-
HP: [mhm] [mhm]
Theodore: = wick], for Scotland, Berwick for [Scotland]. They was trying
HP: [mhm]
Theodore: = to get it into Scotland for donkey’s years. (..) But
Guy: mhm
HP: mhm
Theodore: = I doØ think it’ll ever happen like, you know.
 Chapter 5.  The construction i don’t think 145

In addition to marking transitions between factual and evaluative propositions,

i don’t think serves to mark the transition to a conclusion that is based on speak-
ers’ preceding subjective reasoning behind it. This use of i don’t think is illus-
trated in (17), where Ryan and Dawn are requested to share their views on local
football and on the perceived importance of the post-devolution border. At the
beginning of their turns, Ryan and Dawn reflect on HP’s question and the appro-
priate answer to it. Towards the end of their turns, we find a combination of two
discourse-pragmatic features, so and i don’t think, followed by the requested
opinion. The combination of so and i don’t think signals: that what follows the
validation sequence is the requested assessment; that this assessment is derived
from the preceding subjective validation (see Schiffrin 1987: 204–207 on the func-
tionality of so); and that the assessment provided is subjective and potentially
unreliable (see the reduced volume in both examples as well as the turn-initial
hedge I dunno and the turn-final proclaimer certainly to me as a worker in Dawn’s
turn). Elsewhere in the BwE data, i don’t think also signals neutral or assertive
conclusions, depending on its prosodic encoding and ambient linguistic context.
(17) a. HP: What do you consider the local football derby to be.
(..)
Ryan: Uh (..) Berwick’s never really had one? Em (.) >it used to be
Meadowbank Thistle when they played in Edinburgh.<
HP: mhm
Ryan: = But they’ve moved to Livingston now, so ºI don’t think <Ber-
wick’s really got one>º.
b. HP: And do you think that the significance of the border has
changed since the since devolution?
(..)
Dawn: (h) ºI dunno? E:m (..) we ge- cos we we still get (..) funding and
I still get to go on Scottish [training] courses. So I don’t think
HP: [mhm]
Dawn: = it’s made that much difference.º Certainly to me as a
HP: yeah
Dawn: = worker.

The information-structuring role of i don’t think is not limited to (near-)turn-

final positioning. As shown in the second pair parts of the question-answer adja-
cency pairs in (18), i don’t think also occurs near-turn-initially following clusters
of hesitation signals or minimally coherent responses to mark the onset of longer
opinion sequences. The beginning of Alfred’s turn in (18a) contains several hesi-
tation markers (well, unfilled pause, outbreath) and an overt indication of the
speaker’s uncertainty as to how to formulate the requested assessment (How could I
146 The Structure of Discourse-Pragmatic Variation

describe this?). The following i don’t think signals a subjective stance towards and
marks the starting point of an extended opinion sequence presented in response
to HP’s question. In (18b), Luke launches his turn by emphatically denying the
proposition inherent in HP’s preceding question. The following i don’t think
marks the starting point of an elaboration which validates Luke’s initial denial and
serves to meet the interactional expectation that turn-initial denials be followed
by accounts or corrections (Ford 2001). At the same time, i don’t think signals
that both the account and the initial denial represent a subjective opinion. Both
instances of i don’t think in (18) are produced without variation in prosody. They
thus convey a neutral to assertive attitude.
(18) a. HP: And would you say that in Berwick that eh young people speak
different from older people?
(…)
Alfred: Well. (..) (.h) How could I describe this. (..) I don’t think they’ve
got much other choice [to be very] frank wi you. % Be-
HP: [mhm] mhm
Alfred: = caus:e some of them (..) I mean (..) you’ll have heard bad lan-
guage [and all the rest of it in the town,] and I mean a lot of the
HP: [mhm mhm]
Alfred: = young ones come out wi some terrible language!
b. HP: But would you, would you say an outsider had the right to (.)
[(?)]
Luke: [No!] I don’t think anybody has the right to come in and criti-
cise [somebody] else’s town! Like (.) No! I w- I like I would
HP: [yeah] yeah
Luke: = never go to some- I would never go somewhere else and say,
“Oh, your town’s shit.” Like that. Cos I think th-there’s
HP: yeah
Luke: = nay need for that.

i don’t think occasionally also marks slight shifts in the topical development of
interactions. This use of the variable is illustrated in (19), where Elsie and Ronald
discuss the fact that Berwick upon Tweed has few immigrants. Elsie conveys her
contentment with this state of affairs because immigration has led to problems in
southern England. When she puts a disclaimer on this proposition (What you read
in the paper, you know.), Ronald suggests that Berwick lacks the infrastructure nec-
essary to attract immigrants. This proposition is prefaced with unaccented i don’t
think which serves to introduce an unsolicited neutral opinion that is explored
in the following discourse (not reproduced here). It thus marks a slight shift in the
topical development of the interaction.
 Chapter 5.  The construction i don’t think 147

(19) Elsie: Aye, we’re lucky. We’ve w-we’re, you know, there’s no immi-
grants here. [There’s] (h) there’s nothing like that. We haven’t
HP: [yeah]
Elsie: = got anything like [(?)]
HP: [yeah]
Ronald: [There’s] no sort of racial tension and
things like that, [you know.]
HP: [yeah]
Elsie: [Aye. We’re] we’re really lucky. I mean, if
you go down south, I mean,
HP: yeah
Ronald: Unfortunately, you know, [I mean], every country has an [aw-
Elsie: [No.] [But
Ronald: = ful lot. But never mind.]
Elsie: = mind,] I think we we wouldn’t be a nice place if they did
come in. I think we I we I think w- there would be a lot of
HP: mhm
Elsie: = (.) aggro, if immi- immigrants came here. You know,
HP: mhm
Ronald: mhm
Elsie: = e::h cos that’s what’s happened down south, you know.
Ronald: +
Elsie: = [They g-], you know. [Well. What you read in the paper,
HP: [mhm]
Ronald: [It’s (?), Elsie, I I think, but (.)]
Elsie: = you know.]
Ronald: = >I doØ think there’s much (.) much here for them [really.<]
HP: [mhm]
Ronald: = [You know,] there’s no heavy industry.
Elsie: [No.]

A final information-structuring use of i don’t think identified in the BwE cor-

pus is exemplified in (20) where the construction contributes to structuring an
opinion sequence. Gabriel provides various points of view on the then England
football manager Sven Göran Eriksson, and his turn is interspersed with the
discourse-pragmatic features I think and i don’t think which, as a result of their
lack of prosodic variation, do not signal a high degree of epistemicity. Instead,
they perform a routinised organisational task: that of bracketing the assessment
sequence and marking the contrast between what Gabriel does and does not
believe to be the case.
148 The Structure of Discourse-Pragmatic Variation

(20) HP: But do you think they should sack Sven (..) [Sven (.) Göran
Gabriel: [E::h (.)]
HP: Eriksson?]
Gabriel: = I probably say yes. I think he’s took the team as far as he can
go. I think (.) he: picks his favourites. I doØ think he’s one
HP: yeah
Gabriel: = of the m
 anagers that says, “Right, you’re not playing well. In
bri- (.) we’re gonna bring in youngsters.”

When i don’t think acts as a frame to mark boundaries and transitions in dis-
course, as described above, it frequently co-occurs with other discourse-pragmatic
variables. These serve to signal precisely what kinds of transitions are under way
(e.g. contrast, elaboration, or conclusion). In all instances, the degree of confidence
conveyed by i don’t think is dependent on its prosodic realisation. By nature of
their positioning vis-à-vis the proposition they modify, post-positioned tokens of
i don’t think never mark transitions in discourse.
The data analysed here contain no tokens of i don’t think signalling its
propositional meaning: ‘it is not the case that I hold the opinion.’ Similar to other
corpora previously investigated for i don’t think usage (Bybee & Scheibman
1999), all tokens of i don’t think in the BwE corpus have pragmatic meanings.

5.4.2 Summary of functions

The preceding description and illustration of the range of functions performed by

i don’t think in the BwE interview data demonstrates well the construction’s in-
trinsic multifunctionality and interactional indispensability. In the interpersonal
domain, i don’t think serves to signal various degrees of confidence speakers
attach to their propositions. Moreover, the construction’s subjective epistemicity is
frequently exploited in this domain to signal politeness and promote good social
rapport. i don’t think serves intersubjectively to redress potential face-threats,
to soften the effect of disaligning turns, to attenuate or avoid blunt and abrupt
denials or rejections, and to signal agreement and affiliation. Importantly, the
preceding analysis has described for the first time the information- and discourse-
structuring functions performed by i don’t think in the interpersonal-textual
domain. While always cuing hearers to interpret following propositions and as-
sessments as inherently subjective, i don’t think can simultaneously act as a
frame to mark a range of discourse transitions (objective > subjective proposi-
tions; explications > provisions of assessments) and boundaries (onsets of opinion
sequences; shifts in topical development; shifts in opinion sequences). Appendix 4
 Chapter 5.  The construction i don’t think 149

lists these functional categories together with short descriptions of their charac-
teristic properties in order to illustrate more concisely that it is the construction’s
prosodic encoding, surrounding linguistic context, turn-positioning and imme-
diate as well as larger sequential context of occurrence that determine which
function it performs.
While the functional inventory of i don’t think outlined above accounts
for every token of the variable in the current dataset, it makes no claim to rep-
resent an exhaustive taxonomy of the functions performed by i don’t think
across varieties, speech events or genres. Additional functions may emerge from
analyses of other dialects, casual conversations or even scripted talk. What this
analysis has demonstrated, though, is that currently available descriptions of the
construction’s usage and functionality are inadequate and incomplete, and that
more work needs to be done to identify the full range of textual functions per-
formed by i don’t think.
However, the main aim of the qualitative data analysis was not to criticise ex-
isting accounts of the use of i don’t think. Rather, the aim was to make possible
inclusion of the factor group function in the quantitative analysis by isolating the
factors typifying it in the present dataset. Table 5.2 outlines the coding protocol
which arose from this exercise and guided the preparation of the data for quan-
titative analysis.

Table 5.2  Coding schema for i don’t think functions

function description
interpersonal epistemic marker signals tentativeness & uncertainty (hedge)
domain signals assertiveness & commitment (booster)
mitigator redresses potential face-threats
softens disalignments & disagreements
attenuates denials & rejections
affiliator displays affiliation & agreement
interpersonal- frame marks transitions & boundaries in discourse
textual domain structures opinion sequences
150 The Structure of Discourse-Pragmatic Variation

5.5 Quantitative analysis of i don’t think

The previous sections have identified the variants of i don’t think, the factor
groups that may constrain their distribution as well as the factors that constitute
these factor groups. This makes possible quantitative data analysis to uncover the
sociolinguistic mechanisms underlying the formal variability of i don’t think in
the data. In order to facilitate comparison of results across variables, the analysis
presented here parallels as far as possible that in Chapter 4. It begins with an out-
line of the overall distribution of variants and presentation of the marginal results
which reveal how the variants are distributed across the following independent
variables: individual speaker, speaker sex, age, opinion predicate, syntactic position,
function and complement type. The marginal results determine the configuration
of the mixed-effects multiple regression analysis which models the simultaneous
effect of sex, age, function and complement on the variable realisation of i don’t
think whilst removing from the outcome individual levels of variance. Inspection
of the results uncovered by the regression analysis is judiciously combined with
careful examination of variant distribution across fine-grained functions to reveal
the organisation of the variable grammar underlying the distribution of variants.
The results yield important new insights into the distribution of non-localised and
localised variants of i don’t think as well as the role of extra- and intra-linguistic
factors in discourse-pragmatic variation and change more generally.

5.5.1 Distributional analysis

Close auditory analysis of the data revealed that the 270 tokens of i don’t think
included in the variable context could be divided into four formal variants (see
Section 5.3.2). Figure 5.1 below shows their relative frequency. Roughly half of all
tokens of i don’t think in the data are realised with the full variant, I don’t think.
The reduced variant, I doØ think, constitutes slightly less than one third of all in-
stances of the variable. These results mirror those reported in Bybee & Scheibman
(1999) for American English: the variable is not consistently reduced even though
its meaning is categorically non-compositional (see Section 5.4.1). The localised
variants, I divn’t think and I dinnae think, are less frequent than the non-localised
variants. I divn’t think represents approximately one fifth of all tokens of the vari-
able. I dinnae think occurs only twice and is only used by one older female speaker
in the sample. Because of its negligible instantiation, this variant is removed from
consideration in the ensuing analysis.
Inspection of the distribution of the more frequently attested variants in the
data reveals high degrees of inter-speaker and inter-group variation in their use.
 Chapter 5.  The construction i don’t think 151

I dinnae think
I divn’t think
1%
18%
(N = 2)
(N = 48)

I don’t think
51%
(N = 138)

I doØ think
30%
(N = 82)

Figure 5.1  Overall distribution of the variants of i don’t think

Firstly, not all speakers in the sample employ all of the variants that have been
isolated in the BwE corpus. One speaker categorically uses the variant I doØ think.
Only 10 speakers make use of the variant I divn’t think. To account for these skewed
distributions and avoid obscuring important constraints on the formal variation
patterns in the data, the ensuing analysis proceeds as follows: the data contributed
by the categorical I doØ think-user (N = 3) are removed from consideration in any
quantifications of the data; only the ten speakers who employ I divn’t think in the
recorded data are included in quantifications of this variant across independent
variables.40 Secondly, not all variants are evenly distributed across social groups.
While social groups differ only minimally in the frequency with which they use
the reduced variant I doØ think, they differ more markedly in their frequency
of use of the full variant I don’t think and the localised variant I divn’t think. The
bars along the y-axis in Figure 5.2 show that the overall predominance of I don’t
think (dark grey bars) over I doØ think (black bars) noted in Figure 5.1 above is
present across sexes and age groups but is more marked amongst male speakers
and speakers from the middle age group. The frequency of I divn’t think (light grey
bars) also differs across social groups, especially across males and females. The
latter have markedly higher rates of I divn’t think than the former. However, the

40. As a result, the percentages provided for I divn’t think in Figures 5.2, 5.3 and 5.4 are derived
from consideration of a subsample of the i don’t think token numbers given next to the factors
listed along the x-axis.
 80
65%
70 62% (33) 61%
(71) (55)
55%
60 50% (10)
(37)
50 41% 40% 41%
(48) (27) (18) 36% don’t think
34%
40 32% 31% 32% (27) doØ think
(23)
(36) (37) (12) divn’t think
26%

Percentage
30 (23)
152 The Structure of Discourse-Pragmatic Variation

15%
20 (7)

10

0
male (N = 114) female (N = 118) old (N = 68) middle (N = 90) young (N = 74)
SEX AGE

Figure 5.2  Distribution of variants of i don’t think across speaker sex and age
(Figures in round brackets indicate raw token numbers. The results for I divn’t think are derived from data produced
by divn’t think-users only [N = 99].)
 Chapter 5.  The construction i don’t think 153

social distribution patterns displayed in Figure 5.2 must be interpreted with care.
The figures ignore intra-group differences in the use of the variable and all of its
variants; the figures for I divn’t think are based on a restricted and socially skewed
subsample of speakers. Nevertheless, Figure 5.2 suggests that provision of accurate
accounts of form variation in the use of i don’t think requires inclusion of social
factors in the multiple regression analysis.
The distributional results in Figure 5.2 are based on a reduced dataset of 232
tokens. It does not include tokens of I dinnae think (N = 2), tokens produced by an
invariable speaker (N = 3), tokens not coded for complement type and/or function-
ality due to insufficient context (inaudible turn continuation, interruption, trunca-
tion) (N = 21), tokens with the predicate suppose (N = 3), and tokens which occur in
post-posed position (N = 9).41 Tokens with suppose were removed from the analysis
due to their relative lack of formal variability (the dataset contain no tokens of
the reduced variant I doØ suppose) as well as their limited productivity which
makes impossible systematic and reliable analysis of the effect of predicate type
on variant selection. Post-posed tokens of i don’t think were removed from the
analysis due to their limited functional versatility (see Section 5.4.1) which might
skew form-function correlations in the data, and due to their negligible occur-
rence which makes impossible systematic and reliable analysis of formal variation
patterns across pre- and post-posed tokens of i don’t think. The implications of
the rarity of suppose-tokens and post-posed tokens for the construction’s unithood
and syntactic status are explored in the discussion of the results in Section 5.6.
In contrast to post-posed tokens of i don’t think which invariably perform
interpersonal functions and invariably modify a clause, pre-positioned tokens of
i don’t think perform interpersonal as well as interpersonal-textual functions
(see Section 5.4.1) and occur with clausal complements as well as the pro-form
so (see Section 5.3.1). The figures provided below the x-axis in Figure 5.3 demon-
strate that pre-positioned tokens of i don’t think are more than twice as likely to
perform interpersonal functions (epistemic marking, mitigation, affiliation) than
they are to perform interpersonal-textual functions (framing with a subjective
stance). They also demonstrate the overall predominance of clausal complements
over pro-form complements. The complete lack of pro-form complements with
interpersonal-textual tokens reflects these tokens’ role in marking transitions to
new information units, i.e., to discourse units that have not previously been men-
tioned or cannot be inferred from the surrounding context. The bars along the

41. Exclusion of the latter sets of tokens from the distributional analysis slightly alters some
percentages but does not affect the overall pattern of variant distribution outlined in Figures 5.2
and 5.3. The reasons for excluding these tokens from all quantitative analyses are outlined in
Chapter 4, footnote 28.
 70
56% 56%
60 48% (23) (35) 52%
(49) (35)
43%
50
(10)
33% 33% don’t think
40 29% (21) (22)
(30) doØ think
30 19% divn’t think

Percentage
(7)
20

10
154 The Structure of Discourse-Pragmatic Variation

0
clause pro-form clause
(N = 102) (N = 63) (N = 67)
INTERPERSONAL INTERP.-TEXT.
(71%, N = 165) (29%, N = 67)

Figure 5.3  Distribution of variants of i don’t think across functional domains and complement types
(Figures in round brackets indicate raw token numbers. The results for I divn’t think are derived from data produced
by divn’t think-users only [N = 99].)
 Chapter 5.  The construction i don’t think 155

y-axis in Figure 5.3 show the frequency of variants across functional domains and
complement types. Neither function nor complement affect the distribution of the
non-localised variants, I don’t think (dark grey bars) and I doØ think (black bars),
to any great extent: full variants always outweigh reduced variants. Conversely, the
localised variant, I divn’t think (light grey bars), manifests a stronger tendency to
occur with clausal complements than with the pro-form. The multiple regression
analysis in Section 5.5.2 will establish whether the effect of complement type on
the occurrence of I divn’t think is statistically significant.
Before proceeding to the multiple regression analyses, we must establish the
necessity of treating individual speaker as a random effect. This is accomplished
by way of comparing individuals’ normalised frequencies of i don’t think plot-
ted along the y-axis in Figure 5.4. Each black bullet represents one of the 18 male
speakers in the sample; each grey square represents one of the 18 female speakers
in the sample. Comparison of the 36 data points reveals dramatic differences in
the frequency of i don’t think usage both within and across age groups (arranged
from young to old along the x-axis). They range from zero to 25 tokens per 10,000
words in the youngest age group (on the left), from one to 34 in the middle age
group (in the centre), and from two to 19 in the oldest age group (on the right).
That highly divergent frequencies are particularly pervasive in the female data is
demonstrated in the fact that two thirds of the male data but only one third of
the female data are compressed into the range of zero to eight tokens per 10,000
words, and the fact that only four males but seven females have rates of i don’t
think higher than 15 per 10,000 words. The highly differential frequencies dictate
that the effect of individual speaker be accounted for in the multiple regression
analyses in Section 5.5.2. They may otherwise yield inaccurate results which over-
or under-estimate the effect of social factors on variant choice (see Chapter 2.3.3).

40
Frequency/10,000 words

35
30
25
male
20
female
15
10
5
0
15 25 35 45 55 65 75 85
speaker age

Figure 5.4  Normalised frequencies of i don’t think across individuals

156 The Structure of Discourse-Pragmatic Variation

5.5.2 Multivariate analysis

To establish the joint contribution of independent variables to the probability

of individual variants of i don’t think while accounting for the high levels of
inter-speaker variability discussed immediately above, mixed-effects multiple re-
gression analyses are conducted with Rbrul (see Chapter 2.3.3 for details about
Rbrul-based statistical analyses). Identical independent analyses for the three
most frequently instantiated variants (I don’t think, I doØ think, I divn’t think)
and close inspection of the direction of effect, statistical significance and relative
contribution of predictors across variants enable us to compare the mechanisms
of variability across variants and establish whether their occurrence is conditioned
by the same parameters. (Detailed explanations on how to interpret the results
yielded by the multivariate analyses in Table 5.3 are provided in Chapter 2.3.3).
Restricted token numbers preclude apparent-time analyses of the data which are
needed to identify any ongoing changes to the linguistic conditioning of variant
selection. However, inclusion of age as a factor group in the analyses enables us
to ascertain whether the probability of variant occurrence is changing in apparent
time. In addition to age and sex, the analyses include two internal factor groups:
function and complement. The factor group function is constituted by the factors
interpersonal and interpersonal-textual. Running the regression analyses across
broad functional categories is necessitated by low token numbers and motivated
by the hypotheses about form-function correlations outlined in Section 5.3.3. The
factor group complement is constituted by the factors clause and pro-form. In
order to avoid empty cells, the former conflates tokens of i don’t think followed
by zero-complementisers and tokens followed by that-complementisers. The rare
occurrence of that-complementisers (N = 2) is discussed in Section 5.6. Finally,
application of mixed-effects modelling and inclusion of speaker as a random
effect in the regressions removes from the outcome of the analyses individual
levels of variance.
Table 5.3 displays the results of three independent multivariate analyses of the
contribution of age, sex, function and complement to the choice of I don’t think,
I doØ think and I divn’t think in the BwE interview data. The variant listed at the
top of a column was the one chosen as the application value for this particular run,
with the other two variants included in the run as non-application values. All vari-
able speakers in the sample were included in the runs for I don’t think and I doØ
think; only the ten speakers who use I divn’t think in the recordings were included
in the run for this variant. Because interpersonal-textual tokens only ever occur
with clausal complements (see Figure 5.3 above), these tokens were excluded from
consideration in the runs for complement. (This explains the lower token numbers
in this factor group.)


Table 5.3  Contribution of external and internal factors to the probability of I don’t think, I doØ think and I divn’t think
(The application value at the top of each column is modelled in opposition to all tokens of the other variants.)

I don’t think I doØ think I divn’t think

input prob. 0.513 0.315 0.297
total N 232 232 99
deviance 259.574 253.744 97.332
factor % N log odds factor % N log odds factor % N log odds
weights weights weights
sex not significant not significant p = 0.00196
female [.453] 40.7 118 [−0.190] [.410] 31.4 118 [−0.365] .769 64.7 51 −1.201
male [.547] 62.3 114 −[0.190] [.590] 31.6 114 −[0.365] .231 14.6 48 −1.201
range .538
age not significant not significant not significant
old [.321] 39.7   68 [−0.751] [.536] 33.8   68 −[0.146] [.490] 40.9 44 [−0.039]
middle [.678] 61.1   90 −[0.754] [.347] 25.6   90 [−0.632] [.456] 31.6 38 [−0.176]
young [.501] 50.0   74 [−0.006] [.619] 36.5   74 −[0.468] [.553] 55.3 17 −[0.214]
range
function not significant not significant not significant
interpersonal [.525] 50.9 165 −  [0.99] [.442] 31.2 165 [−0.232] [.547] 43.5 69 −[0.190]
interp.-text. [.475] 52.2   67   [−0.99] [.558] 32.0   67 −[0.232] [.453] 33.3 30 [−0.190]
range
complement not significant not significant p = 0.0105
clause [.488] 48.0 102 [−0.049] [.395] 29.4 102 [−0.428] .732 56.1 41 −[1.006]
pro-form [.512] 55.6   63 −[0.049] [.605] 33.3   63 −[0.428] .268 19.4 36 [−1.006]
range .464
speaker random st. dv 1.741 random st. dv 1.57 random st. dv 0.238
Chapter 5.  The construction i don’t think 157
158 The Structure of Discourse-Pragmatic Variation

Despite the patterns depicted in Figure 5.2 and the divergent factor weights
given in Table 5.3 which associate I don’t think with male speakers and speakers
from the middle age group, the results of the regression analysis reveal that social
factors do not make a statistically significant contribution to the occurrence of the
full variant. (The appearance of social effects is created by unbalanced data.) As
predicted by the distributional analyses in Section 5.5.1, internal factors are not
implicated in the distribution of I don’t think either. The variant occurs with similar
probabilities across broad functional categories and both complement types. (The
interaction effect in the factor group function is due to high degrees of individual
variation in the distribution of I don’t think across functional categories. These
differences are likely to be a fabrication of low token numbers in the data.) The
results shown in Table 5.3 thus reject the hypothesis formulated in Section 5.3.3
that tokens followed by the pro-form so are less likely to be formally reduced by
virtue of the negative constituting the new information contained in i don’t think
so. In sum, none of the factor groups included in the analysis for I don’t think have
been selected as making a significant contribution to its occurrence.
The same obtains for I doØ think. The results of the regression analysis shown
in Table 5.3 are in line with the patterns depicted in Figures 5.2 and 5.3. Neither
external nor internal factors significantly affect the distribution of the reduced
variant. The specious favouring effects for male and young speakers are not real
when all predictors are considered together and within-group differences in variant
choice and token frequency are accounted for. The factor weights for age fluctuate
but they are inconsistent with any inferences of change in progress. Due to a combi-
nation of limited data and intra-speaker variation, complement type does not affect
the occurrence of I doØ think despite divergent factor weights. Finally, the result
that I doØ think occurs with roughly similar probabilities across interpersonal and
interpersonal-textual functions rejects the hypothesis formulated in Section 5.3.3:
that tokens which function across discourse domains to act simultaneously as epis-
temic markers and discourse frames are more likely to occur with don’t-reduction
than tokens which function in the interpersonal domain only.
In the case of the localised variant, I divn’t think, the multivariate results in
Table 5.3 provide robust confirmation of the marginal results shown in Figures 5.2
and 5.3. Its occurrence is more constrained than that of the non-localised variants.
As indicated by the range values, speaker sex makes the most important contribu-
tion to the occurrence of I divn’t think, with the six female users of this variant in
the data exhibiting a favouring effect over the four male users. Thus, while two
thirds of the female speakers interviewed for this project never employ the variant
I divn’t think in the recordings, those who do employ it do so with remarkable fre-
quency. Complement type is the second strongest constraint on the use of I divn’t
think. Clausal complements strongly favour its use while the pro-form disfavours
 Chapter 5.  The construction i don’t think 159

it. Age and discourse-pragmatic function do not make a significant contribution

to the occurrence of I divn’t think. Its usage is not associated with a particular age
group in the speaker sample and it occurs with similar probabilities across broad
functional categories.
The results discussed above suggest that in contrast to the localised variant,
I divn’t think, the occurrence of the non-localised variants, I don’t think and I doØ
think, is largely unconstrained by the factor groups included in the multivariate
analyses. They have refuted hypotheses about the impact of social factors on the
choice of full and reduced variants as well as hypotheses that full variants may
be favoured with the pro-form and reduced variants with interpersonal-textual
functions (see Section 5.3.3). However, we have not yet tested the hypothesis that
the distribution of variants may be conditioned by more fine-grained functional
differences in their use (see Section 5.3.3). Figure 5.5 explores this possibility. It
lists along the x-axis the range of micro-functions established for i don’t think in
Section 5.4: hedges (tokens attenuating propositions), boosters (tokens strengthen-
ing propositions), affiliators (tokens displaying affiliation or agreement), mitigators
(tokens redressing face-threats, signalling disalignments, or attenuating denials/
rejections), and frames (tokens marking transitions/boundaries or structuring
opinion sequences). The frequency with which each variant occurs across these
functions is indicated by the bars along the y-axis.
Although low token numbers preclude firm conclusions to be drawn, the re-
sults suggest that the occurrence of I divn’t think (light grey bars) is not greatly
constrained by micro-functional distinctions. Disregarding its absence from the
category of boosters, the variant is fairly evenly distributed across the categories
listed on the x-axis, though it is somewhat less frequent in framing functions. By
contrast, I don’t think (dark grey bars) and I doØ think (black bars) display some
function-specific patterning in line with Yaeger-Dror’s (1985, 1997) Cognitive
Prominence Principle (henceforth CPP) and Social Agreement Principle (hence-
forth SAP). These principles predict that negatives in interactional registers are:
(1) uncontracted and prominent in terms of pitch, amplitude and/or duration
when they carry important focal information (CPP) or show support of inter-
locutors (SAP); (2) contracted and non-prominent in terms of prosodic emphasis
when they are non-supportive of co-conversationalists (SAP). Evidence for the
operation of these principles on the realisation of i don’t think is provided by
the three categories in the middle of Figure 5.5. As predicted by the CPP, the
negative particle in boosters is always fully enunciated because it is the bearer of
important focal information, i.e., that something is not the case. As predicted by
the SAP, the clitic in affiliators tends to be fully enunciated because the negative is
supportive of interlocutors; conversely, the negative clitic in mitigators tends to be
reduced to soften the unwelcome effect of disagreements or overt rejections. My
 100%
(8)
100
90
80
64%
70 (7)
53% 52%
60 53% 50% (35)
47% (8) don’t think
(59) (3)
50 41% (16) doØ think
(19) 34%
30% 33% divn’t think
40 29%

Percentage
(10)
(34) (10) (22)
160 The Structure of Discourse-Pragmatic Variation

30
20 9%
(1)
10 0% 0%
(0) (0)
0
hedge booster affiliator mitigator frame
(48%, N = 112) (3%, N = 8) (5%, N = 11) (15%, N = 34) (29%, N = 67)

Figure 5.5  Distribution of variants of i don’t think across micro-functions

(Figures in brackets indicate raw token numbers. The results for I divn’t think are derived from data produced
by divn’t think-users only [N = 99].)
 Chapter 5.  The construction i don’t think 161

suggestion that the variation between full and reduced variants of i don’t think
is conditioned by Yaeger-Dror’s (1985, 1997) CPP and SAP is necessarily tentative
due to restricted data precluding statistical testing. Nonetheless, the results indicate
that the non-localised variants of i don’t think display some function-specific
patterning at the micro-level of analysis.
The operation in the data of the CPP and SAP also helps explain why tokens
acting as frames in discourse are not consistently reduced, as initially hypoth-
esised. When these tokens mark transitions in discourse, they can convey either
a neutral, tentative or assertive subjective stance towards the discourse unit they
frame (see Examples (16), (17) and (18b) above in that order). Alternatively, they
can also mitigate the proposition they frame (see Example (19) in Section 5.4.1).
Re-inspection of these tokens demonstrates that interactional frames that unam-
biguously signal an assertive stance tend to be non-reduced while those that si-
multaneously perform some mitigating task tend to be reduced. These patterns
demonstrate that even with multifunctional tokens that operate across discourse
domains, it is the function i don’t think performs in the interpersonal domain
that tends to determine these tokens’ realisation.
Finally, re-inspection of the hedging tokens of i don’t think suggests that
prosodic factors may be implicated in the formal variation of non-localised vari-
ants as well. Bybee (2003: 616–617) points out that automatisation and meaning
changes affect the prosody of grammaticalising words and constructions. Together
with the decrease in the referential load of the grammaticalising items, the cogni-
tive representation of multi-word constructions as fused storage and processing
units leads to prosodic reduction. In frequent NP-negative do-V collocations such
as i don’t think, loss of stress affects the medial syllable, i.e., negative periphras-
tic do, which in turn leads to reduction of this syllable, as observed above (see
also Bybee & Scheibman 1999: 581–582). However, some non-boosting tokens
of i don’t think in the data carry pitch prominence on don’t (see, for example,
Extract (15b) in Section 5.4.1). Raised pitch on negative do functions to signal
high degrees of uncertainty. It could therefore be argued that the signalling of
fine-grained nuances of meaning through prosody affects the formal variation
in the data and contributes to the lack of consistent attrition of i don’t think.
A more thorough investigation of i don’t think’s prosodic realisation and of the
degrees of uncertainty conveyed by the variable than was possible within the con-
straints of this project will confirm the extent of this effect on formal variability
in discourse-pragmatics.
162 The Structure of Discourse-Pragmatic Variation

5.5.3 Summary of results

The key insights to emerge from the quantitative analysis into the formal variability
of i don’t think include the following:
– Despite its non-compositional status in the data, i don’t think is not consist-
ently reduced: Full variants of i don’t think far outweigh reduced variants,
even where the construction is used to perform routinised organisational
tasks.
– The variation between full and reduced variants of i don’t think is exploited
to signal micro-level functional distinctions: The reduced variant I doØ think
is associated with uses where the negative is non-supportive of interlocu-
tors. The full variant I don’t think is associated with uses where the nega-
tive signals support of interlocutors or carries important focal information.
Reduction is also impeded by pitch prominence on don’t to signal high de-
grees of tentativeness.
– Variants with different geographical distributions do not share the same variable
grammar: While non-localised variants evince some function-specific pattern-
ing in the data, the localised variant I divn’t think largely eschews association
with functional categories. Its use is strongly constrained by those predictors
that have no effect on the occurrence of I don’t think and I doØ think, namely
speaker sex and complement type.
Inspection of variant distribution across individual speakers and cohorts of speak-
ers, inclusion of individual speaker as a random effect in the multivariate analyses,
and comparison of the frequency of i don’t think across and within social cohorts
yield another key insight into the distribution of i don’t think in BwE:
– The frequency and form of i don’t think exhibit a considerable amount of
inter-group and inter-speaker variability: Social cohorts and individual speakers
differ in the frequency with which they use the construction i don’t think
as well as their preference for individual variants. This is particularly true of
the variant I divn’t think which is only used by a subsample of ten speakers
recorded for this project.

5.6 Discussion

This chapter set out to uncover the sociolinguistic mechanisms giving rise to the
formal variability of i don’t think in BwE, a peripheral variety of British English
which features both non-localised and localised variants of the targeted variable.
Principled and accountable analyses of the construction’s usage and distribution
 Chapter 5.  The construction i don’t think 163

in the data have provided important new insights into its functional versatility as
well as the constraints on its formal variability. This section discusses these insights,
compares them to those obtained in previous studies of i don’t think-variation
and don’t-reduction in other varieties, and explores several indices of unithood
to support claims that i don’t think constitutes a routinised discourse formula.
The painstaking qualitative data analysis in Section 5.4 has established the func-
tions performed by i don’t think in the BwE data in order to make possible testing
hypotheses that function might be implicated in the distribution of the construc-
tion’s formal variants. Rigorous attempts to account for the usage of every token
of the variable in the data revealed that i don’t think performs a wider range of
functions than previously described in the literature. Beyond signalling reduced
epistemicity and mitigating the illocutionary force of dispreferred interactional
moves, i don’t think functions in the BwE data to boost the degree of commit-
ment expressed, to implement interactionally preferred responses, and to mark
transitions and boundaries in discourse. The distributional analyses in Section 5.5
showed the interactional importance of these previously undocumented functions.
Although the majority of tokens of i don’t think in the BwE corpus function to
hedge propositions, more than a quarter of tokens also perform some routinised
work in the organisation of discourse. Future studies will demonstrate whether the
usage of i don’t think as a textual frame is indicative of its ongoing grammaticali-
sation whereby the construction becomes increasingly desemanticised, and its epis-
temic and mitigating meanings gradually fade away (see, for example, Narrog [2012]
who argues that textual functions can constitute endpoints of semantic-pragmatic
change). In the present data, the subjective epistemicity conventionally associated
with i don’t think is very much part of its framing uses and there is no indication
that its use as a hedging or mitigating device is decreasing and its use as a framing
signal is increasing in apparent time. However, this may be due to the nature of the
data on which the present analysis is based. The construction’s epistemic meanings
may be less prevalent in recording contexts that focus on the communication of
referential information or the construction of coherent narratives than those that
focus on the elicitation of personal opinions and subjective evaluations.
In addition to highlighting the shortcomings of existing functional taxon-
omies of i don’t think usage, isolating fine-grained and previously undocu-
mented functions of i don’t think has proved crucial to accounting for the
distribution of its non-localised variants. As was the case in Bybee & Scheibman’s
(1999) American English data, in the BwE data, i don’t think is not consist-
ently reduced either, even though the construction is exclusively used to signal
non-compositional meanings. Close investigation of the distribution of full and
reduced variants of i don’t think across the fine-grained functions isolated in
this study revealed that far from being randomly distributed in discourse, I don’t
164 The Structure of Discourse-Pragmatic Variation

think and I doØ think evince some function-specific patterning on a micro-level

of analysis which tallies with Yaeger-Dror’s (1985, 1997) CPP and SAP. Full vari-
ants are associated with contexts where the negative is the bearer of important
focal information (boosters) or where it is interactionally preferred (aligning
turns); reduced variants are associated with contexts in which the negative is
interactionally dispreferred (face-threats, disaligning turns, outright rejections/
denials). These constraints also affect the distribution of those tokens which in
addition to signalling epistemicity act as frames at transitions or boundaries in
discourse. Rather than being determined by their status as conventionalised dis-
course frames (see Section 5.3.3), the realisation of these tokens is at least to some
extent dependent on whether they hedge, boost or mitigate the proposition they
frame. These findings demonstrate the complex nature of form-meaning relations
in discourse and counsel scholars against assuming a straightforward connection
between discourse function and phonological substance. This view is supported
by the observation that don’t-reduction in i don’t think is sometimes hampered
by the construction’s prosodic encoding with raised pitch on don’t which signals
high degrees of tentativeness. While the patterns outlined above may not explain
the full extent of formal variation observed in the data, they demonstrate that the
lack of consistent reduction of non-compositional i don’t think tokens is at least
to some extent interactionally motivated.
Inclusion of speaker sex and age in the multivariate analyses in Section 5.5
revealed that use of the non-localised variants of i don’t think is not socially
indexical. Although the frequency of full and reduced variants fluctuates slightly
across speakers of different sexes and ages, the differences do not reach statistical
significance and are not compatible with inferences of change in progress. These
results contrast with those reported for secondary contraction in Yorkshire English
where the deletion or reduction of the negative clitic particle -n’t after vowels (e.g.
don’t, can’t, won’t) is favoured by females and is increasing in apparent time (Petyt
1978; Richards 2008; Whisker 2007; Whisker-Taylor in prep.). The differential
results are likely due either to regional dialect variation or to methodological dif-
ferences. The studies into secondary contraction listed above did not differentiate
between individual auxiliary types or between high- and low-frequency construc-
tions containing negative do. The fact that there is no indication of any apparent-
time monotonic increase or decrease of full and reduced variants of i don’t think
in the BwE data suggests that the discourse-functional distribution of variants
outlined above is an inherent part of the linguistic system of this dialect rather than
a reflection of an unstable and transitional state between a system of primarily full
variants and a system of primarily reduced variants.
The results in Section 5.5 further reveal that the occurrence of the localised
variant I divn’t think is conditioned by different parameters than the occurrence
 Chapter 5.  The construction i don’t think 165

of the non-localised variants. Amongst the subsample of speakers employing the

localised variant in the BwE corpus, it is females who strongly favour its use. The
female predominance in the use of the localised variant contradicts the pattern
widely reported for localised phonological variants which are generally favoured
by men (see, for example, Llamas 2007b; Milroy et al. 1994). Potential causes for
this unexpected social pattern as well as for the low number of speakers using the
variant are explored in Chapter 7. Whatever the nature of the underlying causes, I
divn’t think is a clear candidate for marking social differentiation in the BwE data.
Another pattern that differentiates the localised variant from the non-localised var-
iants is the fact that the occurrence of I divn’t think is not conditioned by function-
ality sensu stricto. Bearing in mind the low token numbers, I divn’t think is roughly
evenly distributed across broad and narrow functional categories. However, after
gender, the second strongest constraint on the use of I divn’t think is complement
type. As shown in Table 5.3 above, I divn’t think is favoured with clausal comple-
ments and disfavoured with the pro-form so. Considering the cohesive function
of the pro-form (see Section 5.3.1), these patterns suggest that I divn’t think is not
primarily used to negate or reject previously stated or inferred propositions but is
used more often to negate discourse-new information. In a sense, then, the occur-
rence of I divn’t think is also constrained by interactional factors.
The finding that the occurrence of variants with different geographical distri-
butions is conditioned by different parameters is supported by closer comparison
of their underlying grammars in Table 5.3 above. Because small datasets such as
the one analysed here are less likely to yield statistically significant results than
bigger data­sets, the hierarchy of constraints constituting non-significant factor
groups can nevertheless be construed as representing the grammar underlying the
surface variability, and thus form the basis for comparing the underlying gram-
mar of individual variants (Poplack & Tagliamonte 2001: 93). The comparison
reveals that non-localised­and localised variants of i don’t think stand in almost
complementary social and linguistic distribution. While I don’t think and I doØ
think are slightly more likely to occur in male speech, in interpersonal-textual uses
and with the pro-form so, I divn’t think is slightly more likely to occur in female
speech, in interpersonal uses and with clausal complements. This demonstrates
that discourse-pragmatic variants with differential geographical distributions
may carry different social and linguistic loads and may therefore not be freely
interchangeable.
Whatever the form selected for encoding i don’t think in BwE, the data
provide compelling evidence that i don’t think constitutes a conventionalised
discourse formula with a fairly fixed internal linguistic structure and a fairly robust
unithood status. As pointed out in Section 5.5.1, I-negative do-opinion predi-
cate constructions are virtually categorically formed with think (99%, N = 267).
166 The Structure of Discourse-Pragmatic Variation

Constructions with suppose are exceptionally rare in the data (1%, N = 3); construc-
tions with other opinion predicates (e.g. believe, find) are non-existent. Skewed
distributions of lexical items are characteristic of frequent constructions and have
been reported for negative and positive epistemic stance constructions elsewhere
(Fetzer 2011; Kärkkäinen 2003, 2007). Together with the fact that in the current
dataset i don’t think is never modified by intervening adverbs, these patterns
demonstrate the lexical formulaicity and internal fixation of negative epistemic
marking with subject-negative-opinion predicate constructions in BwE.
The evidence provided above can be taken as indicative of the non-composi-
tional status of i don’t think, a claim that is supported by the 21: 1 ratio of i don’t
think to ‘I think not-p’ constructions in the data (N = 270 vs. N = 13). Bublitz
(1992) reports similar results and argues that the preference for negative raising
with opinion predicates reflects its usefulness as an attenuation and mitigation
strategy in contexts where speakers strive to develop and maintain good social
relationships. Contra Bublitz, I would argue that the marked preponderance of
i don’t think over ‘I think not-p’ reflects a cognitively rather than interactionally
motivated preference. When speakers wish to signal a subjective epistemic stance
towards negative propositions, they access i don’t think as a whole from memory
rather than constructing a negative subjective stance from individual lexical items
at the moment of speaking. Rare instances in the data of ‘I think not-p’ can be ac-
counted for by interactional factors hampering negative raising. Firstly, ‘I think
not-p’ tends to occur as a result of planning difficulties. In (21) below, the unfilled
pause after I think it was and the following recast of the proposition in negative
terms suggest that the speaker did not know at the outset that what would follow
was a negative proposition (it wasn’t done properly). Consequently, the negative
could not be raised to the preceding matrix clause but instead had to occur with
the proposition that it was construed with.
(21) But to me, I think it was (..) it wasn’t done properly.

Secondly, the occurrence of ‘I think not-p’ in the data can be explained on the
grounds that it allows the addition of more layers of modification than is possible
with i don’t think, as illustrated in (22):
(22) a. I think I probably wouldn’t change the way I talk.
b. *I don’t think I probably would change the way I talk.

The preceding discussion demonstrates that despite the huge variety of available
linguistic resources, negative epistemic stance marking in BwE is highly regular-
ised and routinised in terms of the linguistic forms and syntactic structures used.
The high degree of internal fixation differentiates i don’t think from positive po-
larity epistemic stance markers which are far more variable in terms of the opinion
 Chapter 5.  The construction i don’t think 167

predicates used (Kärkkäinen 2003: 37) and also allow a much higher degree of
TAM variation than we encounter with negative epistemic stance markers in BwE
(see van Bogaert 2011). Moreover, i don’t think enjoys less syntactic freedom
and positional mobility than positive polarity constructions such as I think, I guess,
etc. While I think and its variants variably occur utterance-initially, -medially or
-finally (see, for example, Thompson & Mulac 1991b), i don’t think never oc-
curs utterance-medially and only rarely utterance-finally (see Section 5.5.1). The
strong preference for utterance-initial pre-positioning is largely interactionally
motivated to secure alignment early on and guide hearers’ interpretation of what
follows. Also, post-positioned modification with i don’t think is only possible
if the qualified proposition has been independently negated, as in (23a). Simply
transporting i don’t think from one utterance position to another would result
in grammatically incorrect utterances (23b) or meaning changes (23c).
(23) a. I don’t think I mind it. → I don’t mind it, I don’t think.
b. I don’t think I mind it. → *I mind it, I don’t think.
c. I don’t mind it, I don’t think. ≠ I don’t think I don’t mind it.

What i don’t think-p constructions share with I think-p constructions, though,

is their near-consistent absence of that-complementisers (see Section 5.5.2). In the
literature, that is generally described as demarcating the boundary between two
clauses with referential content, while zero occurs when information-carrying fi-
nite clauses are preceded by an epistemic frame (Kearns 2007; Thompson & Mulac
1991a, b; Torres Cacoullos & Walker 2011: 234). The near-categorical use of zero-
complementisers in the current data can thus be interpreted as reflecting the syn-
tactic and informational status of i don’t think-p constructions. When i don’t
think performs its conventionalised role as a discourse-pragmatic variable, it con-
stitutes an epistemic frame with scope over the following information-carrying
finite clause rather than a matrix clause introducing a subordinate complement
clause (see Thompson 2002). The fact that the negative in i don’t think, through
transferred negation, takes scope over the following proposition lends additional
support to the non-existence of a clause boundary.
The preceding discussion was based on the assumption that the construc-
tion’s modal usage derived through semantic bleaching and conventionalisation
of implicatures from its propositional reading ‘I do not hold the opinion that.’ In
other words, the modal or pragmatic meaning of i don’t think was taken to be
a grammaticalised, weak reading of negative raising (Horn 1978). While some
scholars would argue that formal reduction is a necessary concomitant of gram-
maticalisation (e.g. Frank-Job 2006), the discussion above has identified several
structural indices of unithood to demonstrate that i don’t think qualifies as
a non-compositional, routinised discourse formula despite its lack of consistent
168 The Structure of Discourse-Pragmatic Variation

reduction of don’t. While the data do not furnish any evidence of ongoing gram-
maticalisation, the synchronic distribution of variants shows all the signs of a
grammaticalisation outcome: the synchronic co-existence of full and reduced vari-
ants; the construction’s non-compositional usage and polysemic layering; and its
syntactically reanalysed status as an epistemic frame rather than a complement-
taking matrix clause.

5.7 Conclusion

The analysis presented in this chapter has confirmed the grammaticalised status
of negative raising with opinion predicates (see Horn 1978) and the lack of con-
sistent don’t-reduction in the pragmatic usage of i don’t think (see Bybee &
Scheibman 1999). Systematic qualitative and quantitative analyses of the data have
revealed several previously undocumented findings: that i don’t think constitutes
a highly conventionalised discourse formula with a wide range of functions in the
interpersonal and textual domains of discourse; that the variation between its full
and reduced variants is conditioned by fine-grained functional differences in the
construction’s use as well as by prosodic factors; and that the occurrence of its
localised variant is socially indexical.
Because of limited data, no apparent-time analyses could be conducted to
reveal how stable the organisation of the variable grammar is. Larger datasets and
comparable longitudinal data from this and other varieties are needed to determine
how stable the fine-grained functional patterning of full and reduced variants de-
scribed in Section 5.5.2 is, how widespread the formal variation patterns outlined
in Section 5.5.2 are, and whether the lexical formulaicity and internal fixation of
i don’t think documented in Section 5.6 is characteristic of (northern) (British)
English more generally.
 chapter 6

Negative polarity question tags (neg-tags)*

6.1 Introduction

A considerable amount of scholarly attention has been devoted to the investigation

of English tag questions (henceforth TQs), which typically consist of an anchor
clause to which an interrogative tag is appended. Scholars generally distinguish
two types of tags: (i) invariant tags such as right?, eh?, isn’t that so?, don’t you think?
which can be attached to any anchor irrespective of its grammatical features; and
(ii) canonical tags such as isn’t it?, can’t you?, won’t we? which generally bear a form-
dependency on the preceding anchor (see further Section 6.3.2). Beyond providing
detailed accounts of their syntactic-semantic properties (Cattell 1973; Hudson
1975; McGregor 1995) and thorough descriptions of their functional versatility
(Algeo 1988, 1990; Andersen 2001; Coates 1996: 191–200; Erman 1998; Holmes
1982, 1984a, 1987, 1995: 79–86; Kimps 2007; Millar & Brown 1979), previous re-
search on English TQs has yielded important insights into the extent to which the
frequency and/or function of question tags varies across speaker sex, age, com-
munities of practice, speaker role or power, and geographical space (Cameron et al.
1988; Cheshire 1981; Coates 1987; Dubois & Crouch 1975; Fishman 1983; Harris
1984; Hoffmann et al. forthc.; Holmes 1984a, 1987, 1995; Meyerhoff 1994; Moore &
Podesva 2009; Tottie & Hoffmann 2006). To complete the present investigation into
the formal variation of negative polarity discourse-pragmatic variables, this chap-
ter builds on the small number of studies concerned with variation in the form of
canonical and canonically-derived tag variants (Cheshire 1981; Moore & Podesva
2009) as well as those concerned with the origin, spread and grammaticalisation
of the innovating tag variant innit (Andersen 2001; Cheshire et al. 2005: 155–159;
Erman 1998; Krug 1998). It thus aims to: uncover the factors conditioning the
formal variation in the BwE neg-tag system illustrated in (1) and introduced in
more detail in Section 6.3.2; establish the entrenchment in the BwE data of the
innovating variant innit; and assess the extent to which the BwE neg-tag system
is affected by ongoing grammaticalisation.

* Throughout this chapter, the label ‘tag question’ is applied to the combination of anchor
clause (e.g. It’s nice) and question tag (e.g. isn’t it). The label ‘(question) tag’ is used to refer to
the question tag element of TQ constructions only.
170 The Structure of Discourse-Pragmatic Variation

(1) a. Jane: I mean really, we could have a conversation, (h) if somebody

came in frae London (.) (h) o:r no even as far [south as Lon-
Helen: [excuse me]
Jane: = don.] And we could baffle them. Couldn’t we.
HP: mhm mhm
Jane: = [They wouldn’t know what we was]
Helen: [Oh aye. Wouldn’t know what] we was talking about.
Jane: (h) They [wouldn’t know] what we was talking about.
HP: [mhm mhm]
Yeah. Uh-huh. (h) And do you think that’s a pity if i- why why
would it be such a pity if if it died out.
Jane: (h) Because I think it’s one of the em (..) no (a perk). (h) But
it’s one of the Berwick (.) parts of the Berwick culture.
HP: mhm
Jane: And history. [In’t it.]
HP: yeah
b. Luke: Aye, it’s no as bad as Eyemouth, though. [I mean,] you get
HP: [yeah]
Luke: = drugs everywhere, [divn’t you.] But it’s no as bad as Eyemouth,
HP: [yeah] yeah
Luke: = like.
c. Jane: Aye, but I think, for the younger ones they call their under-
pants kecks. But we called them trousers, did we no?
d. Daniel: And Britain e:h (.) had to had to declare war. That was the
thing. It’s absolutely crazy, innit. If he had stopped at (..) I
HP: yeah
Daniel: = mean, after the First World War th- they didn’t give Germa-
ny enough. They should have given the land back. And Britain
was for that. E::h (..) but for the Germans, <when they
HP: mhm yeah
Daniel: = when they got stronger they wanted to take back their own
lands.> (h) And that’s understandable, [th-] we would have
HP: [yeah]
Daniel: = been the same. But th- he went further than that, didn’t he?
e. Gregory: And he speaks completely different to me, din he.
June: [Mhm. He does. Aye.] Uh-huh.
HP: [Mhm. Mhm.]
Gregory: And he was my best man at my wedding and [you’d] (.)
June: [Aye.]
Gregory: = think we was poles apart, wouldn’t you.
 Chapter 6.  Negative polarity question tags (neg-tags) 171

Systematic analysis of the BwE neg-tag system reveals a dramatic social and func-
tional split between canonical tag variants (e.g. isn’t it, couldn’t we, didn’t he) and
the canonically-derived tag variant innit. The results also show that while innit is
more grammaticalised than canonical tags in terms of functionality, this is not the
case with regard to analogical levelling. Neither innit nor its co-variant isn’t it are
regularly used outside their original context of use. Comparison of these results
with those reported for innit in contemporary London English (Andersen 2001)
provides important new insights into the social, geographical and linguistic dif-
fusion of innovating discourse-pragmatic variants which complement and extend
the insights uncovered in Chapters 4 and 5 into the sociolinguistic mechanisms
of discourse-pragmatic variation and change.
This chapter is organised in the same way as Chapters 4 and 5. Section 6.2 of-
fers a review of the literature which has provided the impetus for the current analy-
sis of neg-tags, and outlines in more detail the aims of this chapter. Section 6.3 sets
out how the variable context was circumscribed. This section and Section 6.4 also
describe how the dependent and independent variables were coded. Section 6.5
presents the results of the quantitative analysis and reveals the factors constrain-
ing observed patterns of variability in the choice of tag variants. The results are
discussed in Section 6.6. Finally, Section 6.7 is the conclusion.

6.2 Previous research on question tags

As pointed out above, English TQs have received a fair amount of scholarly at-
tention. This section reviews the literature on TQs that is directly relevant to the
present investigation. The focus is on studies that have described question tags’
functional properties, tag variants’ distributional characteristics, and the tag sys-
tem’s increasing grammaticalisation. There follows at the end of this section an
outline of how the present investigation constitutes a development beyond previ-
ous research by studying the neg-tag system as a whole and including social,
functional as well as syntactic-semantic factors in the quantitative analysis of both
canonical and canonically-derived tag variants.
172 The Structure of Discourse-Pragmatic Variation

6.2.1 Qualitative studies: Discourse-pragmatic functions of question tags42

Because they have been studied within a range of theoretical frameworks, a mul-
titude of functional taxonomies have been proposed for English question tags
(for the most comprehensive ones, see Holmes 1982; Kimps 2007). The follow-
ing section is limited largely to reviewing those accounts which have impacted
directly on establishing the functions performed by neg-tags in the BwE data
in Section 6.4 as well as those which are relevant to the discussion of the results
in Section 6.6.
Academic interest in the functionality of question tags was triggered by Lakoff ’s
(1973) intuitive account of TQs which distinguished between legitimate and ille-
gitimate uses of tags. The former seek confirmation or verification of propositions
in which speakers lack confidence; the latter are attached to propositions of which
speakers are sure, thus signalling lack of assertiveness rather than lack of knowl-
edge. Holmes’s (1982, 1984a, 1987) in-depth analysis of TQs in a corpus of tape-
recorded speech data challenges Lakoff ’s taxonomy and demonstrates that Lakoff ’s
illegitimate tags do in fact perform important interpersonal functions in discourse.
They express solidarity and function to encourage or make easier addressees’ entry
into or contribution to the discourse (facilitative tags); and they work towards miti-
gating the force of negatively affective speech acts such as directives, complaints
or criticisms (softening tags). In her later work, Holmes (1995: 80–82) qualifies
her earlier proposal that tags always serve as hedges attenuating the illocutionary
force of preceding propositions, and shows that tags can also be confrontational
and serve to boost the force of potential face-threats (challenging tags).
The view that tags constitute politeness as well as impoliteness devices is sup-
ported by others. The tag functions described by Algeo (1988, 1990) fall into three
broad categories distinguished by their (im)politeness impact: tags seeking verifi-
cation or corroboration are polite; punctuational tags serving to highlight speakers’
propositions are neutral; and peremptory or aggressive tags signalling hostility
are impolite. Cheshire (1981: 375–376) also identifies tags which serve to convey
a hostile, aggressive or sarcastic overtone. Algeo and Cheshire concur that what
distinguishes hostile or impolite tags from polite tags is the fact that the former
are not response-eliciting. In (2) below, Jenny was not in a position to know that
Cathy was going on holiday, and Cathy knew that Jenny did not know. The sincerity
conditions on interrogatives (“The speaker believes that the hearer knows at least as
well as the speaker does whether the proposition is true or false,” Hudson 1975: 12)

42. Because the functional repertoire of neg-tags is illustrated in detail in Section 6.4, few
examples are provided here of the functions identified for it in the literature.
 Chapter 6.  Negative polarity question tags (neg-tags) 173

do not apply to this tag and it is non-conducive as a result. Only a TQ such as that
in (3), which meets the sincerity conditions of both interrogatives (see above) and
declaratives (“The speaker believes that the proposition is true,” Hudson 1975: 24),
solicits a response from interlocutors.
(2) Jacky: We’re going to Southsea on the seventeenth of next month. And
on Sunday they …
Cathy: Yeah, and I can’t bloody go.
Jenny: Why not?
Cathy: Cos I’m going on fucking holiday, in I?  (Cheshire 1981: 375)
(3) Matthew: I was all brought up to say please and thank you. Wasn’t I?
HP: mhm
Gerald: Ah, you’re ok there.

Andersen (2001: 118–138) demonstrates that in COLT the tag variant innit per-
forms a wide range of functions identical or similar to those described above.
However, he analyses the functionality of innit within a relevance-theoretic frame-
work which postulates that utterances are contextualised on the basis of existing
beliefs so as to produce the greatest cognitive effect for minimal processing effort
(see further Blakemore 1988; Sperber & Wilson 1995). Consequently, he describes
the functions performed by innit not in terms of politeness or conduciveness but
in terms of contextual alignment. Depending on its prosodic encoding, innit func-
tions either to establish or to signal speakers’ presumptions of common ground
with their interlocutors.
The overview provided above is necessarily brief and does not do justice to the
volume of literature investigating the functionality of TQs. Yet it successfully dem-
onstrates that scholars’ categorisations of tag functions largely overlap. The main
differences pertain to the labels attached to individual functions and the number
of functions identified. These differences are due at least in part to the differen-
tial methodological and theoretical frameworks employed and to the contrasting
research goals pursued across studies. They should not distract from the fact that
interrogative tags are consistently reported to perform important interpersonal
functions in discourse. Yet because existing accounts of their functionality are
largely based on the analysis of casual conversations, they may not capture tags’
functional spectrum in other speech events or genres. The analysis in Section 6.4
explores this possibility and establishes the functions performed by interrogative
tags in interview data.
174 The Structure of Discourse-Pragmatic Variation

6.2.2 Quantitative studies: Variation and change in the use of question tags

Contrary to other discourse-pragmatic variables, including those analysed in

Chapters 4 and 5, question tags have a long history in quantitative variationist
research. Scholars have investigated their variable frequency, function and/or
form across a range of varieties in attempts to uncover the contextual constraints
on their use. The following review of relevant studies provides important hy-
potheses about the conditioning of tag variation in the current data and suggests
that an analysis which tests the combined effect of multiple contextual factors on
variant choice may be urgently needed to further current understanding of TQ
variation and change.
The initial impetus for quantifying the use of TQs was provided by Lakoff ’s
(1973) account mentioned in Section 6.2.1 above, in particular by her intuitive
observation that women’s frequent tag usage reflects their lack of assertiveness
and their desire for confirmation and approval. Empirically based studies of TQ
usage challenged this claim by demonstrating that tags cannot be straightforwardly
associated with women’s language (Cameron et al. 1988; Dubois & Crouch 1975),
and that men are in fact more likely to use tags for signalling reduced epistemicity
than women who tend to use tags primarily to signal solidarity with co-conver-
sationalists (Holmes 1984a, 1995). The social indexicality of canonical tags has
been repeatedly confirmed over the last few decades, most recently in Moore &
Podesva’s (2009) ethnographic study of female adolescents in north-west England.
Their in-depth analysis of 778 tags produced by members of four communities
of practice demonstrates that tags derive their social meaning not just from their
frequency of occurrence but first and foremost from a combination of their dis-
course properties (whether they solicit agreement in turn-medial position), their
formal properties (whether the plosive in the negative clitic is released, glottalised
or deleted), and their grammatical properties (whether the tag is produced with
standard or non-standard morpho-syntax). Moore & Podesva (2009) argue that
it is members’ divergence across these three properties which enables them to
construct distinct group styles.
The social meaning potential of tags’ formal and grammatical variability has
also been explored elsewhere, notably in the context of neg-tag variants that de-
rive through morpho-phonological changes from the canonical tag isn’t it: ain’t it,
in’t it and innit (see further Section 6.3.2). Krug’s (1998) analysis of these variants
in the spoken subpart of the BNC, a ten million word corpus of spoken British
English collected in 1991–1994, reveals that their proportional frequency varies
dramatically across social groups and in apparent time. In stark contrast to isn’t
it, the canonically-derived variants ain’t it, in’t it and innit occur most frequently
among speakers from the lower socio-economic classes and, in the case of innit,
 Chapter 6.  Negative polarity question tags (neg-tags) 175

among speakers with shorter periods in full-time education. In addition to a pro-

pensity for innit to occur in female speech, Krug also notes an apparent-time
increase in the rate of innit and a concomitant decrease in the rate of isn’t it. He
argues that innit is gradually replacing isn’t it as well as ain’t it and in’t it which are
already marginal. The association of innit with working-class, female and young
speakers is confirmed in Andersen’s (2001) in-depth analysis of the distribution of
323 tokens of the tag variant innit in COLT (recorded in 1993) and the distribu-
tion of 96 tokens of innit in a specifically designed subset of the spoken part of the
BNC/London. Importantly, Andersen’s (2001) analysis reveals that it is not just
the frequency of occurrence of innit that is socially indexical but also its syntactic-
semantic distribution.
Through analogical levelling the canonical tag isn’t it and its derivational co-
variants ain’t it, in’t it and innit have begun to spread across the inflectional para-
digm to non-third person singular neuter contexts of be (see further Andersen
2001: 98; Cheshire 1982: 62; Krug 1998).43 Evidence for this process is provided
by what Andersen (2001: 104–105) calls ‘non-paradigmatic’ uses of the above
variants: where they occur in contexts which according to Standard English tag
formation rules require an auxiliary-pronoun combination other than is + it (see
further Section 6.3.1). In Andersen’s data, non-paradigmatic uses of innit are
limited to speakers under the age of 30 as well as speakers in inner- and outer-
London, and are more frequent amongst speakers from lower socio-economic
backgrounds and female speakers. The most marked differences in the use of
non-paradigmatic innit, though, are found across ethnic groups. Andersen dem-
onstrates that ethnic minority speakers in London have markedly higher rates of
non-paradigmatic innit than white speakers. Based on these results and previous
research which showed that tags not matching the syntactic-semantic properties
of their anchor tend to thrive in multilingual environments (see inter alia Hewitt
1986; Todd & Hancock 1986; Trudgill & Hannah 1982), Andersen (2001: 112–114,
190–192) argues that the social forces driving the analogical levelling of innit are
ethnic minority speakers. Elsewhere, the spread of isn’t it, ain’t it, in’t it and innit
across the inflectional paradigm has been attributed to their frequency in dis-
course (Cheshire 1982: 62; Krug 1998). In historical and contemporary data, third
person singular neuter tokens with present tense be are proportionally the most

43. Andersen (2001: 98) uses the term ‘invariabilisation’ to refer to “the process of reanalysis
by which a form which was originally restricted to a particular syntactic environment comes
to be used in all syntactic environments across the inflectional paradigm.” This terminology is
not adopted here as it is essentially co-referential with the more widespread and familiar term
‘analogical levelling.’
176 The Structure of Discourse-Pragmatic Variation

­dominant auxiliary-pronoun combination in neg-tags (Hoffmann 2006: 42–43;

Krug 1998:152–153; Tottie & Hoffmann 2006: 296).
Yet the fact that innit as well as isn’t it, ain’t it and in’t it are levelling across the
paradigm does not entail that their syntactic-semantic distribution is random.
In keeping with Hopper’s (1991: 28–30) principle of persistence, which predicts
that the distribution of grammaticalising features is constrained by their origin,
these tags have the highest rate of occurrence in those contexts that are syn-
tactically and semantically closest to their original environment: after anchors
with it/that and is. In COLT, innit is favoured in contexts that require a tag with
negative polarity, the pronoun it, be-verbs, present tense verbs, and you-subjects
(Andersen 2001: 162–179). In’t-tags are preferred in contexts that require a tag
with be in Scottish varieties (Macafee 1992), and contexts that require a tag
with either be or have in Cheshire’s (1981) Reading data. These results demon-
strate that although the distribution of tags consisting of or derived from is +
-n’t + it is not limited to what is assumed to be their original context of use, it
is nevertheless systematically constrained by the syntactic-semantic properties
of the preceding anchor clause. Also, the observation that these variants are
most frequent in those environments that bear the closest resemblance to their
original environment suggests that their spread across the paradigm is gradual
and ongoing (see Krug [1998: 171–172] for predictions about the variants’ future
syntactic-semantic trajectories of change).
In Andersen’s (2001) teenage data from London, innit is used by all socio-
economic groups. In Cheshire et al.’s (2005: 155–159) teenage data from Reading,
Milton Keynes and Hull, by contrast, it is used only by working-class speakers.
This suggests that the precise social meaning of innit might differ across varieties.
A possible explanation for this is the variant’s longer history and more advanced
status of grammaticalisation in London compared to elsewhere (Cheshire et al.
2005: 156). This hypothesis is supported by the following observations: that innit
occurs with a higher frequency in southern than northern UK varieties (Krug
1998); that paradigmatic and especially non-paradigmatic innit rarely occur in
Hertfordshire, Reading, Hull and Milton Keynes (Andersen 2001; Cheshire et al.
2005: 155–159). The fact that all but one token of innit in the highest social classes
and in Hertfordshire were produced by male speakers leads Andersen (2001) to
argue that although females have a consistently higher rate of innit usage than
males, it is males and not females who lead its social and geographical diffusion.
Beyond social and linguistic-structural factors, discourse-functional fac-
tors may also be implicated in question tag variability. Numerous studies have
shown that social groups differ in the strategic use to which they put tags (see
the references in Section 6.1 above). Of importance to the present investigation
 Chapter 6.  Negative polarity question tags (neg-tags) 177

is Cheshire’s (1981) study exploring the variation between [eɪnt]- and [ɪnt]-tags
among working-class adolescents in Reading. Cheshire distinguishes between
tags that function to seek confirmation from addressees and tags that convey
overtones of aggression, assertion and hostility towards addressees (see Section
6.2.1). Her quantitative analysis reveals that the occurrence of [eɪnt]-tags is un-
constrained by function, while the occurrence of [ɪnt]-tags is limited to signalling
hostility and aggression. Because of its strong association with hostility and ag-
gression, the tag form [ɪnt] is interpreted by Cheshire (1981) as a marker of the
adolescents’ vernacular culture which is dominated by violence. Beyond signal-
ling functional meanings, the variant thus also communicates social meanings.
Elsewhere, Cheshire (1982: 61) reports that her Reading corpus (collected in the
late 1970s) contains a few examples of the tag form that she represents as in it
(She’s too good for you, in it? She makes her laugh, in it?). Cheshire implies that,
similar to the in’t-tags discussed above, these tags are non-conducive and serve
to signal hostility and aggression. We may tentatively deduce from this that it is
both in’t-tags and in it that are functionally differentiated from other tag forms
in Cheshire’s data. Cheshire’s association of in it with hostility and aggression
corroborates Krug’s (1998: 151–152) suggestion that disyllabic tags (e.g. innit) are
favoured for peremptory and aggressive tag uses.
The results outlined above demonstrate the effect of language-external, linguis-
tic-structural and discourse-functional constraints on the frequency and formal
variability of tag usage in contemporary British English. Beyond furnishing impor-
tant hypotheses for the present analysis into formal neg-tag variation in BwE, they
call into question the validity of the results produced by Torgersen et al.’s (2011)
recent analysis of innit in sub-corpora of COLT (recorded in inner- and outer-
city London and Hertfordshire in 1993) and LIC (Linguistics Innovator Corpus,
recorded in inner- and outer-city London in the period 2005–2006). Their cross-
corpora comparison of the normalised frequency of innit shows that innit occurs
with similar frequencies in both corpora (1,676 per million words in COLT; 1,569
per million words in LIC) and across most social groups. This leads Torgersen et
al. (2011: 107) to conclude that innit is an established and stable feature of London
teenage English and that “the extent of use of innit is no longer characteristic of a
particular group of speakers.” Because this analysis ignores any language-internal
constraints on the occurrence of innit and any potential changes to these con-
straints in the period 1993–2006, its results may yield a potentially misleading
measure of language variation and change in contemporary London English. In
contrast to Torgersen et al.’s (2011) investigation, this study includes external and
internal predictors in the quantitative analysis of neg-tags in BwE. It thus affords
valid insights into the organisation and robustness of neg-tag variation in BwE.
178 The Structure of Discourse-Pragmatic Variation

6.2.3 Aims and contribution of this chapter

As pointed out at the beginning, this chapter investigates formal variation patterns
in the use of neg-tags in the BwE corpus. It represents a development beyond
previous quantitative and qualitative studies of (negative polarity) TQs by: inves-
tigating their use in a peripheral variety of English; testing the combined effect of
internal as well as external factors on variant selection; including in the envelope
of variation the entire system of neg-tag variants; and investigating the functions
performed by neg-tags in interview data.
Investigation of neg-tag variation in BwE, a variety of British English re-
mote from the alleged source of innovating tag uses in London and distinct from
other varieties previously subjected to tag variation analysis, allows us to test two
hypotheses advanced above: that innit might be less grammaticalised in north-
ern than southern varieties of British English; and that some tag variants might
have different social meanings across different varieties. Further, by conducting a
multivariate analysis of neg-tags, the present study overcomes the limitations of
previous studies of question tag variation which have plotted the distribution of tag
variants across multiple external and/or internal factors without testing the relative
effect of multiple predictors when they are considered simultaneously (Andersen
2001; Krug 1998). This study thus enables us to provide a more reliable and precise
description of the sociolinguistic conditioning of tag variation than previous stud-
ies. Moreover, inclusion in the variable context of all neg-tag variants found in
the data makes it possible to determine how many and which variants are affected
by analogical levelling and to assess how the grammaticalisation of some variants
affects the frequency and sociolinguistic distribution of other variants. Finally,
thorough functional analysis of neg-tags in the current corpus of interview data
allows us to establish the functional inventory of neg-tags in this speech event
and assess the comprehensiveness of functional taxonomies based on the analysis
of conversational data. In sum, the analysis presented in this chapter extends and
develops previous qualitative and quantitative studies of (negative polarity) TQs in
order to advance current insights into their usage, distribution and development.
The next section identifies the contexts in which tag variants may alternate,
introduces the variants isolated for neg-tags in the current dataset, and explains
the operationalisation of hypotheses about dependent variables that might condi-
tion the variation.
 Chapter 6.  Negative polarity question tags (neg-tags) 179

6.3 The variable context and data coding

6.3.1 The variable and the envelope of variation

The variable under investigation in this chapter is the BwE neg-tag system, i.e.,
the canon of reduced negative polarity interrogatives which are appended to a
clausal or non-clausal anchor and are derived from the following linear string of
components: (auxiliary) + (negative clitic) + (pronominal subject) or (auxiliary) +
(pronominal subject) + (isolate negator) (see Chapter 2.3.2 and Section 6.3.2). To
yield principled and accountable results, every instance of a neg-tag was extracted
from the corpus. Because the focus of this project is on discourse-pragmatic varia-
bles with negative polarity, positive polarity tags, as in (4), were not included in the
analysis. Invariant interrogative tags, as in (5), and invariant lexical tags, as in (6),
were not included in the variable context either. These tags are diachronically and
structurally unrelated to canonical or canonically-derived neg-tags. Therefore,
their inclusion in the variable context would not improve our understanding of
the factors triggering the grammaticalisation of neg-tags as I have defined them
here. Omission from the analysis of the tags exemplified in (4)–(6) may affect the
comprehensiveness of the qualitative analysis in Section 6.4 as well as intra- and
inter-speaker frequency rates of TQ usage. However, their omission does not affect
the conditioning of formal neg-tag variation.44
(4) You have heard that, have you?
It doesn’t sound very good, does it?
(5) Why can’t they do something for them and then it would be so much better,
do you know what I mean?
But I also think if Berwick was in the English league, they wouldn’t be as good,
you know what I mean?
(6) Would just be ugly if you’re unattractive, e?
I’ve said it for like five years, right?

The extraction process described above yielded instances of neg-tags occur-

ring in different syntactic-semantic contexts previously described by Andersen
(2001: 104–105) as paradigmatic and non-paradigmatic. Paradigmatic neg-tags,
as exemplified in (7), match the syntactic-semantic properties of their anchor

44. For a comprehensive overview of positive polarity TQs, see Kimps (2007). For a discussion
of invariant interrogative tags, see Torgersen et al. (2011). For a detailed account of invariant
lexical tags, see Stenström et al. (2002: 165–191).
180 The Structure of Discourse-Pragmatic Variation

and thus follow the dependency rules of Standard English question tag formation
outlined in Section 6.3.2 below. Non-paradigmatic neg-tags, as exemplified in
(8), do not match the syntactic-semantic properties of their anchor and violate
the dependency rules of Standard English question tag formation.45 Moreover,
the data contain tokens of tag variants that are only partially paradigmatic. As
shown in (9), these tags replicate the subject pronoun of their anchor but do not
mirror the person and/or number of the auxiliary in the anchor, (9a), or the type
of auxiliary used in the anchor, (9b) (in’t/in derive from the third person singular
form isn’t, see further Section 6.3.2 below). I will therefore categorise these tokens
as occurring in semi-paradigmatic contexts. Finally, the present dataset contains
neg-tags that occur with non-clausal anchors, i.e., anchors that do not contain
an overt subject and finite verb, as in (10). However, the missing subject and verb
of these neg-tags can be straightforwardly inferred from the context (e.g. It’s an
imaginary line, isn’t it? – It’s just the way they talk, innit? – I think it was last year,
wasn’t it?). Therefore, they can be treated like tags following clausal anchors and
be categorised according to paradigmaticity (see above). All instances of neg-
tags – paradigmatic, non-paradigmatic, semi-paradigmatic and those following a
non-clausal anchor – are included in the variable context. This decision is dictated
by the principle of accountability (see Chapter 2.3.2) and is further motivated by
one of the aims of this chapter which is to assess the degree of analogical levelling
across the BwE neg-tag system.
(7) It’s a different world when you go along there, isn’t it?
I suppose it’s just the same, in’t it?
It’s easier to get to and get parked, innit?
(8) Newcastle and that, they call you a Jock, isn’t it, Paula?
Oh, I’ve answered this one before last time, innit?

45. In analogy to Holmes’s (1982) distinction between canonical tags (e.g. don’t you, can’t we,
hasn’t he) and invariant tags (e.g. eh, right), Andersen (2001: 98, 104) applies the label ‘invariant’
to tag variants such as innit which frequently occur not just in paradigmatic but also in non-
paradigmatic contexts. This terminology is not adopted here. In contrast to invariant lexical tags
such as eh, right etc. and invariant interrogative tags such as (do) you know what I mean, the use
of innovating tags such as innit is not entirely unconstrained by linguistic-structural factors.
The diffusion of innit from its original context of use after anchors with third person singular
subjects and present tense forms of be is gradual and ongoing (Andersen 2001: 173–183; see also
Section 6.5 below). I therefore reserve the label ‘invariant’ to refer to lexical or interrogative tags
like those listed in (5) and (6) above, which – to the best of my knowledge – are not diffusing
from an original context of use. I will refer to tokens such as those in (8) as non-paradigmatic
instances of canonical or canonically-derived tag variants.
 Chapter 6.  Negative polarity question tags (neg-tags) 181

(9) a. You’re quite near Hungary then, in you?

Aye, we’re Northumberland in’t we?
They’re wise, in’t they?
b. They’ve got less pool to draw their players, in’t they?
We got a little visitor centre, in’t we?
(10) An imaginary line really, isn’t it?
Just the way they talk and that, innit?
I think last year, wasn’t it?

Initial data observations further reveal that all neg-tags in the data occur after
declaratives. The vast majority of these follow an affirmative anchor clause, i.e.,
they form part of a reverse negative polarity TQ. Only one neg-tag in the data
occurs after a negative polarity anchor, thus forming part of a constant negative
polarity TQ, as illustrated in (11). Evelyn’s proposition the young ones don’t want
to go echoes HP’s preceding utterance regarding young Berwickers’ reluctance to
leave Berwick. The appended tag do they no? serves to signal Evelyn’s surprise at the
information offered by HP. This function, which has in the past also been identified
for constant positive polarity tags (Kimps 2007: 282–283), is not performed by any
of the reverse polarity neg-tags in the data (see Section 6.4 below). In view of its
rarity and functional anomalousness compared to reverse polarity neg-tags, the
token in (11) is not included in the envelope of variation.46
(11) HP: I’ve interviewed a lot of young people who said, “No. Maybe
I have to go away for a job, [but I don’t] [really want to.”]
Mary: [mhm]
Evelyn: [Th-the] young ones
don’t want to go, do they no?

As in the preceding chapters, the following types of tokens were also excluded
from the database: tokens of the variable whose form could not be determined;
and tokens which occurred in quoted speech and thus could not be confidently
assigned to the quoter’s linguistic repertoire. In preparation for the quantitative
analysis in Section 6.5, each of the 316 tokens of neg-tags retained in the database
was coded for its form and a number of predictors hypothesised to constrain neg-
tag variation. The next sections set out the coding procedures.

46. The fact that only one neg-tag in the data follows a negative polarity anchor clause confirms
the rarity of this construction noted elsewhere (Algeo 1988: 178; Hoffmann 2006: 35, 43; Kimps
2007: 271; McGregor 1995: 99; Quirk et al. 1985: 813; Tottie & Hoffmann 2006: 284, 289).
182 The Structure of Discourse-Pragmatic Variation

6.3.2 The dependent variable: neg-tag variants in BwE

The extracts from the data in (1) in Section 6.1 illustrate that BwE neg-tags ex-
hibit rich variation in their form. This is partly a result of the complex rules of tag
formation in Standard English which affect the type, tense, number and person of
the auxiliary as well as the person, number and gender of the pronoun in the tag.
Yet it is also caused by the variable attrition, fusion and overall phonetic realisation
of the construction’s component parts. Moreover, in BwE neg-tags it is not just
the placement but also the form of the negative particle that is variable. To ensure
consistent and accurate coding of what are often very similar realisations (e.g.
didn’t he vs. din he, isn’t it vs. in’t it), all neg-tag tokens were subjected to repeated
auditory analysis, and the coding of a random sub-sample of tokens was verified
by an independent coder. This section introduces the BwE neg-tag variants which
have been identified through this procedure, and divides them according to their
morpho-phonological encoding and geographical dispersion.
As pointed out in Section 6.1 above, canonical tags are grammatically depend-
ent on their preceding anchor: the auxiliary in the tag is the same as the auxiliary
in the anchor clause (12a); when the anchor contains no auxiliary, the tag is formed
with do-support, mirroring the tense, number and person of the finite verb in the
anchor (12b); the pronoun in the tag repeats that from the anchor (12c), or agrees
with the preceding nominal subject in number, person and gender (12d); the polar-
ity of the anchor is usually reversed in the tag (12e). In contrast to the high degree
of auxiliary-, pronoun- and polarity-variation, the tag-internal order of elements
is fixed: (auxiliary) + (enclitic negator) + (personal pronoun). (As discussed further
in Section 6.5.1, the BwE data contain no neg-tags with not which, like no in (14)
below, has a different placement than -n’t.) Because they are formed with the range
of English auxiliaries and personal pronouns, the neg-tag variants exemplified in
(12) do not strictly speaking constitute a uniform formal category. Nevertheless,
they are allocated here to one variant category and differentiated from other vari-
ants on the basis of their relative lack of formal reduction, their adherence to
Standard English rules of tag formation, and their use of non-localised negators
and negative auxiliaries. By virtue of sharing these characteristics, the canonical
tags exemplified in (12) are found across England and the English-speaking world
and as such are non-localisable.
(12) a. It’s nice and quiet, isn’t it?
They’re lovely flyers, aren’t they?
They will go on, won’t they?
You can pick them up, can’t you?
 Chapter 6.  Negative polarity question tags (neg-tags) 183

b. We even used the contacts from the post-office, didn’t we?

They think it is part of Scotland, don’t they?
But war brings out the worst in people, doesn’t it?
c. We would say gear, wouldn’t we?
They’re like a home nation, aren’t they?
d. Lydia was born here, wasn’t she?
The young ones have too much now, haven’t they?
e. It was a knacker doing them herrings, wasn’t it?
We divn’t go under the border community, do we?

Rare instances of neg-tags which violate the rules of subject-verb agreement, as

in (13), are also included in the category of non-localised canonical tag variants
introduced above. These tags adhere to the rules of Standard English tag formation
summarised above by mirroring the properties of the auxiliary (and subject) in the
preceding anchor, and they reflect a tendency towards was-levelling which occurs
throughout the English-speaking world (Adger & Smith 2005: 155).
(13) We was doing that, wasn’t we?

In addition to the variants introduced above, the BwE corpus contains canonical
tags that follow the dependency rules of the non-localisable canonical tags illus-
trated in (12) but deviate from these in that they contain the negative auxiliary
variant divn’t (14), or are formed with the independent negator no (15). Rather
than occurring between the auxiliary and the pronoun, no is positioned after
the pronoun. Because of the strong association of divn’t with the north(-east) of
England and of no with Scotland (see Chapter 3.3.2), these variants are subsumed
here under the label localised canonical tags.
(14) I mean, you get drugs everywhere, divn’t you?
They like the words, divn’t they?
(15) I think kecks really is underpants, is it no?
But we called them trousers, did we no?

Finally, we find in the BwE corpus variants that derive from (non-localised) ca-
nonical tag variants through phonetic attrition. The variant din [dɪn] + (pronoun)
in (16) is a reduced form of either don’t (fronting of the vowel and loss of [t]) or
of divn’t (loss of [v] and [t]). The variants in’t [ɪnt] or [ɪnʔ] + (pronoun) and in
[ɪn] + (pronoun) in (17) and (18) derive through regular sound change from isn’t it
(loss of [z] and [t]) (Andersen 2001: 196–200; Cheshire 1981: 366–367). Following
Cheshire (1981), int/in-variants are conflated in the remainder of this chapter to
constitute one variant of in’t-tags.
184 The Structure of Discourse-Pragmatic Variation

(16) He speaks completely different to me, din he?

(17) Well, it’s proper cooking, in’t it?
They’re wise, in’t they?
(18) Oh yeah, there’s a whole heap of them, in there?

Another variant that derives through phonetic reduction and, in this case, struc-
tural reanalysis from non-localised canonical tags is innit in (19). The develop-
mental pathway of innit, pronounced [ɪnɪt] or [ɪnɪʔ], is somewhat less certain
than that of the variants in (16)–(18). Andersen (2001: 196–201) puts forward
two alternative hypotheses which are sketched out in (20) and (21). The cline in
(20) suggests that innit derives through phonetic attrition from isn’t it, with loss
of [t] preceding loss of [z]; the two formerly independent morphemes, [ɪn] and
[ɪt], are subsequently reanalysed (or rebracketed) as a single unit, [ɪnɪt]. The cline
in (21) suggests that innit derives through phonetic attrition from ain’t it (raising
of the initial vowel of the diphthong > monophthongisation > loss of final [t]).
Andersen’s (2001) data support the pathway in (20). He argues that if innit derived
from ain’t it, which in turn derived by regular sound change from negative present
tense forms of be and have (Cheshire 1981: 366), it would be used for hasn’t it in
the initial stages of its spread. However, this is not borne out in his data, where
innit first spreads in third person singular neuter contexts of be before spreading
to other contexts, including hasn’t it. Also, he argues, the forms in’t and in occur
too infrequently in his data to be considered stepping stones on the path to innit.
(19) Well, it’s only an hour away from Edinburgh and Newcastle, innit?
(20) isn’t it [ɪznt ɪt] > isn’t it [ɪzn ɪt] > in it [ɪn ɪt]  (Andersen 2001: 197)
(21) ain’t it [eɪnt ɪt] > int it [ɪnt ɪt] > in it [ɪn ɪt] (Andersen 2001: 197)

The variants introduced in (16)–(18) are found across many varieties of British
English. Din has been attested in Cambridgeshire (Peitsara & Vatso 2002) and
northern England (Llamas 2001: 127, fn. 8: Moore & Podesva 2009). In’t-tags have
been reported for southern England (Andersen 2001; Cheshire 1981), northern
England (Cheshire et al. 2005: 157; Moore & Podesva 2009), and south-eastern
Scotland (Brown & Millar 1980: 118). Their widespread, albeit numerically vari-
able, use across Britain is also confirmed in Krug (1998: 193–194). The use of innit
in (19) is often associated with London and the south of England (Krug 1998: 193–
195; Stenström et al. 2002: 168), but it has also been reported for varieties beyond
London and the England-Scotland border (Cheshire 1982: 61; Cheshire et al.
2005: 155–158; Llamas 2001: 127, fn. 8; Moore & Podesva 2009). Because they are
used in many varieties of English throughout the British Isles but are not gener-
ally associated with Standard English usage, the canonically-derived tag variants
illustrated in (16)–(19) are subsumed here under the label supra-local variants.
 Chapter 6.  Negative polarity question tags (neg-tags) 185

Table 6.1 summarises the BwE inventory of neg-tag variants. It allocates vari-
ants to one of three categories which are differentiated on the basis of variants’
morpho-phonological properties and geographical spread: non-localised canoni-
cal tags, localised canonical tags, and supra-local canonically-derived tags. Where
appropriate, Sections 6.5 and 6.6 will make more fine-grained distinctions within
these categories in order to draw attention to and account for the details of variants’
composition and realisation.

Table 6.1  Inventory of neg-tag variants in BwE

non-localised localised supra-local
canonical tags canonical tags canonically-derived tags
auxiliary + -n’t + pronoun auxiliary + pronoun + no din + pronoun
divn’t + pronoun in’t + pronoun
innit

6.3.3 Independent variables: Data coding

Following the coding of each neg-tag token for its form (both in terms of vari-
ant category and constituent elements), the next step in the analysis was to create
a coding system for the independent variables believed to be implicated in the
conditioning of formal neg-tag variation in BwE.
The research on TQs reviewed in Section 6.2 has demonstrated that the
­morpho-phonological encoding of tag variants is strongly constrained by a range
of social factors. It has also shown that the spread of innit across the inflectional
paradigm is socially embedded. To test the hypothesis that social factors are also
implicated in the formal variability of neg-tags in BwE and to probe the social
indexicality of the variants in the data, every token of the variable was coded for
speaker sex and age. If the neg-tag system in BwE is undergoing grammaticalisa-
tion, as is the case in London English (see Section 6.2.2), inclusion of speaker sex
and age as independent predictors also allows us to identify the innovators and
leaders of any such changes, and assess whether innovating variants such as innit
have identical social meanings across varieties. In order to avoid producing results
that are distorted by high degrees of intra-group variation, all tokens were further
coded for individual speaker. This makes possible application of mixed-effects
modelling to accurately identify the rules governing variant selection.
The literature on TQs also highlights that the syntactic-semantic characteris-
tics of the anchor clause have a bearing on the form of the appended tag, even when
tag variants are levelling to contexts where their occurrence is not sanctioned by
the rules of Standard English tag formation. To establish whether BwE neg-tags
186 The Structure of Discourse-Pragmatic Variation

are at all affected by levelling processes and whether these are gradual in nature,
all tokens of the variable in the data were coded firstly for their syntactic-semantic
context, i.e., paradigmatic, non-paradigmatic and semi-paradigmatic, as defined
and illustrated in Section 6.3.1 and Extracts (7)–(10) above; and secondly for the
syntactic-semantic properties of their anchors, i.e., person, number and gender
of the subject as well as type, tense, person and number of the finite verb. Beyond
establishing whether BwE neg-tags are undergoing grammaticalisation, this may
allow us to identify any trajectories of tags’ spread across the inflectional paradigm.
Moreover, if multiple neg-tag variants are undergoing levelling processes simul-
taneoulsy, consistent coding for the semantic-syntactic properties of the anchor of
all neg-tag tokens enables us to assess whether different variants are competing
for the same syntactic-semantic contexts.
Finally, the results reported in Cheshire (1981) for ain’t- and in’t-tags demon-
strate that variation in the formal realisation of neg-tags may signal important
functional differences in their use. To test whether function is implicated in the
formal variability of neg-tags in BwE, Section 6.4 develops a functional taxonomy
for BwE neg-tags. This taxonomy serves as a coding schema for the factor group
function and makes it possible to test the effect of function on variant choice in
Section 6.5.

6.4 Qualitative analysis of neg-tags

The functions performed by neg-tags in the BwE interview data are uncovered
through development of the functional taxonomies available in the literature and
application of the methods outlined in Chapter 2.5. Crucial to this endeavour is
the close reading of tags in their larger sequential and interactional contexts of oc-
currence, especially with regard to listeners’ next-turn responses. They constitute
a guide towards the interpretation of tags on-line in interaction and often suggest
a particular reading of the data (see Chapter 2.5.2). Furthermore, meticulous at-
tention is paid in utterance interpretation to the ambient linguistic material in
the anchor and the tag. This material is faithfully reproduced in the data extracts
provided below (see page xvii for a key to the orthographic conventions used
for replicating non-verbal materials). The extracts serve to illustrate the range
of functions isolated in the data and to demonstrate the validity of the analysis.
The ultimate aim of the analysis is to operationalise function as a factor group for
quantitative analysis.
 Chapter 6.  Negative polarity question tags (neg-tags) 187

6.4.1 Functions performed by neg-tags

Although I embrace the general view that interrogative tags perform important
interpersonal functions in social interaction, I depart from previous functional
accounts which categorise tags in terms of modal vs. affective meanings (Holmes
1984a, 1987, 1995), polite vs. impolite functions (Algeo 1988, 1990), or conven-
tional vs. non-conventional uses (Cheshire 1981). Instead, I broadly divide the
interpersonal uses of neg-tags into a subjective, speaker-oriented and an inter-
subjective, hearer-oriented category (see Chapter 2.5), recognising at all times that
there is some overlap between the two.
The literature reviewed in Section 6.2.1 concurs in attributing an epistemic
function to interrogative tags. The tag serves to reduce speakers’ commitment to
their propositions and to seek verification of these propositions from addressees.
BwE neg-tags regularly perform this function, as exemplified in (22) below. The
tag can’t you follows Charlene’s proposition that she can be identified as being from
Berwick despite not using many Berwick slang words. The opinion preceding the
tag is expressed with confidence and assertiveness. Neither the anchor clause to
which the tag is appended nor the preceding utterance contain any linguistic fea-
tures that may be indicative of doubt or hesitancy, such as false starts, repetitions,
epistemic markers, (un)filled pauses, high pitch, or rising intonation contours.
The tag can’t you, by contrast, is produced in higher than usual pitch and with a
rising intonational contour. Both prosodic features convey a strong impression
of tentativeness and doubt; the latter also serves to seek hearer validation. With
this prosodic encoding, the tag serves a dual function. It retrospectively qualifies
Charlene’s commitment to her proposition, while at the same time seeking verifica-
tion of this proposition from her co-interviewee.
(22) Charlene: I’ve lived in Berwick all my life but I don’t use all the slang
words. But you you can still tell I come from Berwick. Can’t you?

In addition to attenuating the strength of assertive propositions, neg-tags also

function to reinforce the tentativeness already inherent in the turn elements over
which they have scope. In (23a), Jane’s halting deliberation about the denotations
of kecks, a dialect word for ‘trousers’ or ‘underpants,’ is replete with linguistic fea-
tures that signal reduced epistemic commitment: filled and unfilled pauses, the
hedges I divn’t knaa and I think, and the final rise on underpants. In (23b), high
pitch signals which constituent of the elliptical anchor clause is the focal element
of uncertainty: the month of the Riding of the Bounds, a long-standing annual
ceremony which involves residents patrolling the boundaries of Berwick on horse-
back. In both examples, the neg-tag invites interlocutors to confirm or refute the
188 The Structure of Discourse-Pragmatic Variation

assessments or facts presented as uncertain in the preceding discourse. As a result

of their conduciveness, the tags also control turn-allocation.
(23) a. HP: Now, what’s kecks. Is kecks [trousers or] underpants?
Jane: [Trousers.]
No, trousers. (.) Eh now, I divn’t knaa? (..) I think kecks really
is underpants, is it no?
b. Evelyn: Have you heard about that story, the Riding [of the Bounds?]
Mary: [Riding of the Bounds.]
HP: [No.]
What’s that?
Evelyn: Well, (.) the first (.) of May? Isn’t it, Mary.
Mary: == It’s the first of May.

As pointed out in the literature review, tags do not by definition signal reduced
epistemicity or invoke a response from addressees. The extracts in (24) and (25)
illustrate the interactional impact of non-tentative and non-conducive neg-tags
in the BwE corpus. In (24), Daniel, the sole interviewee, tells HP about an occa-
sion on which he spoke to some Scottish people who had emigrated to the United
States twenty years before his encounter with them. When Daniel reveals that the
Scottish expatriates were pleased with his observation that they had retained their
native accent, he follows this assessment with innit. HP was not present at the
encounter that Daniel recounts and hence is not in a position to either confirm or
refute Daniel’s version of the events. The sincerity conditions on interrogatives are
not met (see Section 6.2.1). Thus, rather than seeking confirmation and involve-
ment, the tag serves to foreground Daniel’s utterance and to reinforce a point he
had been stressing throughout the interview: that Scottish people take great pride
in their nationality. HP’s acknowledgement tokens (mhm, yeah, uh-huh), which
are produced in overlap with and in succession to Daniel’s innit, do not preclude a
non-conducive reading of innit. They do not constitute confirmatory responses but
signal HP’s continued interest in Daniel’s narrative and encourage him to continue
talking (Jefferson 2002; Schegloff 1982).
(24) Daniel: And they was still broa- I could tell, still tell they was Scots
HP: mhm
Daniel: = And I say, “Oh. You haven’t (.) you haven’t lost your accent.”
Oh, they was pleased as punch, [innit]   [They] always say,
HP: [mhm] yeah [uh-huh]
Daniel: = “We’ll no do that.” Because they live amongst Scots.
 Chapter 6.  Negative polarity question tags (neg-tags) 189

In (25), Keith presents a proposition that is generally accepted to be true: home’s

always home. The appended tag innit is non-conducive and functions to underline
the obviousness of the preceding statement (see also Coates 1987: 117–118; Millar
& Brown 1979: 34–35). Tag uses such as this often occur turn-finally at the end of
an extended topic sequence. Because they do not invite addressees to contribute
their view on the topic, they are generally followed by topic-change.
(25) Keith: Home’s always (..) home innit.

The extracts from the data in (24) and (25), then, illustrate that neg-tags can
function to emphasise speakers’ attitudinal stance towards propositions to which
they are fully committed. Following Tottie & Hoffmann (2006: 300–301), I label
these tag uses attitudinal stance markers. They do not usually seek verification or
confirmation from addressees, and are non-conducive as a result. Unlike the non-
conducive tags described in Algeo (1988, 1990) and Cheshire (1981), however, they
do not generally carry a hostile or antagonistic overtone.
In addition to the subjective functions outlined above (epistemic marking, at-
titudinal stance marking), neg-tags also function intersubjectively. Tags function-
ing as mitigating devices exploit the tentativeness and corroboration-seeking effect
associated with tags to soften the negative force of interactionally dispreferred
moves (see also Andersen 2001: 124–128; Holmes 1982: 58–61). In (26a), Rebecca
displays a negative stance towards HP’s suggestion that incomers to Berwick might
be requested to adopt the local dialect. Her turn-final utterance cos that’s just silly,
isn’t it is produced with moderately reduced loudness. In addition, the lexical item
silly and the tag isn’t it are produced on a moderately high pitch level. With this
pitch contour, the tag functions to soften the negative effect of Rebecca’s potentially
quite face-threatening dismissal. In (26b) and (26c), Germaine and Keith dispute
the congruity and relevance of the dialect words suggested by HP and Adam. The
rising contour on in’t it in (26b) conveys that Germaine entertains some doubt re-
garding her correction. This reduces the face-threat constituted by her disaligning
turn. In (26c), the slight fall on divn’t they signals that the disagreement expressed
in Keith’s anchor is not an aggressive or hostile challenge but a matter-of-fact ob-
servation. The tag thus works towards averting interactional conflict. Finally, in
(26d), Patrick ridicules HP’s question regarding the fashionableness of Berwick
upon Tweed. He offers fictional evidence for Berwick’s trendiness so exaggerated
as to indicate its lack of trendiness. By adding the tag didn’t we he invites HP to
pretend to agree with this proposition that he does not endorse himself. This at-
tenuates the face-threat constituted by his sarcastic remark (see also Andersen
2001: 127–128). The examples discussed above demonstrate that tags used as sof-
teners and mitigation devices can be conducive when they challenge addressees
190 The Structure of Discourse-Pragmatic Variation

to justify the proposition the speaker disagrees with (26c), or non-conducive and
topic-curtailing when they signal that the co-conversationalist’s preceding proposi-
tion is in some way wrong or inappropriate (26a).
(26) a. HP: But would you say, if you want to live in Berwick, you have
to speak like us.
(.)
Rebecca: No:.
HP: No.
Alicia: Not really. [(h)]
Rebecca: [There’s] lots of different people live in Berwick
(.) [as it is (and)] and you never say, “Right. If you want to
Alicia: [Yeah.]
Rebecca: = live in Berwick you have to speak like a Berwicker.” Cos
(..) ºthat’s just silly, isn’t it.º
b. ((Discussing dialect words for standard English ‘woman’.))
HP: Have you ever heard hantle [(.) for woman?]
Germaine: [Hantle.] Em (..) is that’s like
for your heed, in’t it? ºHantle. [A hantle.]º
HP: [Is it?] Uh-huh.
c. ((Discussing dialect words for ‘attractive’.))
Adam: Eh fit.
HP: ºMhm.º
Keith: Fit’s like any- e- (.) everywhere says fit though, [divn’t they.]
Adam: [Aye.] But it’s
(.h) I was just trying to think of what I would say.
d. HP: Do you think Berwick is a fashionable place to be?
(…)
Patrick: We got a telephone <£ last week, didn’t we. £>

Tokens that I refer to as involvement inducers (called facilitative or confirmatory

tags in the literature, see Section 6.2.1) are always conducive. They draw listeners
into the discourse by seeking their support or corroboration of a proposition. The
tags in (27) are produced with a falling intonation contour that reflects speak-
ers’ confidence in the truth of their propositions. These do not therefore serve
to modify the propositions expressed in the preceding anchor, but instead they
invite addressees to express their agreement with these propositions. In (27a),
Cody’s invited contribution to the discourse amounts to no more than a minimal
affirmation which serves to acknowledge Glenn’s proposition but does not trigger
turn-exchange. Cody thus performs a supportive rather than active role. In (27b),
by contrast, Charlene’s corroboration-seeking tag isn’t it yields the turn to Natalie,
 Chapter 6.  Negative polarity question tags (neg-tags) 191

allowing her to become an active and equal participant in the discussion of the
topic. The corroboration-seeking interpretation of Charlene’s tag is supported by
the structure of Natalie’s disaligning turn: it is delayed and hedged (I’m not sure),
indicating that an affirmative response had in fact been invited. By inviting and
encouraging co-interviewees’ agreement and participation in talk, neg-tags with
a falling intonation contour serve to secure and maintain listeners’ involvement in
the interaction (hence the label) (see also Norrick 1995).
(27) a. Glenn: There’s like (..) the young people (.) I know we’re no old. (h)
But you know like the (.) thirteen- fourteen-year-olds,
HP: mhm
Glenn: = they’re (.) even though they’re from Berwick, they’re more
Geordie aren’t they. [Cos] they walk around and they
Cody: Oh yeah.
HP: [Yeah?]
Glenn: = talk Geordie, but you know, they (.) that’s no Berwick what
they talk.
b. ((Talking about different housing estates in Berwick.))
Charlene: Well, Highcliffe’s on one side and Highcliffe’s nice. But on
the other side is Eastcliffe, which is a bit, I would
HP: uh-huh
Charlene: = say that’s about the the [roughest] in the town. Isn’t it.
HP: [uh-huh]
(.)
Natalie: I’m not sure. Some parts of Highfields is quite
Charlene: Highfields
Natalie: = run down.
Charlene: = uh-huh

neg-tags are not just a means by which speakers seek to ensure common ground
with their listeners and a means of creating listeners’ involvement in discourse.
Like negation in general (Cheshire 1997, 1998), they can also be used by speakers
to signal their alignment and involvement with prior speakers’ talk. In the extract
from the data in (28), HP expresses her fondness of the expletive doylem, a Berwick
word for ‘idiot.’ Godfrey follows HP’s statement with a tag-appended clause. The
anchor in his turn (that’s a good one) echoes HP’s sentiment. The appended tag can
therefore hardly be interpreted as seeking corroboration of this attitude. Rather,
it serves to reinforce Godfrey’s alignment with HP’s turn that is already explicit in
the anchor. By signalling Godfrey’s alignment with and approval of HP’s stance,
the tag constitutes a positive politeness device.
192 The Structure of Discourse-Pragmatic Variation

(28) Godfrey: Doylem. @ [@]

HP: [<@ That’s my one of my favourite ones. @>]
Godfrey: <@ That’s a good one innit @>

It was shown earlier that neg-tags sometimes function to emphasise speakers’

stance (see Example (24) above). In certain sequential contexts, assertive and em-
phatic tags can signal speakers’ involvement with what is being said. When Lori
in (29) tells HP about an occasion where her family deliberately attracted pigeons
to her feet despite her fear of birds, Godfrey provides an unsolicited confirmation
of the account: yeah, we was doing that. Godfrey terminates his confirmation with
a non-conducive tag (wasn’t we) that functions to signal his attitude towards the
truth of Lori’s turn and to manifest his interpersonal involvement in the interaction
(see also Norrick 1995: 687). Godfrey’s tag in (29) is similar to the token in (28) in
that it signals involvement and alignment and is not response-eliciting.
(29) Lori: (h) I said I divn’t like birds. (..) And they all (shouted). You
know, we were walking along, and they threw bread and that at
my feet! <@ So that all swarmed on us. [I was petrified.] @>
HP: [@]
Godfrey: <@ Yeah. We was doing that, wasn’t we. @>
Lori: <@ Isn’t that nasty? @>

The examples provided above highlight the interactional importance of neg-tags

in discourse. By signalling speakers’ shared orientation to a topic and signalling
involvement, they constitute a positive politeness device, contribute to the collabo-
rative development of interaction, and work towards maintaining good rapport
between interlocutors.

6.4.2 Summary of functions

The functional analysis of neg-tags in the BwE interview data has shown that
they perform important interpersonal functions in discourse. Subjectively, they
function to signal speakers’ degree of commitment and attitudinal stance towards
their propositions. Intersubjectively, they serve to mitigate potential face-threats,
to draw listeners into the discourse and maintain their active involvement, and
to signal speakers’ alignment with prior talk and active involvement in the in-
teraction. The analysis thus confirms the interactional importance of tags and
negation more generally, and provides supporting evidence for the non-condu-
civeness of some tag uses (see also Andersen 1998). There is, of course, some
overlap between the functional categories established above, for tags – like other
 Chapter 6.  Negative polarity question tags (neg-tags) 193

discourse-pragmatic variables – are intrinsically multifunctional (see also Coates

1987: 130). By describing neg-tags not just in terms of the specific interpersonal
functions they perform but also in terms of their conducive or non-conducive
effects, the analysis demonstrates that individual instances of neg-tags differ suf-
ficiently to warrant allocation to distinct functional categories. A concise sum-
mary of the variable’s functional inventory and recurring co-occurrence patterns
is provided in Appendix 5.
The functional taxonomy of neg-tags developed above accounts for every
token of the variable in the current dataset. However, it must not be taken as
representing an exhaustive list of the functions performed by question tags in
everyday social interaction. Firstly, the analysis has focused exclusively on neg-
tags and has not considered the functional repertoires of positive polarity or
invariant tags. These have been shown to perform functions not described above
(Kimps 2007; Stenström et al. 2002: Chapter 7). Secondly, the analysis was based
on interview data which may also have limited the scope of the analysis. What
this analysis has demonstrated, though, is that with the exception of hostile
or aggressive tags which have not been found to occur in the BwE corpus, the
functionality of neg-tags does not diverge to any great extent across the inter-
view data analysed here and the more casual and spontaneous data previously
analysed in the literature.
Table 6.2 below outlines the coding protocol which arose from the detailed
functional analysis in Section 6.4.1. Following this protocol, each neg-tag in the
data was allocated to the functional category which best described its use. This
procedure makes it possible to test the contribution of function to variant choice.

Table 6.2  Coding schema for neg-tag functions

conducive non-conducive either
subjective epistemic marker attitudinal stance m.
intersubjective involvement inducer alignment signal mitigation device

6.5 Quantitative analysis of neg-tags

The preceding sections have served to prepare the data for quantitative analysis.
They have isolated the variants available in the BwE pool of neg-tags and opera-
tionalised hypotheses about their sociolinguistic distribution as factor groups for
quantitative analysis. This section presents the results of the quantitative analy-
sis. For ease of cross-variable comparison, its structure largely mirrors that of
194 The Structure of Discourse-Pragmatic Variation

preceding chapters. Section 6.5.1 begins with an outline of the overall distribu-
tion of variants in the corpus and an account of the highly uneven distribution
of variants within and across variant categories. There follows a short description
of how the variable and its most frequently instantiated variants are distributed
across social groups, syntactic-semantic contexts, functional categories and/or
individual speakers. These distributions affect the configuration of the mixed-
effects multiple regression analyses in Section 6.5.2 which model the simultaneous
effect of the independent variables on the observed patterns of formal variation
whilst at all times catering for the effect of intra-group variation. The multiple re-
gression analyses will reveal the divergent distribution of non-localised canonical
and supra-local canonically-derived neg-tag variants in BwE. Close inspection
of these results affords important new insights into the usage and diffusion of
innovating tag variants as well as the structure of discourse-pragmatic variation
more generally.

6.5.1 Distributional analysis

In Section 6.3.2, each of the 316 neg-tag tokens in the data was allocated to one
of the three variant categories listed in the left-hand column of Table 6.3. The
table shows the numerical breakdown of individual variants across and within
the variant categories. Non-localised canonical tag variants of the form auxiliary
+ -n’t + pronoun dominate the BwE neg-tag system. Constituting roughly two
thirds of all tokens of the variable, they dwarf the frequency of other variants,
especially those in the category of localised canonical tags where the negator takes
the form no or where negative periphrastic do takes the form divn’t. These two
variants account for less than 4% of all neg-tags in the data. At 28.3%, the sum
of tokens in the category of supra-local canonically-derived variants occupies a
more prominent position in the neg-tag system. However, variants within this
category occur with unequal frequencies. Din + pronoun occurs only once in the
whole dataset. With 17 tokens, in’t + pronoun is slightly more frequent but still
amounts to a mere 5% of all neg-tags. The most frequent supra-local canonically-
derived variant by far is innit. It constitutes almost a quarter of the data and is the
second most frequent neg-tag variant in the BwE corpus overall. The distribu-
tion in Table 6.3, then, demonstrates that despite the range of options available to
BwE speakers for realising neg-tags, a full 91% of neg-tags are constituted by
just two variants: non-localised canonical tags and the supra-local canonically-
derived variant innit.
 Chapter 6.  Negative polarity question tags (neg-tags) 195

Table 6.3  Overall distribution of neg-tag variants

variant category variant N %
non-localised canonical tags auxiliary + -n’t + pronoun 216 68
localised canonical tags auxiliary + pronoun + no    2   0.7
divn’t + pronoun    9  3
supra-local canonically-derived tags din + pronoun    1   0.3
in’t + pronoun   17  5
innit   71 23

The observed paucity of tag variants with no, divn’t, din and in’t is largely a product
of social, structural and functional constraints on their use. The negligible occur-
rence of tag variants with no (N = 2) and the complete absence of variants with not
is consistent with the widely reported pattern whereby isolate negative particles
are virtually missing from present-day English question tags (Beal & Corrigan
2005: 149; Hoffmann 2006: 46; Tagliamonte & Smith 2002: 263). The more general
demise of the negator no in BwE (see Table 3.4 in Chapter 3.4) as well as the strong
association of no-tags with the rather infrequently attested function of signalling
epistemicity (see Figure 6.3 below) may exert additional inhibiting effects on the
occurrence of no-tags. With din- and divn’t-variants (N = 1 and N = 9 respectively),
it is syntactic-semantic factors that limit their frequency in the BwE neg-tag sys-
tem. Only 34 tokens of neg-tags in the data are preceded by anchor clauses that
require tags with present tense negative do. When these constraints are considered,
divn’t-tags are not in fact all that marginal, accounting for 26% of all relevant tags.
The rate of in’t it (N = 6) is not restricted by the infrequency of anchors that require
a negative tag with is and it (N = 136) but by the high productivity of its co-variants
in this environment: isn’t it and innit (see further below). Also, the inconspicuous
occurrence of in’t-tags overall (N = 17, including those with the pronoun it above)
and among young speakers in particular parallels its marginality and social distri-
bution in other corpora (Andersen 2001: 199; Anderwald 2002: 131; Krug 1998).
It may therefore be symptomatic of the variant’s more general paucity in contem-
porary British English. The distribution of tag variants with no, divn’t, din and
in’t calls for further investigation in order to test the hypotheses proposed above.
Due to low token numbers it cannot be pursued here. The following analysis will
therefore focus on the distribution of the two most frequently attested variants
in the BwE corpus: non-localised canonical tags and the supra-local canonically-
derived tag innit.
Initial data runs reveal that these variants are not equally distributed across
the 33 speakers in the sample who tag their propositions. Firstly, while all speak-
ers use non-localised canonical tags, albeit with fluctuating frequencies, only 13
196 The Structure of Discourse-Pragmatic Variation

speakers in the sample use the supra-local canonically-derived tag variant innit. To
ensure valid and reliable results, only data from the 13 speakers who use innit in
the recordings are included in this variant’s tabulation across independent variables
(N = 145 neg-tag tokens). Secondly, the social cohorts represented in the corpus
data differ in their frequency of use of isn’t it and innit, two variants which by virtue
of being composed of or derived from the same auxiliary-pronoun combination
compete for the same syntactic-semantic environment. The grey bars along the
y-axis in Figure 6.1 reveal a marked female predominance in the use of isn’t it. They
also show that the rate of isn’t it is fairly constant across older and younger speakers
but dips in the middle age group. By contrast, the black bars show a sharp increase
in the rate of innit in the middle age group which stagnates in the young age group.
Due to badly distributed data, the marginal results for innit across speaker sex are
grossly misleading. Contrary to the patterns displayed in Figure 6.1 and those re-
ported for COLT and the BNC (Andersen 2001: 184–186; Krug 1998: 186), innit is
strongly associated with male speakers in the BwE data. Not only are eleven out of
the 13 innit-users in the data male but their normalised frequency of innit is almost
three times that of the two female innit-users’ (7.5 vs. 2.6 per 10,000 words). These
results and those depicted for isn’t it suggest that social factors are implicated in
the formal variability of the BwE neg-tag system.
The distributional analysis in Figure 6.1 is based on the quantification of the
whole set of neg-tag tokens in the data. Infrequent variants, i.e., no-, divn’t-, din-
and in’t-tags, are included in the quantification of the data in Figures 6.1 to 6.3
because they might affect the distribution of the more frequent variants. However,
because any results obtained for these variants are potentially unreliable due to
limited data (N = 29), they are not displayed in the figures summarising variant dis-
tributions. Non-localised canonical tags that do not contain the auxiliary is and the
pronoun it (see Table 6.4 below) are also consistently included in the quantification
of the data. The reason why their distribution across social groups is not displayed
in Figure 6.1 is that it would provide little insight into the social mechanisms of
formal tag variability. Rather, it would reflect social groups’ differential tendencies
to tag anchors that do not require tags with is and it, a phenomenon not pertinent
to the aims of the present analysis.
Separating out isn’t it from the other non-localised canonical tags in Figures 6.1
to 6.3 is motivated by the aims of the study as well as the results shown in Table 6.4
which examines the relative frequency of the 43 different auxiliary-pronoun com-
binations found in the BwE neg-tag system. As in other corpora (Hoffmann
2006: 42–43; Krug 1998: 152–153; Tottie & Hoffmann 2006: 296), isn’t it is pro-
portionally the most dominant auxiliary-pronoun combination in the category of
non-localised canonical tag variants. It occurs five to six times more often than the


70
56%
60 53% (18)
49% 50% (42)
(68) (3)
50

35%
40 33%
(52) 29% isn’t it
(11)
25% (21) innit

Percentage
30
(33)
16%
20 12% (18)
(20)
10

0
male female old middle young
(N = 166) (N = 150) (N = 130) (N = 113) (N = 73)
SEX AGE

Figure 6.1  Distribution of neg-tag variants across speaker sex and age
(Figures in round brackets indicate raw token numbers, including all non-localised canonical and infrequent tag
variants. The results for innit are derived from data produced by innit-users only [N = 145].)
Chapter 6.  Negative polarity question tags (neg-tags) 197
198 The Structure of Discourse-Pragmatic Variation

next most frequent combinations: aren’t they, don’t they and don’t you. The com-
parative frequency of isn’t it goes some way towards explaining why this form has
developed reduced variants (in’t it, innit) and started to level across the inflectional
paradigm (see Bybee [2003: 615–617, 2011: 66] on the role of frequency effects in
phonological change and analogical levelling). Treating isn’t it and the aggregate
of other non-localised canonical tags as two separate variants enables us to: (i)
explore whether the distribution of isn’t it tallies with that of other canonical tags
or with that of its reduced co-variant innit; and (ii) assess whether it is reduced
as well as non-reduced variants of third person singular neuter tags with present
tense be that are levelling across the inflectional paradigm.

Table 6.4  Inventory and frequency of auxiliary-negative-pronoun combinations

in non-localised canonical tags
N % N % N %
isn’t it 72 33.3 aren’t we 3 1.4 couldn’t it 1 0.5
aren’t they 14 6.5 didn’t (s)he 3 1.4 couldn’t we 1 0.5
don’t they 13 6.0 didn’t it 3 1.4 didn’t they 1 0.5
don’t you 12 5.6 haven’t they 3 1.4 didn’t you 1 0.5
didn’t we 8 3.7 wasn’t (s)he 3 1.4 doesn’t (s)he 1 0.5
wasn’t it 8 3.7 can’t they 2 0.9 haven’t we 1 0.5
isn’t he 6 2.8 doesn’t it 2 0.9 shouldn’t I 1 0.5
wouldn’t you 6 2.8 don’t we 2 0.9 wasn’t I 1 0.5
can’t you 5 2.3 hasn’t (s)he 2 0.9 wasn’t they 1 0.5
aren’t you 4 1.8 haven’t you 2 0.9 weren’t it 1 0.5
couldn’t you 4 1.8 shouldn’t you 2 0.9 weren’t they 1 0.5
isn’t there 4 1.8 wasn’t we 2 0.9 won’t it 1 0.5
wouldn’t it 4 1.8 weren’t you 2 0.9 won’t they 1 0.5
wouldn’t they 4 1.8 won’t you 2 0.9
hasn’t it 4 1.8 wouldn’t we 2 0.9 TOTAL 216

The figures provided below the x-axis in Figure 6.2 reveal very little evidence of
analogical levelling in the BwE neg-tag system. 91% of all neg-tag tokens in-
cluded in the tabulation occur in paradigmatic contexts. The fact that only 9% of
all neg-tag tokens occur in non-paradigmatic contexts suggests the following.
The overall predominance in the data of neg-tags with is and it (47%, N = 149,
including isn’t it, innit, in’t it) is not diagnostic of the neg-tag system’s advanced
stage of analogical levelling. Rather, it is indicative of a more general trend whereby
speakers are more likely to tag propositions containing third person singular neu-
ter subjects and present tense be than they are to tag those with other subject-
verb combinations. When the overall frequency of non-paradigmatic tag usage is
tabulated separately for individual variants, they turn out slightly higher for innit
 60
50%
48%
(63)
(136) 44%
50 (8) polarity question tags (neg-tags) 199
Chapter 6.  Negative

40
Percentage

27% canonical (w/out isn’t it)

30 23% (8) isn’t it
(67) innit
17%
20 60 (5)
50%
48%
(63)
(136) 44%
10 50
(8)

0 40
Percentage

paradigmatic (91%, N = 287) non-paradigmatic (9%,27%

N = 29) canonical (w/out isn’t it)
30 23% (8) isn’t it
(67) innit
17%
20 (5)

10

0
paradigmatic (91%, N = 287) non-paradigmatic (9%, N = 29)

Figure 6.2  Distribution of neg-tag variants across syntactic-semantic contexts

(Figures in round brackets indicate the raw token numbers, including infrequent variants.
The results for innit are derived from data produced by innit-users only [N = 145].)

(11%, N = 8 out of 71) than for isn’t it (7%, N = 5 out of 72). Taking into account
the restricted data on which these figures are based, the results tentatively suggest
two findings. Firstly, full and reduced variants of third person singular neuter tags
with be may not be spreading across the inflectional paradigm at the same speed.
Secondly, the degree of innit’s spread across the inflectional paradigm in BwE lags
far behind that reported for COLT (56%, N = 181 out of 323) (Andersen 2001: 108).
Comparison of the differently shaded bars along the y-axis in Figure 6.2 con-
firms that different neg-tag variants are differently distributed across syntactic-
semantic contexts. While the aggregate of other canonical tags (light grey bars)
manifest a strong propensity to occur in paradigmatic contexts, the distribution
of innit (black bars) and isn’t it (dark grey bars) is relatively unconstrained by the
syntactic-semantic properties of the preceding anchor.47 What innit and isn’t it

47. Closer inspection of the canonical tag tokens that were coded as not matching the seman-
tic-syntactic properties of their preceding anchors indicates that they cannot be construed as
straightforward evidence of analogical levelling. In the example We could go down to the beach
all day, didn’t we?, the speaker’s use of did instead of could in the tag may result from the fact
that in spontaneous conversation speakers tend to repeat the general content rather than the
exact words of propositions (Axelsson 2011: 36). The general idea expressed in the anchor is that
when the speaker and her co-interviewee were young, they were able or allowed to visit the beach
unsupervised. It may therefore be the case that in the tag, the speaker no longer focuses on the
freedom she and her friend enjoyed in their youth but on the actions they performed as a result
200 The Structure of Discourse-Pragmatic Variation

also share is their tendency to replace canonical tags of the form doesn’t it (e.g.
It depends, innit? That kind of puts the span on the works, innit? It still means at-
tractive, though, isn’t it?). 50% of all environments in the data which, according
to the rules of Standard English tag formation, require the tag form doesn’t it are
realised with innit or isn’t it. No other canonical tag is substituted with innit or isn’t
it to the same extent. This result conforms to Andersen’s (2001: 175) prediction
that non-paradigmatic innit uses are favoured in environments with it-pronouns
and present tense verbs.
Figure 6.3 examines the distribution of the variable and its variants across
discourse-functional factors. The percentages provided below the x-axis reveal that
almost half of all neg-tags in the data serve to induce addressees’ involvement in
the interaction and more than a quarter function to signal an attitudinal stance. At
5–9%, the remaining tag functions, i.e., the signalling of epistemicity, alignment
and mitigation, are comparatively infrequent.48 The distribution of variants across
the functional categories is shown by the bars along the y-axis. They demonstrate
a tendency for canonical tags (light grey bars) to perform conducive functions,
i.e., marking epistemicity and inducing involvement, and for innit (black bars) to
perform non-conducive functions, i.e., signalling stance and alignment. Isn’t it is
more evenly distributed across function than other variants but hardly appears in
the function most strongly associated with innit, i.e., attitudinal stance marking.
Formal variation in neg-tag usage, then, is at least to some extent constrained by
the function they perform in discourse.
Chapters 4 and 5 have shown how high degrees of inter-speaker and intra-
group variation in the rate of use of discourse-pragmatic variables (and their vari-
ants) can distort the results of multiple regression analyses if their effect is not
catered for through inclusion of individual speaker as a random effect in mixed-
effects analyses. By plotting the frequency of neg-tag usage across the 18 male
speakers (black bullets) and the 18 female speakers (grey squares) in the sample,

of this freedom. In He only passed away em this year, wasn’t it?, it seems that when the speaker
utters the filled pause em to signal her on-line presence, she mentally inserts it was in her utter-
ance. The following tag matches this mental, though not verbal, insertion (It was this year, wasn’t
it?). Finally, in He died with cancer, wasn’t it?, the tag matches the syntactically reanalysed clause
It was with cancer that he died. In short, these anomalous tags seem to reflect (interactionally
motivated) performance errors rather than speakers’ indiscriminate use of selected tag forms
across the inflectional paradigm.
48. Since the results in Figure 6.3 are based on neg-tags only, we must not infer from them
that they confirm the general marginality of epistemic tags and the overall persuasiveness of
involvement inducing tags reported, for example, in Holmes (1982: 24) or Tottie & Hoffmann
(2006: 301–302).


70 62%
60% 61% 60%
(49) (93)
(3) (9)
60 52%
(15) 50%
(5)
50
37%
40 (6)
canonical (w/out isn’t it)
31% 30% 30%
(9) isn’t it
26% 27% (8) (8)
25% innit
30 (24) (41)

Percentage
(4)

20 14%
(5)
9%
(8)
10

0
epistemic m. attitudinal st. m. involvement i. mitigation d. alignment s.
(9%, N = 29) (29%, N = 92) (48%, N = 152) (5%, N = 16) (9%, N = 27)

Figure 6.3  Distribution of neg-tag variants across functional categories

(Figures in round brackets indicate raw token numbers, including all infrequent variants. The results for innit are derived from data
produced by innit-users only [N = 145].)
Chapter 6.  Negative polarity question tags (neg-tags) 201
202 The Structure of Discourse-Pragmatic Variation

Figure 6.4 establishes that mixed-effects analysis is also required to produce reli-
able results for neg-tag variation. The scatter plot shows that neg-tag frequencies
range from zero to 38 tokens per 10,000 words in the youngest age group (on the
left), from zero to 53 in the middle age group (in the centre), and from two to 46
in the oldest age group (on the right). However, most young and old speakers’ rate
of neg-tag usage is below 20 tokens per 10,000 words. In the middle age group,
which has the most frequent and most excessive outliers, most speakers’ rate is
below 10 tokens per 10,000 words. To minimise the risk of providing inaccurate
significance estimates of social factors, the following multivariate analysis is con-
ducted with Rbrul.

60
Frequency per 10,000 words

50

40
male
30
female

20

10

0
15 25 35 45 55 65 75 85
speaker age

Figure 6.4  Normalised frequencies of neg-tags across individuals

6.5.2 Multivariate analysis

The mixed-effects multivariate analyses reported in this section will establish the
joint impact of the independent variables speaker sex, age, syntax and function to
the probability of the most frequently instantiated neg-tag variants while con-
sistently accounting for the high degrees of inter-speaker variation in neg-tag
usage discussed immediately above (see further Chapter 2.3.3 for details on Rbrul
analysis). Pitfalls associated with limited data are mitigated by recoding the five
functional categories identified in Section 6.4 into two broader categories: con-
ducive tags, i.e., those marking epistemicity and inducing involvement, and non-
conducive tags, i.e., those signalling alignment or an attitudinal stance. Because
they can be either conducive or non-conducive, tags initially categorised as miti-
gation devices were revisited to establish their conduciveness effect in their par-
ticular contexts of occurrence and ensure their accurate re-categorisation as either
 Chapter 6.  Negative polarity question tags (neg-tags) 203

conducive or non-conducive tags. Also, for the purpose of the regression analysis,
the variant isn’t it is conflated with the aggregate of other non-localised canonical
tags. While this re-configuration may conceal the differential patterning of isn’t it
and other canonical tags across syntax, it will not adversely affect the results but
rather improve their reliability by increasing cell sizes (Guy 1980).
Table 6.5 displays the results of two independent multivariate analyses of the
contribution of age, sex, syntax and function to the choice of innit and canonical
tags in the BwE interview data. The variant listed at the top of a column is the
one that was chosen as the application value for this particular run, with all other
neg-tag variants included in the run as non-application values. Only the data
from the thirteen speakers who use innit were included in the run for this vari-
ant; the data from all 33 neg-tag users in the sample were included in the run
for canonical tags. (For further details on how to interpret the results in Table 6.5,
see Chapter 2.3.3.)
The high input value for innit reveals that although this variant constitutes only
a quarter of all neg-tags in the data as a whole, it competes quite vigorously with
other tag variants when only those speakers are considered who actively use innit
in the data. Function is the only predictor included in the run that is selected as
making a significant contribution to the occurrence of innit. It is strongly favoured
for non-conducive tag uses such as marking an attitudinal stance and signalling
alignment. Cross-tabulations (not shown here) reveal that this effect is consistent
across the age groups represented in the data. Syntax does not exert a significant
effect to the occurrence of innit. The variant is only weakly preferred in paradig-
matic contexts.49 The effect of age on variant selection does not reach statistical
significance either, despite the constraint rankings and highly divergent factor
weights creating the appearance of an apparent-time rise in the use of innit. The
em-rule in Table 6.5 indicates that speaker sex was not included as a factor group
in the multivariate analysis for innit. This decision was taken because of severely
unbalanced data (see Section 6.5.1) and the fact that the token numbers for female
speakers (N = 6) are well below the threshold for obtaining reliable results (Guy
1980). Nevertheless, several indicators suggest that speaker sex is the most impor-
tant explanatory factor group for the occurrence of innit by far. Not only is innit
used almost exclusively by male speakers in the sample but males also contribute
all but 3 tokens of innit to the data.

49. The results for syntax are based on very low token numbers in non-paradigmatic contexts.
However, since acceptable levels of accuracy can be obtained with numbers in excess of ten
tokens per cell (Guy 1980), they must not be dismissed out of hand.
204 The Structure of Discourse-Pragmatic Variation

Table 6.5  Contribution of external and internal factors to the probability of innit and non-
localised canonical tags (with infrequent variants included as non-application values)
innit non-localised canonical tags
(including isn’t it)
input prob. 0.417 0.754
total N 145 316
deviance 177.508 289.789
factor % N log odds factor % N log odds
weights weights
sex p = 0.0172
male – – 139 – .328 50.3 167 [−0.716]
female – –    6 – .672 88.6 149 −[0.716]
range .344
age not significant not significant
young [.642] 56.2   32 −[0.586] [.380] 37.0   73 [−0.488]
middle [.573] 52.5   80 −[0.294] [.410] 38.1 113 [−0.365]
old [.293] 33.3   33 [−0.880] [.701] 56.9 130 −[0.853]
range
syntax not significant not significant
paradigmatic [.562] 49.6 127 −[0.248] [.624] 70.7 287 −[0.508]
non-paradigmatic [.438] 44.4   18 [−0.248] [.376] 44.8   29 [−0.508]
range
function p = 1.96e-06 p = 7.12e-07
non-conducive .737 61.0 100 −1.030 .292 42.3 130 ]]−0.885
conducive .263 22.2   45 −1.030 .708 86.6 186 −]]0.885
range .474 .416
speaker random st. dv 0.57 random st. dv 1.082

As is the case with innit, function and speaker sex play an important role in af-
fecting the occurrence of non-localised canonical tags. However, in stark contrast
to innit, canonical tags are strongly favoured for conducive tag uses such as sig-
nalling epistemicity and inducing hearers’ involvement in the interaction. This
effect is consistent across age and speaker sex, as revealed by cross-tabulations
not reproduced here. The range values show that speaker sex exerts the second
most important effect to the occurrence of canonical tags. While innit is strongly
associated with male speakers, canonical tags are favoured by the female speak-
ers in the data, most noticeably by those in the middle and young age groups (as
revealed by hidden cross-tabulations). Despite the divergent factor weights which
associate canonical tags with older speakers and paradigmatic contexts, age and
syntax are not selected as making a significant contribution to the occurrence of
 Chapter 6.  Negative polarity question tags (neg-tags) 205

these tag variants. The appearance of an age effect is produced by unbalanced data
in the middle age group. The non-significant effect of syntax is a result of conflating
isn’t it with the other non-localised canonical tag forms. As shown in Figure 6.2
above, the latter show a markedly stronger tendency than the former to occur in
paradigmatic contexts. Also, the eight canonical tags (excluding isn’t it) that do
not match the syntactic-semantic properties of their preceding anchors may be
construed as performance errors rather than non-paradigmatic tokens sensu stricto
(see footnote 47 above). Despite the factor weights, then, there is strong reason to
assume that canonical tags are generally favoured in paradigmatic contexts.

6.5.3 Summary of results

The preceding analysis into the formal variability of neg-tags has yielded the
following key findings:
– The variable context is unevenly partitioned between variants: The neg-tag
system in BwE is largely split between non-localised canonical tag variants and
the supra-local canonically-derived variant innit. The remaining supra-local
variants with din and in’t and the localised canonical tag variants with divn’t
and no occupy a trivial niche in the pool of available variants.
– The most frequent variants in the data have complementary social and func-
tional meanings: The use of non-localised canonical tags in the BwE corpus
is correlated with female speakers and conducive functions. Conversely, the
use of innit is strongly associated with male speakers and non-conducive
functions.
– The BwE neg-tag system has not been affected by analogical levelling to any
great extent: The vast majority of neg-tags in the data occur in contexts where
they match the syntactic-semantic properties of their preceding anchor. The
variant most likely to occur in non-paradigmatic contexts is innit, followed by
isn’t it.
Another important finding, generated through inspection of both the variable and
its variants’ distribution across individual speakers, is the following:
– The data exhibit a considerable amount of inter- and intra-group variability
in neg-tag usage: Although the pragmatic meaning attached to the most
frequent variants is consistent across the speaker sample, the frequency of
neg-tags and individual variants varies markedly across social cohorts and
individual speakers. This is particularly true of the variant innit which is only
used by a subsample of thirteen speakers, eleven of whom are male.
206 The Structure of Discourse-Pragmatic Variation

6.6 Discussion

This chapter set out to establish the rules governing the formal variability of neg-
tags and the entrenchment of innovating tag uses in BwE, a variety of British
English remote from those varieties from which tag innovations are hypothesised
to be diffusing. Qualitative and quantitative methods were combined to develop
a functional taxonomy of neg-tags in the BwE corpus and to model the choice
process between neg-tag variants. This section discusses the key findings of the
preceding analysis, situates them vis-á-vis previous studies of neg-tag variation
in other varieties, and explores the origins of innit in BwE.
The detailed qualitative data analysis in Section 6.4 served to establish the
functional inventory of neg-tags in order to allow testing the effect of function
on variant selection. It confirmed several findings obtained in previous functional
analyses of tag usage. Firstly, neg-tags are used by speakers to signal their atti-
tudes towards propositions as well as hearers. Secondly, conduciveness is not a de-
fining property of their use. In the BwE dataset, the tags that function subjectively
to seek hearers’ verification and confirmation of propositions or intersubjectively
to draw co-conversationalists into the discourse tend to invoke a response from
listeners. By contrast, the tags that function subjectively to emphasise speakers’
attitudinal stance towards their propositions or intersubjectively to signal their
alignment with prior talk do not solicit a response from listeners. Beyond con-
firming the strong link between the function tags perform and the conducive-
ness they entail (see in particular Kimps 2007: 274–280), these results show that
non-conduciveness is not limited to those tags that signal hostility or aggression,
as implied by Algeo (1988, 1990) and Cheshire (1981), but extends to those tags
that signal a neutral or assertive stance as well as those that constitute a positive
politeness device (e.g. when they signal approval of or involvement with prior
speakers’ talk). Moreover, the fact that some 40% of neg-tags in the data analysed
here are non-conducive calls into question previous functional accounts such as
those by Cheshire (1981) and Holmes (1982) which imply that conducive tags
are the unmarked and non-conducive tags the marked usage (see Andersen 1998
for a similar criticism).
The distinction between conducive and non-conducive tag functions has
proved essential in accounting for the distribution of non-localised canonical tags
and the supra-local tag variant innit. The former are strongly favoured for condu-
cive functions, in particular for securing and maintaining listeners’ involvement
in the interaction; the latter are strongly favoured for non-conducive functions,
in particular for emphasising speakers’ attitudinal stance towards their proposi-
tions. The strong association of innit with non-conduciveness and subjectivity
 Chapter 6.  Negative polarity question tags (neg-tags) 207

may be symptomatic of a more general pattern whereby reduced tag variants are
not usually response-soliciting or hearer-oriented. The BwE data contain 14 tag
tokens of the form in’t + pronoun where in’t is a reduced form of either isn’t (e.g.
It’s unbelievable, in’t it?) or aren’t (e.g. They’re wise, in’t they?). Similar to innit, the
majority of these tokens are non-conducive (71%, N = 10), half of them signalling
an attitudinal stance (50%, N = 7). Cheshire’s (1981) Reading data contain tags with
the reduced form [eɪnt] + pronoun as well as tags with the even more reduced form
[ɪnt] + pronoun. The latter are more strongly associated with lack of conduciveness
than the former.
Assuming, with Hoffmann (2006) and Tottie & Hoffmann (2006), that the
original meaning of question tags is interrogative, non-conducive tag uses can be
construed as representing a more advanced stage in the grammaticalisation of tag
meanings than conducive tag uses. Verification- and corroboration-seeking tags
(Examples (22)–(23) and (27) above) constitute genuine questions by virtue of
invoking listener response. Conversely, stance- and involvement-signalling tags
(Examples (24)–(25) and (28)–(29) above) are devoid of interrogative qualities
because they have undergone desemanticisation. In line with Traugott & Heine’s
(1991) implicational hierarchy of the type semantic > pragmatic > less semantic-
pragmatic, the interrogative meaning associated with conducive question tags has
faded away to the point that non-conducive tags do not have a response-elicitation
effect. In the current dataset, the loss of interrogative qualities in advanced stages
of the tags’ semantic-pragmatic development is reflected in the form and struc-
ture of the variant most closely associated with non-conducive tag usage: innit.
As a result of its structural reanalysis from three morphemes (is, -n’t, it) into a
single unit, innit is no longer easily identifiable as a reduced interrogative clause
composed of an auxiliary, negator and pronoun. Evidence from analyses of innit
in London English suggests that the variant’s loss of analysability and composi-
tionality leads to the gradual development in this variety of functions even more
remote from tags’ original verification- and corroboration-seeking meaning than
the non-conducive­stance- and involvement-signalling meanings discussed above.
Andersen (2001: 139–156) reports instances of innit that constitute utterances in
their own right, as in (30). Pichler & Torgersen (2012) report instances of innit that
occur in non-clause final position, as in (31). Like the non-conducive neg-tags
described in Section 6.4.1, the innit tokens in (30) and (31) do not constitute genu-
ine questions aimed at eliciting verbal contributions from listeners. Rather, they
serve to mark the current speaker’s agreement with the previous speaker’s proposi-
tion (30), or to foreground new information (31). They thus differ phonologically,
semantically and pragmatically from their source form, i.e., response-soliciting
canonical tags of the type auxiliary + -n’t + pronoun.
208 The Structure of Discourse-Pragmatic Variation

(30) Ken: They’re about a year behind us in fashion.

Selum: Innit man.  (Andersen 2001: 149)
(31) Tina: I’m a lot smaller than all of them man and who were like, “whoa.”
I mean, the sister, innit, she’s about five times bigger than you.
 (Pichler & Torgersen 2012)

Returning to the functions identified for neg-tags in the BwE data, the devel-
opment of non-conducive tags from conducive tags sketched out above coun-
terexemplifies widely attested hypotheses about the unidirectionality of semantic
change: that subjective uses of constructions chronologically precede and give
rise to intersubjective uses (Traugott 2003b, 2010; Traugott & Dasher 2002: 40);
and that a reverse order of development from intersubjective to subjective can be
ruled out (Traugott & Dasher 2002: 281). Although the conducive neg-tags in
the data are not categorically intersubjective (see Section 6.4.1 and Figure 6.3 in
Section 6.5.1), it is clear that neg-tags are intersubjective from the outset (see also
Traugott 2012: 10–11). By seeking listeners’ support or corroboration of proposi-
tions (see Example (27) above), they signal speakers’ attention to their hearers.
At the same time, the data examined above demonstrate that in the process of
desemanticisation neg-tags develop new meanings which are primarily, though
not exclusively, subjective (see Section 6.4.1 and Figure 6.3 in Section 6.5.1). By
conveying an attitudinal stance and speaker commitment (see Examples (24) and
(25) above), they signal speakers’ orientation not towards their hearers but towards
their propositions. The development of neg-tags can therefore not be described
as conforming to a strict unidirectionality between subjective and intersubjective
meanings. It suggests that widely reported regularities of the type subjective > in-
tersubjective are at best a weak hypothesis and that grammaticalising constructions
do not by definition follow a single, rigid sequence of meaning changes.
In addition to the functional split discussed earlier, the preceding quantitative
analysis has also uncovered a social split between non-localised canonical tags and
the supra-local tag variant innit. Canonical tags are favoured by females; innit is
used almost exclusively by males. A possible explanation for the strong association
of innit with male speech is its functional compartmentalisation as an attitudinal
stance marker noted in Section 6.5.2. There is ample evidence in the literature that
women prefer a more addressee-oriented conversational style than men (see, for
example, Coates 2004: 85ff; Woods 1997: 117). Question tags in particular tend to
be used by women more than by men to signal politeness and involve addressees in
the interaction (Cameron et al. 1988: 85; Holmes 1984a: 54, 1987: 73, 1995: 83–85).
This trend is echoed in the current dataset. Women produced 66% (N = 101) of
all 152 tags with a corroboration-seeking function but only 13% (N = 12) of all 92
tags with a stance-signalling function. The virtual non-existence of innit in female
 Chapter 6.  Negative polarity question tags (neg-tags) 209

speech may therefore be due largely to its strong association with an assertive,
speaker-oriented conversational style that is not generally embraced by women (or,
if embraced, is achieved by means other than the use of question tags). More gen-
erally, these patterns demonstrate that not only do the women and men recorded
for this project use different tag variants but they exploit the availability of these
variants to pursue different interactional styles.
If we adopt, for now, the prevalent view that innit is innovating in and diffusing
from varieties in southern England (Cheshire et al. 2005: 155–158; Krug 1998), its
functional compartmentalisation as a stance marker in BwE might explain why it
is adopted by men at a faster rate and/or to a greater extent than women. Several
studies have pointed out that supra-local phonological variants are more likely
to be adopted by women than men (see, inter alia, Foulkes & Docherty 1999: 16;
Kerswill 2003: 226–227; Milroy et al. 1994: 26; Watt & Milroy 1999: 40, 43). The
fact that the supra-local variant innit departs from this well-established sex pattern
may be diagnostic of more general differences between the spread of discourse-
pragmatic and phonological variants as well as the important role of function in
discourse-pragmatic change. Because discourse-pragmatic variants are never void
of pragmatic meaning, their adoption may be intrinsically tied up with their func-
tionality and whether this functionality is compatible with the interactional style
pursued by potential adopters. Whatever the precise role of function in the diffu-
sion of innit, the fact that the male speakers in the BwE corpus are at the forefront
of the use of innit is consistent with Andersen’s (2001: 207) hypothesis that females
are the innovators and males the leaders in changes affecting innit (see also Pichler
& Torgersen 2012). (Alternatively, male Berwickers’ favouring of innit over female
Berwickers might be due to the fact that the variant carries covert prestige and
is stigmatised. The OED (2010) describes innit as the “vulgar form of isn’t it.” As
Stubbe & Holmes (1995: 81) point out, discourse-pragmatic innovations involving
covert prestige are frequently adopted by men first.)
Several observations from the BwE data support the view that innit is diffusing
from southern varieties of English, in particular London English. Firstly, innit is
far less frequent in the northern variety of English investigated here than it is in
varieties spoken in southern England. In the BwE corpus (collected in 2003–2005),
innit accounts for 16% of all negative and positive polarity tags (N = 71 out of 432)
and for 48% of all tags formed with or derived from is + -n’t + it (N = 71 out of
149). Its average rate of occurrence in BwE is 2.7 tokens per 10,000 words. In COLT
(collected in 1993), innit accounts for 27% (N = 323 out of 1205) of all negative and
positive polarity tags in the data and for 52% of all tags formed with or derived
from is + -n’t + it (N = 323 out of 621). Its average rate of occurrence in COLT is
7.3 tokens per 10,000 words (Andersen 2001). The BNC (collected in 1991–1994)
210 The Structure of Discourse-Pragmatic Variation

records the highest frequency of innit out of all tags formed with or derived from
is + -n’t + it for south-west England (Krug 1998: 193–194). Secondly, the spread of
innit across the inflectional paradigm is by far not as advanced in the BwE data as
it is in COLT. The rate of non-paradigmatic innit tokens in the BwE corpus is 11%
(N = 8 out of 71) compared to 56% (N = 181 out of 323) in COLT. (This explains
to some extent the cross-variety frequency differences described above. Innit is
more frequent in London English because it occurs in a wider range of syntactic-
semantic environments.) Cheshire et al. (2005: 156) also report low overall and low
non-paradigmatic rates of innit in their adolescent data recorded in 1995–1999 in
Reading, Milton Keynes and Hull (14%; N = 5 out of 36). They take these figures to
suggest that innit has had a longer history and is more grammaticalised in London
compared to the communities they studied.
Yet the results presented in the current investigation caution against assessing
a form’s degree of grammaticalisation solely on the basis of its syntactic-semantic
distribution. As argued above, in BwE innit performs pragmatic functions that
can be located towards the more advanced end of the cline of semantic-pragmatic
change. This is not the case in COLT. Andersen (2001) does not describe innit as a
speaker-oriented variable which signals an assertive stance but as a hearer-oriented
variable which signals social rapport and aligned contextual assumptions (see also
Cheshire et al. 2005). Comparison of the BwE and COLT data, then, suggests that
innit is situated at opposite ends of the semantic-pragmatic and syntactic-semantic
clines of change in these varieties. The only other variant in the BwE data that
shows a propensity for use as an attitudinal stance marker is in’t + pronoun which
is only used by a small minority of speakers, mostly from the old and middle age
groups. Innit may have been adopted in BwE to fill a niche in the system of neg-
tag functions that has become vacant as a result of the decreasing use of in’t +
pronoun tags. In London English, on the other hand, innit may have developed as a
marker of social or ethnic differentiation. As pointed out by Andersen (2001: 192),
the actual form innit is not an ethnic minority feature because all ethnic groups in
London use it. However, non-paradigmatic uses of the form definitely are ethnic
identity markers because their frequency is vastly higher in ethnic minority speech.
The fact that innit has a different discourse-functional and syntactic-semantic
profile in BwE than in COLT does not in and of itself preclude a diffusion view of
the form’s distribution in the UK. Britain (2002: 617–618) points out that supra-
local innovations are not necessarily adopted wholesale but that they often interact
with local structures to produce local outcomes. Buchstaller (2008) provides evi-
dence for the outcome of such interaction on the level of discourse-pragmatics. She
demonstrates that the innovative quotative variants be like and go do not have the
same distribution in British English as they do in American English from where
 Chapter 6.  Negative polarity question tags (neg-tags) 211

they have diffused. She attributes these differences to the fact that not all properties
of be like and go were transferred in their spatial diffusion across the Atlantic but
that some social, formal and functional attributes were created anew as the vari-
ants were adopted into British English dialects. It is at least possible that a similar
process might have been at play in the adoption of innit into BwE. Innit might
have been taken on as a linguistic form without its pragmatic envelope. Instead of
putting innit to the same strategic use as their (mostly female) compatriots in the
south, male Berwickers might have attributed a new function to innit, i.e., one that
had been associated with a variant that has started to wane in BwE (see above),
thus giving a supra-local form a local meaning.
However, we need to acknowledge the possibility of an alternative origin for
innit. Rather than being the result of an outside adoption, innit might have devel-
oped through reduction from isn’t it and/or in’t it from within the dialect itself and
irrespective of any developments in southern British English. There is no reason to
assume that attrition and fusion, which are generally taken to be natural processes
of linguistic change resulting from frequency of use and repetition (Bybee 2003),
should only occur in one variety from which reduced forms spread rather than
occur simultaneously and independently across varieties. Among those Berwickers
who use the variant, innit is a well-established feature of BwE rather than a recent
development that is only gradually gaining ground. This is evident from the results
of the regression analysis in Section 6.5. The input value demonstrates innit’s high
probability of occurrence amongst the 13 innit-users in the sample; the constraint
hierarchy within the factor group age reveals only a non-significant rise in the
use of innit in apparent time. However, non-paradigmatic tokens of innit are rare
in the near-categorically mono-ethnic town of Berwick upon Tweed (99.6% of
the population were white British at the time of data collection), especially when
compared to multi-ethnic and multi-cultural London (see above). Because, ac-
cording to Andersen (2001), non-paradigmatic uses of question tags thrive in
multilingual communities and because ethnic minority speakers are the social
forces driving tags’ analogical levelling, we could argue that the form innit per se
may be an independent development across varieties of UK English and that it
may only be the form’s non-paradigmatic uses that are diffusing from multi-ethnic
London (and possibly other multi-ethnic cities such as Birmingham). The low rate
of non-paradigmatic tokens of innit as opposed to paradigmatic tokens in the BwE
data may then be attributable to the ethnic composition of Berwick upon Tweed
and the lack of ethnic minority speakers who would generally lead the analogi-
cal levelling of innit. The non-existence of non-paradigmatic innit in Andersen’s
(2001) exclusively white British speaker sample from Hertfordshire tentatively
supports this hypothesis.
212 The Structure of Discourse-Pragmatic Variation

6.7 Conclusion

The preceding analysis of neg-tags has shown that variation in their realisation is
not random but systematically constrained by extra- and intra-linguistic factors.
The overarching finding to emerge from the analysis is the complementary func-
tional and social distribution of non-localised canonical tags and the supra-local
tag variant innit. Comparison of the results obtained in the present study with
those obtained elsewhere, notably Andersen (2001), has demonstrated the com-
plexity and function-dependency of discourse-pragmatic variation and change. It
warns researchers against assuming that the occurrence of a discourse-pragmatic
variant in different varieties is necessarily a result of its social and geographical dif-
fusion, and that identical variants necessarily carry the same social and functional
meanings across varieties. The results thus highlight the importance of closely in-
vestigating the sociolinguistic distribution of discourse-pragmatic variants before
drawing any cross-variety comparisons about their use.
Without access to more data from across Britain any hypotheses about the
origins, distribution and dissemination of innit must necessarily remain tentative.
Accountable analyses of its extra- and intra-linguistic distribution across a range
of British English varieties are needed to fully understand how and whether innit
is diffusing across the UK. The study of question tag variation and change would
also benefit from an increase in diachronic studies which closely investigate the
semantic-pragmatic trajectory of tag usage over extended time periods.
part iii
 chapter 7

Discussion & conclusion

7.1 Introduction

This project was devised to demonstrate the theoretical insights that can be gained
into the structure of synchronic language variation and the interactional mecha-
nisms creating it by subjecting discourse-pragmatic variables to systematic vari-
ationist analysis. To this end, the book has developed an innovative methodology
which draws on the combined resources of variationist sociolinguistics, gram-
maticalisation studies and conversation analysis in order to uncover the complex
organisation of discourse-pragmatic variation in synchronic dialect data. The
method was applied to the analysis of the three variables i don’t know, i don’t
think and neg-tags which were chosen for analysis on the basis of sharing the
following properties: they function predominantly on the discourse-pragmatic
level of the linguistic system where they perform multiple interpersonal and textual
functions; they evince a high degree of formal variability which primarily affects
their morphemic structure and/or the realisation of their negative auxiliaries; and,
in the variety investigated here, they have variants which are differentiated by
their geographical distribution within England and the English-speaking world.
The analysis of these variables in a corpus of BwE, a peripheral variety of English
spoken in the far north-east of England, has established the grammar underlying
their formal variability in the data and demonstrated that they constitute for-
mulaic discourse units whose synchronic distribution in the data is a product of
changes associated with grammaticalisation. A key insight afforded by focusing the
analysis on three variables which are composed of a similar string of components
((pronoun) + (auxiliary) + (negator) + (verb); (auxiliary) + (negator) + (pronoun))
is that the sociolinguistic organisation of formal discourse variation is not entirely
uniform even across variables which are structurally related.
This chapter synthesises the results obtained in Chapters 4 to 6 with a view
to illustrating more fully the important new insights into language variation and
change which the rigorous and multi-faceted analyses in these chapters have
produced. Following a short account of parallel distribution patterns in the data,
Section 7.2 explores the mechanisms which give rise to the heterogeneous pat-
terning of formal variants across the three variables investigated in this book.
Section 7.3 discusses the far-reaching methodological and theoretical implications
216 The Structure of Discourse-Pragmatic Variation

of the results obtained in this study for grammaticalisation studies, variationist

sociolinguistics and linguistic theory more generally. It also examines why the
patterns of variation and change uncovered here do not correspond fully to those
reported for variables at other levels of the linguistic system, and outlines why
formulation of general principles of discourse-pragmatic variation and change
may be an objective that is not easily attained. Finally, Section 7.4 is the conclu-
sion to this chapter.

7.2 Synthesis of the results

Investigation into the formal variability of i don’t know, i don’t think and
neg-tags required identifying the whole inventory of variants for each variable,
operationalising hypotheses about selection constraints as independent predictors
and isolating the range of functions performed by the variables in the data, and, fi-
nally, quantifying the distribution of variants across the extra- and intra-linguistic
predictors hypothesised to constrain variant distribution. This procedure yielded
several important findings. Firstly, the speakers included in the sample have the
option of encoding the selected variables with a range of variants which can be
broadly divided into non-localised, supra-local and localised variants. Secondly,
the targeted variables perform a wide range of functions which, for the purpose
of quantification, can be divided into broad categories (e.g. referential vs. prag-
matic; interpersonal vs. textual; conducive vs. non-conducive) or more narrow
categories (e.g. hedge, booster, affiliator, mitigator, frame). Thirdly, the formal
variability in the data is highly structured and systematically constrained by mul-
tiple contextual predictors. Comparison of the results obtained in Part II reveals
that the structure of variation exhibits important similarities and dissimilarities
across variables and variants.
One of the overarching parallelisms to emerge from the preceding analyses is
the fact that the occurrence of non-localised and supra-local variants of i don’t
know, i don’t think and neg-tags is consistently conditioned by discourse-
functional constraints (see, however, I dono), with the result that: I don’t know
is favoured for referential meanings and I dunno for pragmatic meanings (see
Tables 4.6 and 4.7 in Chapter 4.5); I don’t think is associated with contexts where
the negator is interactionally preferred and I doØ think with those where it is dis-
preferred (see Figure 5.5 in Chapter 5.5); canonical tags are favoured for conducive
functions and innit for non-conducive functions (see Table 6.5 in Chapter 6.5). It
follows from this that although the pool of formal variants available in BwE for en-
coding i don’t know, i don’t think and neg-tags is quite large, speakers’ choice
of variant is somewhat limited once functional constraints on variant usage are
 Chapter 7.  Discussion & conclusion 217

taken into account. i don’t know and neg-tags were also found to resemble each
other in that their full variants (I don’t know, canonical tag forms) and their fused
and reduced variants (I dunno, innit) encode distinct and complementary mean-
ings which represent different stages of semantic-pragmatic change. Full variants
are favoured for less semantically bleached meanings (referential, conducive); re-
duced variants are favoured for more semantically bleached meanings (pragmatic,
non-conducive). The parallelisms highlighted here demonstrate that functionality
plays a crucial role in constraining formal variability in discourse-pragmatics, and
that formal reduction is strongly suggestive of desemanticisation (see, however, I
doØ think discussed in Section 7.3 below).
Moreover, multivariate analyses of the non-localised and supra-local variants
discussed above have shown that when social constraints are at all implicated
in variant distribution, their effect is strongly linked with variants’ functionality
(canonical tags, innit) and/or is less important than that of function (I don’t know,
canonical tags) (see Table 4.6 in Chapter 4.5; Table 6.5 in Chapter 6.5). The weak
or non-existent conditioning effect of the broad social factors age and speaker
sex can be attributed to the remarkable constancy of variant meanings across the
speaker sample. Cross-tabulations of the data have uncovered that although social
groups and individuals vary in their frequency of use of non-localised and supra-
local variants, they consistently use them with the same broad meanings. (Minor
meaning fluctuations are observable within broad functional categories. However,
they are neither significant nor symptomatic of apparent-time changes.) The social
consistency in the strategic use of non-localised and supra-local variants suggests
that the functional usage patterns described above have come to be fossilised in
BwE. Their fossilisation may be motivated by the fact that consistent usage patterns
of the type described above promote communicative efficiency.
This is not to say, however, that the formal variability of the targeted variables
is completely void of social meanings. The regression analyses in Chapters 4 and 5
have uncovered important parallelisms relating to the localised variants of i don’t
know and i don’t think (I divn’t knaa, I divn’t think) which show that their occur-
rence is consistently affected by broad social factors (see Table 4.9 in Chapter 4.5;
Table 5.3 in Chapter 5.5). In addition to confirming the social meaning-making
potential of formal discourse variation previously documented in the literature
(Cheshire 1981; Drager 2011; Moore & Podesva 2009), the analysis of localised
variants has shown that I divn’t knaa and I divn’t think resemble each other in terms
of their functional distribution. Both variants occur with roughly equal probability
across the functional categories established for their respective variables.
The preceding outline of the similarities in the conditioning of formal vari-
ation patterns in the data could not avoid drawing attention to their dissimilari-
ties. Due to the consistent conditioning effect of discourse-functional factors on
218 The Structure of Discourse-Pragmatic Variation

non-localised and supra-local variants and the consistent effect of extra-linguistic

factors on localised variants, the occurrence of variants with different geographical
spreads is constrained by different parameters. As a result, their functional and
social meanings are necessarily opposing. Non-localised variants (of i don’t know
and i don’t think) are functionally compartmentalised markers of social cohe-
sion (see Tables 4.6 and 4.7 in Chapter 4.5; Table 5.3 and Figure 5.5 in Chapter 5.5);
localised variants (of i don’t know and i don’t think) are functionally versa-
tile markers of social differentiation (see Table 4.8 in Chapter 4.5; Table 5.3 and
Figure 5.5. in Chapter 5.5). (The distribution of neg-tag variants is asymmetrical
in a different way and will be sketched out further below.)
Additional and more intricate disparities emerge when we examine in more
detail the precise functional and social meanings attached to individual vari-
ants. As alluded to above, the painstaking qualitative and quantitative analyses of
i don’t know and i don’t think in Chapters 4 and 5 have revealed strong cor-
relations between the variables’ realisations and their functions. However, these
correlations do not result in identical form-meaning correlations across the two
variables. With i don’t know, formal variability tallies with broad functional cat-
egories at a macro-level of analysis. Full variants pattern with the variable’s use as
a referential expression of insufficient knowledge; reduced variants pattern with
its use as a discourse-pragmatic feature that serves a wide range of interpersonal
and textual functions (see Tables 4.6 and 4.7 in Chapter 4.5). With i don’t think,
formal variability reflects fine-grained functional differences at a micro-level of
analysis. Full variants pattern with interactional contexts where the negator is
the bearer of important focal information or where it is interactionally preferred;
reduced variants pattern with contexts where the negator constitutes a potential
face-threat (see Figure 5.5 in Chapter 5.5). These correlations demonstrate firstly
that the pragmatic values of full and reduced variants of don’t are not constant
but are contingent on their collocational context of occurrence, and secondly that
variation between reduction or lack of reduction is not random but is carefully
monitored and controlled by speakers (see also Bybee & Hopper 2001: 11; Plug
2010). With i don’t know, speakers exploit the formal variation to mark the differ-
ence between referential and pragmatic meanings. With i don’t think, speakers
either allow or suppress reduction depending on the immediate interactional and
social demands of the speech situation.
Another inconsistent pattern in the data involves the social indexicality of the
localised variants I divn’t knaa and I divn’t think. The former is strongly favoured
by young males (and, to a lesser extent, old females) (see Table 4.9 in Chapter 4.5);
the latter is strongly favoured by females (see Table 5.3 in Chapter 5.5). In the ag-
gregate of productive constructions in the data, social factors were found not to
impact on the occurrence of divn’t (see Table 3.4 in Chapter 3.4.2). The divergent
 Chapter 7.  Discussion & conclusion 219

social value of divn’t across the two formulaic constructions, i don’t know and
i don’t think, and the aggregate of productive NP-negative do-V constructions
analysed in Chapter 3 may be a product of well-established gendered interactional
styles. Men typically adopt a competitive, confrontational and dominant style;
women characteristically adopt a tentative, involved and affective style (see, inter
alia, Coates 1996, 2004; Holmes 1995; Tannen 1993). Young males are the most
productive users of the variable i don’t know as well as the variant I divn’t knaa
(see Figure 4.5 and Table 4.9 in Chapter 4.5). Because i don’t know is used in the
data predominantly for textual functions such as turn-regulation and topic-devel-
opment (see Figure 4.3 in Chapter 4.5), young males’ high use of I divn’t knaa may
be attributable to their pursuing an interactional style of control and dominance.
Conversely, females are the more productive users of the variable i don’t think
as well as the variant I divn’t think (see Figure 5.4 and Table 5.3 in Chapter 5.5).
Because i don’t think is used in the data predominantly for interpersonal func-
tions such as epistemicity and mitigation (see Figure 5.5 in Chapter 5.5), females’
high use of I divn’t think may be attributable to their pursuing a tentative and
hearer-oriented interactional style. In sum, divn’t, which carries no social mean-
ing in productive constructions (see Table 3.4 in Chapter 3.5), acquires social
meanings in formulaic constructions as a result of these constructions’ inherent
functionality. Alternatively, the divergent social value of divn’t in i don’t know
and i don’t think may be a product of the high variability and great instability in
the formal implementation of pragmatic functions (see, inter alia, Barbieri 2008;
Brinton 2001; Precht 2008). Social groups other than young males may rarely use
I divn’t knaa because they draw on features other than i don’t know to perform
textual functions. Similarly, male speakers may rarely use I divn’t think because
they exploit features other than i don’t think to signal interpersonal meanings.
The hypotheses formulated above are necessarily speculative and require empirical
verification. Also, they can be criticised for being essentialist and reductive because
they are based on binary gender distinctions and ignore the potential diversity of
gendered practices. Nevertheless, they demonstrate well that the social embedding
of variation in discourse-pragmatics is highly complex, and that the social meaning
of variants cannot easily be separated from variables’ functionality.
The analysis of neg-tags in Chapter 6 provides additional and more conclusive
evidence that discourse-pragmatic variants achieve their social meaning through
their functionality. Multivariate analyses of the most frequent tag variants in the
data revealed that unlike the occurrence of the majority of discourse-pragmatic­
variants investigated in this book, the occurrence of canonical tag variants and
the variant innit is governed by both extra-linguistic and discourse-functional
constraints. Canonical tags are favoured by females and for conducive functions
such as inducing others’ involvement in the interaction; innit is strongly associated
220 The Structure of Discourse-Pragmatic Variation

with males and favoured for non-conducive functions such as signalling an as-
sertive stance (see Table 6.5 in Chapter 6.5). As pointed out in Chapter 6.6, the
association of canonical tags with women may be ascribed to women’s preference
for a facilitative, addressee-oriented interactional style; the association of innit
with men may be ascribed to men’s preference for an assertive, speaker-oriented
interactional style. Because the social indexicality of neg-tag variants is parasitic
on their functionality, their socio-pragmatic meaning differs from that sketched
out above for non-localised and localised variants of i don’t know and i don’t
think. neg-tag variants are functionally compartmentalised markers of social
differentiation.

7.3 Implications of the results

The results synthesised in Section 7.2 have important methodological and theoreti-
cal implications for grammaticalisation studies and variationist sociolinguistics.
This section explores these implications with a view to illustrating the great im-
portance of studying discourse-pragmatic variables for developing current theories
of language.
Painstaking analyses in Chapters 4 and 5 have established that i don’t know
and i don’t think constitute formulaic discourse units whose synchronic dis-
tribution is a product of grammaticalisation changes operative prior to the time-
span covered by the BwE corpus. The co-existence in the data of full and reduced
variants of i don’t know and i don’t think was therefore attributed to attrition,
a grammaticalisation process reflecting changes to the cognitive representation
of constructions (compositional > non-compositional) (Bybee 2006). In the BwE
data as well as Bybee & Scheibman’s (1999) American English data, the rate of
don’t-reduction was found to be higher in i don’t know than i don’t think (67%
vs. 37% in BwE; 74% vs. 35% in American English).50 Bybee & Scheibman (1999)
attribute the higher rate of formal reduction in i don’t know in their data to the
fact that i don’t know has a higher discourse frequency than i don’t think
(N = 37 vs. N = 19). High-frequency words and constructions regularly undergo
reduction and fusion at faster rates and to greater extents than low-frequency
words and constructions (see, inter alia, Bybee 2001; Gregory et al. 1999; Jurafsky
et al. 2001). The BwE data support Bybee & Scheibman’s (1999) theory. The token
frequency of i don’t know, which is at the forefront of the reduction of don’t, is

50. The figures for BwE are based on consideration of non-localised variants only. For the pur-
pose of generating these figures, the semi-reduced variant I dono was conflated with the full
variant I don’t know.
 Chapter 7.  Discussion & conclusion 221

more than twice as high as that of i don’t think (N = 600 vs. N = 270). Yet at the
same time the results of the present investigation suggest that frequency may not
be the sole or main determinant of the differential degree to which i don’t know
and i don’t think are affected by on-line reduction processes.
As pointed out in Section 7.2 above, reduction and lack of reduction have a
different meaning in i don’t know and i don’t think. In i don’t know, they
mark the broad distinction between pragmatic and referential meaning; in i don’t
think, they signal fine-grained differences in pragmatic meaning which are inter-
actionally motivated by referential weight and concerns about face. Moreover, it
has been argued in Chapter 5.6 that the signalling of fine-grained nuances of mean-
ing through prosody affects the realisation of don’t in i don’t think. Reduction is
blocked when non-boosting tokens carry pitch prominence on don’t to signal high
degrees of doubt. The lack of consistent reduction in i don’t think is therefore
not exclusively due to its lower rate of occurrence compared to i don’t know, as
argued by Bybee & Scheibman (1999). It is at least to some degree also conditioned
by pragmatic factors and the way the variable is used in interaction as well as by
prosodic factors and the way prosody is implemented to affect meaning.51 These
findings have important implications which I will discuss below.
Although it has not been possible to observe grammaticalisation changes un-
fold in the time-span covered by the BwE corpus, the effects of these processes can
be observed in the synchronic variation patterns uncovered by the multivariate
analyses. With i don’t know, they demonstrate a strong link between semantic-
pragmatic shift and formal attrition: desemanticised uses are near-consistently
fused and reduced. The functional split between I don’t know and I dunno in the
BwE data can thus be conceptualised as the synchronic offshoot of form-meaning
changes which either have unfolded concurrently throughout the evolution of
i don’t know as a discourse-pragmatic variable or happen to coincide in the
time-frame analysed here. (The fact that older male speakers were found to be
more likely than other speakers to use I don’t know for pragmatic uses weakly
suggests that semantic-pragmatic changes may have preceded phonetic changes
by one generation.) With i don’t think, the patterns uncovered by the multi-
variate analyses demonstrate only a weak link between semantic-pragmatic shift
and formal attrition: desemanticised uses are mostly non-reduced. In light of the
findings highlighted in the preceding paragraph, the functional distribution of
I don’t think and I doØ think can thus be conceptualised as the synchronic offshoot
of form-meaning changes which, due to the strong interference of pragmatic and
prosodic factors, have not progressed in lockstep. Beyond adding to the canon

51. The differential rates of reduction may also be affected by the differential phonemic struc-
tures across word boundaries in i don’t know and i don’t think.
222 The Structure of Discourse-Pragmatic Variation

of literature challenging Croft’s (1990: 244) notion that the changes constituting

grammaticalisation are strictly synchronised (see inter alia Bisang 2004; Romaine
1995), the divergent distribution patterns for i don’t know and i don’t think
highlighted above have the following implications for grammaticalisation studies.
Firstly, they demonstrate that grammaticalisation does not affect constructions
homogenously, even if they are composed of near-identical strings of components.
Secondly, they show that the existence of non-concurrent grammaticalisation
changes is not random but structured and interactionally motivated, thereby pro-
viding important hypotheses about why grammaticalisation changes do not by
definition unfold concurrently.
Additional hypotheses about the asynchronous development of grammaticali-
sation changes were provided by the close analysis of the supra-local canonically-
derived tag variant innit in Chapter 6. Close scrutiny of its synchronic distribution
patterns revealed that although innit exhibits an advanced stage of formal reduc-
tion and semantic-pragmatic change, it has not been affected by analogical level-
ling to any great extent. Andersen (2001) argues that the spread of innit across the
inflectional paradigm thrives in multilingual environments and that in London
English it is driven by ethnic minority speakers. In view of Andersen’s proposal, the
near-categorical absence of non-paradigmatic tokens of innit in BwE was attributed
to the virtual non-existence of ethnic minority speakers in Berwick upon Tweed. It
follows that if different social groups activate and advance different changes consti-
tuting grammaticalisation (see Chapter 1.4) and if these groups are not universally
present across communities, the set of changes constituting grammaticalisation
may not co-evolve synchronously within, let alone across, varieties. As a result,
the synchronic distribution of grammaticalising features may not be parallel even
across closely related varieties. To fully illuminate the nature and progression of
grammaticalisation changes, it is therefore necessary to closely examine their social
dimension within and across varieties (see also Janda 2001).
Beyond grammaticalisation studies, the structured variability uncovered for
the variables investigated in Part II has important implications for variationist
sociolinguistics. As outlined at the beginning of this section and illustrated in
more detail in Chapters 4.5 and 5.5, the conditioning effect of extra- and intra-
linguistic predictors on patterns of formal variability is not homogeneous across
the constructions i don’t know and i don’t think. Non-localised variants have
differential semantic-pragmatic meanings across the two constructions; their local-
ised variants take on differential social meanings depending on the construction in
which they occur. Additional comparisons of the two formulaic constructions with
the aggregate of productive NP-negative do-V constructions in the data (e.g. they
divn’t like, you don’t see, we divn’t do, people don’t bother) show that the frequency
of don’t-reduction and divn’t-usage varies dramatically not just across formulaic
 Chapter 7.  Discussion & conclusion 223

constructions but also across formulaic and productive constructions. At 12%, the
aggregate of productive constructions have a much lower rate of don’t-reduction
than either i don’t know or i don’t think (67% and 37% respectively), which
may indicate that it is i don’t know and i don’t think that advance don’t-reduc-
tion.52 Conversely, divn’t has a markedly higher rate of occurrence in the aggregate
of productive constructions than in i don’t know or i don’t think (36% vs. 21%
and 18% respectively), which may be linked to the internal fixation and holistic
storage of these constructions in their non-localised forms. Moreover, the social
distribution of divn’t is more restricted in formulaic constructions than it is in
productive constructions. In the former, its use is strongly associated with one or
two social groups; in the latter, its use is largely unconstrained by the social factors
sex and age. The frequency and constraint differences outlined above demonstrate
the important role of formulaic constructions in shaping both the internal or-
ganisation as well as the social indexicality of linguistic variation, thus bolstering
the arguments presented in Chapter 1.3 with regard to the non-peripheral and
fully grammatical status of discourse-pragmatic variables in the linguistic system.
They also demonstrate that provision of accurate accounts of linguistic variation
crucially depends on studying grammatical morphemes in the constructions in
which they occur and isolating formulaic constructions from each other and from
productive constructions. Non-separation of formulaic or frequent constructions
risks obscuring important constraints on variant choice and leaving unaccounted
for their role in moulding linguistic variation and social meanings.
Equally important to the provision of accurate accounts of sociolinguistic vari-
ation in discourse-pragmatics is inclusion of function as an independent variable
in the quantitative data analysis. As highlighted in Section 7.2 above, function is
one of the key constraints on the formal variability of the discourse-pragmatic
features analysed in Part II of the book. Crucially, consideration of function has
shown that discourse-pragmatic variables do not achieve their social meanings
in the same way as phonological or morpho-syntactic variables (see Cheshire
[2005] and Cheshire et al. [2005] for similar arguments regarding higher-level
syntactic variables). The frequency with which formal discourse variants occur
across social groups is ultimately determined either by the function associated
with the variant in question (if it is a single-purpose variant such as innit), or by
the functional profile associated with its variable (if it is a multi-purpose variant
such as I divn’t knaa or I divn’t think). Because of their dependency on discourse-
functional conditioning and the fact that different social groups do not necessarily
pursue identical discourse styles (see Section 7.2 above), the use of supra-local or

52. These figures are based on consideration of the variant don’t only and do not include produc-
tive constructions with divn’t or dinnae.
224 The Structure of Discourse-Pragmatic Variation

localised discourse variants cannot be ascribed definitively to one social group as

is the case for example with supra-local or localised phonological variants which
have been consistently associated with female and male speakers respectively (see,
for example, Foulkes & Docherty 1999; Llamas 2007b; Milroy et al. 1994; Watt &
Milroy 1999). It follows that the generalisations drawn by Labov (1998) and others
regarding the social embedding of phonological variation and change are unlikely
to apply without restrictions to discourse-pragmatic variables. What is more, it
may prove unfeasible to formulate generalisations about the distinct social embed-
ding of discourse-pragmatic variation and change. Which social group it is that
drives semantic-pragmatic change will to a large part depend on the functions as-
sociated with evolving variables at different stages in their diachronic development.
This is not to imply that we should abandon the variationist analysis of dis-
course-pragmatic features. As demonstrated throughout this book, subjecting
discourse-pragmatic variables to accountable quantitative analysis provides im-
portant new insights into the creation and organisation of synchronic language
variation. Full investigation of every aspect of discourse-pragmatic variability is
therefore imperative to testing and developing current theories of language which,
by definition, must be based on systematic analyses of all levels of the linguistic
system. One of the key offerings of this book is a robust body of empirical evidence
for viewing discourse-pragmatic features as non-peripheral and fully grammatical
elements of the linguistic system which interact with more productive construc-
tions in shaping the structure of synchronic language variation.

7.4 Conclusion

In addition to advancing an analytical model for future investigations of discourse-

pragmatic variables which accommodates their full complexity, this book has il-
lustrated the descriptive, explanatory and theoretical potential of quantitative
discourse variation studies. The corpus-based variationist analysis of three vari-
ables that are closely related by virtue of being constituted of similar components
(i don’t know, i don’t think and neg-tags) has offered a number of previously
undocumented findings. By way of summary, they include:
– The occurrence of discourse-pragmatic variants with different geographical
spreads is governed by different parameters: Among the variables investigated in
this book, the occurrence of non-localised variants is governed by discourse-
functional constraints; that of localised variants by extra-linguistic constraints;
and that of supra-local variants by both discourse-functional and extra-lin-
guistic constraints.
 Chapter 7.  Discussion & conclusion 225

– The social indexicality of discourse-pragmatic variation is strongly dependent

on variables’ and/or variants’ functional compartmentalisation in a variety:
Because their use is motivated by their functionality, discourse-pragmatic vari-
ants derive their social meanings either from the functions they encode in a
given variety, or from the functional profiles associated with their variables in
that variety.
– Discourse-pragmatic variables play a key role in shaping linguistic variation and
social meanings: The frequency and sociolinguistic conditioning of linguistic
variants is affected by whether they occur in productive constructions or in
different formulaic constructions.
– Grammaticalisation does not affect variables and variants homogeneously:
Interactional constraints bar processes of grammaticalisation from operating
uniformly across closely related variables within a variety. Social constraints
hamper the uniform progression of grammaticalisation changes across identi-
cal variants in different varieties.
The results summarised above demonstrate the range and depth of insights that
can be gained from subjecting discourse-pragmatic features to rigorous qualitative
and accountable quantitative analysis. As discussed in Section 7.3, these insights
have far-reaching implications for current theories of language. Yet despite their
indisputable theoretical value, the number of variationist analyses of discourse-
pragmatic variables remains dwarfed compared to the number of variationist
analyses of phonological and morpho-syntactic variables. To conclude this book,
the next and last chapter explores some of the challenges scholars face in their
attempts to advance the field of discourse variation analysis beyond its current
rudimentary state and provide a more robust account of patterns of variation and
change in discourse-pragmatics.
 chapter 8

Challenges for the future

8.1 The state of the art and beyond

Just over a decade ago, Macaulay (2002a: 298) painted a rather bleak picture when
he remarked of the state of discourse variation analysis at the time:
[t]he study of discourse variation is still at an elementary stage. […] there are many
different approaches to the sociolinguistic investigation of discourse, and it would
take a braver person than I am to assert with confidence that we have much solid
information on gender, age, or social class differences. What we have are a number
of intriguing claims that need to be tested again and again, by the same or different
methods, in similar or different settings, with similar or different samples […].

Although concerns remain about the robustness of the findings obtained in dis-
course variation research, considerable progress has been made since the turn
of the century in the investigation of previously relatively unexplored aspects of
discourse-pragmatic variation and change. Among others, these include:
– the social trajectories of discourse-pragmatic change (see inter alia Tagliamonte
& D’Arcy 2009; Denis 2011);
– the role of youth in the creation and spread of new discourse-pragmatic vari-
ants (see inter alia Andersen 2001; Roth-Gordon 2007);
– social and geographical diffusion patterns of innovative and innovating dis-
course-pragmatic variants (see inter alia Andersen 2001; Buchstaller 2008;
Buchstaller & D’Arcy 2009; Cheshire et al. 2005);
– mechanisms of discourse-pragmatic change in language contact situations (see
inter alia Andersen 2001; Cheshire et al. 2011; Fuller 2001; Goss & Salmons
2000; Hllavac 2006; Levey et al. forthc.; Torres 2002);
– patterns of discourse-pragmatic variation across interactional, situational and
technological settings (see inter alia Escalera 2009; Fuller 2003; Schleef 2008;
Tagliamonte & Denis 2008; Verdonik et al. 2009);
– patterns of discourse-pragmatic variation and change in child and learner lan-
guages (see inter alia Furman & Özyürek 2007; Levey 2006, 2012; Liao 2009;
Müller 2005; Terraschke 2007);
228 The Structure of Discourse-Pragmatic Variation

– the role of discourse-pragmatic variables in the construction and negotiation

of social identities (see inter alia Bakht 2010; Drager 2011; Moore & Podesva
2009; Trester 2009); and
– social attitudes towards and perceptual loads of discourse-pragmatic variables
(see inter alia Buchstaller 2006b; Parton et al. 2002; Russell et al. 2008).
The studies cited above address key questions about the nature and mechanisms of
discourse-pragmatic variation and change. Together with the present investigation,
they thus constitute an important step in the development of variationist discourse
analysis and its expansion in new directions. However, for further advancement of
the field, it will not be sufficient for scholars to gather more empirical and experi-
mental evidence to substantiate or refute previous findings. They will also need to
overcome a number of challenges which continue to obstruct progress.
This chapter explores some of these challenges, focusing specifically on: the
low-frequency occurrence of many discourse-pragmatic features; the lack of a
uniform set of reliable methods for discourse variation analysis; the focus on
external and the insufficient attention paid to internal factors in accounting for
discourse-pragmatic variation; the shallow time-depth of synchronic corpora; and
the persistent trend to study a limited range of discourse-pragmatic variables while
neglecting the wide expanse of other variables. In what follows I will identify the
nature of these challenges and, where relevant, their origins; outline the reasons
why they need to be overcome to allow progress to be made; and, where feasible,
provide some suggestions as to how they might be overcome. What follows does
not constitute an exhaustive list of all the challenges facing the field of discourse
variation analysis. Rather, it represents a selective account of those challenges
which seem particularly salient to me and which I believe need to be addressed
most urgently if our ultimate aims are: firstly, to illuminate the sociolinguistic em-
bedding of variation and change in discourse-pragmatics; and secondly, to progress
towards a holistic theory of language variation and change which spans all levels
of the linguistic system.

8.2 Dealing with low token frequency

By their very nature, phonological and some lower-level morpho-syntactic variables

occur with high frequency even in relatively small corpora. Conversely, many dis-
course-pragmatic (and higher-level syntactic) variables tend to occur with far lower
frequency even in corpora of moderate size. This is due to the fact that their use
is strongly constrained by the operation of semantic-pragmatic and interactional-
situational factors (e.g. details of the speech event and situation; topic and purpose
of the interaction; speaker roles and relationships). Although understanding of
 Chapter 8.  Challenges for the future 229

these constraints has gradually improved (see Section 8.1 above for references), it is
not feasible to increase token numbers by predicting typical contexts of occurrence
of individual variables and concentrating data collection and analysis accordingly.
Not only do the pragmatic functions performed by discourse-pragmatic features
often overlap but they can be encoded by linguistic features from other levels of
the linguistic system (see Chapter 1.3). Alas, what exactly it is that conditions the
choice of one feature over another is, as yet, little understood.
As shown in Chapters 5 and 6 of this book, access to large token numbers need
not be critical to yielding intriguing results and generating testable hypotheses about
the mechanisms of discourse-pragmatic variation and change. Nevertheless, the
typically low-frequency occurrence of many discourse-pragmatic features in dia-
lect corpora may be problematic in at least two important ways. Firstly, the range of
discourse-pragmatic features involved in sociolinguistic variation and change may
remain hidden from observation if token numbers are severely limited. As a result,
it may not be possible to produce accurate accounts of the full extent of discourse-
pragmatic variation and change within and across varieties. Secondly, even where
variables occur with sufficient frequency to make possible detection of variation,
dialect corpora may not produce the amount of data required to uncover the gram-
mar underlying the observed variation. Without detailed knowledge of how vari-
ables and their variants are distributed in the linguistic system and/or the speech
community, understanding of the sociolinguistic embedding of discourse-pragmatic
variation and change cannot advance. In short, the overall paucity of many discourse-
pragmatic features may have descriptive as well as theoretical ramifications.
Studies of low-frequency morpho-syntactic variables have benefited from
the increasing availability of public corpora such as the British National Corpus
(Burnard 2000) or the International Corpus of English (Greenbaum 1992). Because
they are much larger than private corpora, they are more likely to yield token files
that are amenable to statistical analysis. However, D’Arcy (2011: 58–59) points out
that despite their extraordinary size, public corpora are not necessarily suitable
for the analysis of discourse-pragmatic variables: “the data are typically compiled
from fragments rather than texts in their entirety [and some texts] may be sys-
tematically cropped to fit with the overall design of the corpus.” This is alarming
because discourse-pragmatic variables tend to occur in clusters rather than being
evenly distributed throughout texts (Walker 2010: 74), and because the function of
discourse-pragmatic­variables cannot be determined independent of their larger
context of occurrence, as demonstrated in Part II of this book. D’Arcy (2011: 59)
also draws attention to Biber et al.’s (1999: 28) observation that the spoken com-
ponents of publicly available corpora have tended to be collected in limited and
artificial settings. This affects the representativeness and authenticity of the data,
and the validity of any conclusions drawn from their analysis.
230 The Structure of Discourse-Pragmatic Variation

Fortunately, the low-frequency problem is not insoluble. Firstly, some publicly

funded variationist projects have resulted in the construction of mega-corpora
such as, for example, the 1.4-million-word Linguistic Innovators Corpus (Kerswill et
al. 2007), the 1.2-million-word Multicultural London English corpus (Kerswill et al.
2011), the 3.5-million-word Ottawa-Hull Corpus (Poplack 1989), the 1.5-million-
word York English Corpus (Tagliamonte 2002b), or the 1.8-million-word Toronto
English Archive (Tagliamonte & D’Arcy 2007). These corpora constitute prolific
resources for the investigation of discourse-pragmatic variables, as demonstrated
in recent studies by Cheshire et al. (2011), D’Arcy (2007), Denis (2011), Ito &
Tagliamonte (2003), Levey et al. (forthc.), Tagliamonte (2005, 2008), Tagliamonte
& Denis (2010) and Waters (2013). Secondly, analyses of discourse-pragmatic vari-
ation and change in traditional sociolinguistic interviews can be supplemented,
or even replaced, with analyses of other data sources. Notwithstanding obvious
drawbacks such as their authenticity and comparability with corpus-based data of
vernacular speech, internet newsgroups, instant messaging and popular television
series provide valuable resources for investigating at least some discourse-pragmatic
variables (e.g. Buchstaller et al. 2010; Tagliamonte & Denis 2008; Tagliamonte &
Roberts 2005). In addition to boosting token numbers, web-based and media data
may add a real-time dimension and wider geographical angle to apparent-time
analyses, and capture short-lived innovations in discourse-pragmatics which may
go undetected in analyses of synchronic private corpora (Buchstaller et al. 2010).
These data thus have the potential not just to facilitate but also to enrich analyses
of (sparse) discourse-pragmatic variables in traditional corpora.

8.3 Developing reliable analytical methods

Discourse-pragmatic features “are used by speakers in such complex and sophisti-

cated ways [that] studying variation [and change] in their use is no straightforward
task” (Stubbe & Holmes 1995: 85). As previously discussed in this book, attempts
to subject discourse-pragmatic features to accountable variationist investigation
are complicated by two factors. Firstly, the Labovian framework is not easily ap-
plicable to the analysis of discourse-pragmatic variables (see Chapters 1.3 and
2.3). Secondly, the complexity of discourse-pragmatic features is not yet fully un-
derstood (Aijmer & Simon-Vandenbergen 2011). It would therefore be wrong to
assume that the advances outlined in Section 8.1 result from scholars having fully
surmounted the methodological difficulties they face in quantifying discourse-
pragmatic variation and change.
In stark contrast to quantitative studies of lower-level phonological and
morpho-syntactic variation and change which are relatively homogeneous and
 Chapter 8.  Challenges for the future 231

congruent in focus and methodology, studies of discourse-pragmatic variation

and change lack a shared set of methodological and analytical principles. Macaulay
(2002a: 284) neatly summarises the problems involved as follows:
[t]here is no general agreement on methods of collecting or analyzing data, on
what features are suitable for investigation, on how to identify possible discourse
features, and what significance to attach to the use of a particular feature […].

The persistent diversity of methods and the resultant lack of cross-corpora compa-
rability impede significant progress. They make it difficult, if not altogether impos-
sible, to synthesise the results of existing studies into a set of coherent findings, and
thus, to formulate empirically grounded generalisations of discourse-pragmatic
variation and change (see further Pichler 2010).
To address these problems, Pichler (2010) has advocated a uniform method
for discourse variation analysis whose consistent application ensures reliability,
intersubjectivity and comparability. The advocated method was elaborated and
refined in Chapter 2 of this book, and its efficacy for uncovering the synchronic
organisation of discourse-pragmatic variability was illustrated in Chapters 4 to 6.
In addition to yielding accountable results and facilitating cross-corpora compari-
son, a great advantage of this methodology is that it can be tailored to the idiosyn-
crasies of a range of discourse-pragmatic variables, and that it can be developed
and expanded to accommodate new theoretical insights into the complex nature
of discourse-pragmatic features.

8.4 Exploring intra-linguistic constraints on discourse-pragmatic variation

In stark contrast to diachronic grammaticalisation studies which pay close atten-

tion to the morpho-syntactic and semantic-pragmatic properties of discourse-
pragmatic features (e.g. Brinton 1996, 2008; van Bogaert 2010, 2011), synchronic
variation studies have tended to focus most attention on these features’ social
meanings (e.g. Barbieri 2008; Crosby & Nyquist 1977; Dailey-O’Cain 2000;
Macaulay 2002b, 2005; Stubbe & Holmes 1995; Woods 1991). The preoccupa-
tion with extra-linguistic factors at the expense of intra-linguistic factors is largely
due to the methodological challenges outlined in Chapter 2.3, particularly those
regarding the quantification of discourse-functionality and the circumscription
of the variable context in linguistic-structural terms. Investigations focusing on
social variation patterns in the overall frequency of discourse-pragmatic variables
sidestep these difficulties because such patterns are identifiable without allocat-
ing tokens to functional categories or determining where in the linguistic system
variables can but do not occur.
232 The Structure of Discourse-Pragmatic Variation

Consideration of social factors is, of course, a key concern of the variationist

enterprise. Linguistic variation marks social differentiation, and linguistic changes
are driven by social forces. However, the persistently narrow focus on social
constraints in discourse variation studies to the near-exclusion of interactional-
situationa­l, discourse-functional and linguistic-structural constraints ignores the
multi-dimensional nature of linguistic variation. As demonstrated in Part II of
this book, the choice process in discourse-pragmatics is rarely constrained by so-
cial factors alone, and changes in discourse-pragmatics regularly manifest them-
selves in changes to variants’ linguistic conditioning. It follows that studies which
ignore the potential operation of non-social factors may yield incomplete, mis-
leading or even inaccurate accounts of discourse-pragmatic variation. Moreover,
they may fail to detect the mechanisms and trajectories of ongoing changes in
discourse-pragmatics.
Successful resolution of the problems outlined at the beginning of this sec-
tion depends on methodological advancements. Some progress in this direction
has already been made. D’Arcy’s (2005, 2007) study of discourse like in Toronto
English has shown that the variable context for discourse-pragmatic variables can
be circumscribed according to structural criteria. This makes possible linguisti-
cally based explanations for apparent-time frequency increases in variables’ and/
or variants’ use. The present study has shown how multifunctionality can be incor-
porated as a parameter in the analysis to gauge reliable insights into the discourse-
functional distribution of discourse-pragmatic variables. Recent studies of general
extenders in contemporary varieties of British English have developed empirically
accountable methods for operationalising the processes associated with gram-
maticalisation as independent variables for quantitative analysis (Cheshire 2007;
Pichler & Levey 2011). Approaches such as these allow scholars to compare the
linguistic distribution of variants across social cells and trace the linguistic trajec-
tory of ongoing changes in apparent time. They thus yield important insights into
the role of discourse-pragmatic features in the linguistic system.

8.5 Exploring patterns of discourse-pragmatic change

The variationist framework enables scholars to identify and trace ongoing linguistic
changes by comparing patterns of synchronic variation across successive genera-
tions of speakers (see Chapters 2.3 and 4.5; Tagliamonte 2002a). Recent studies of
intensifiers and quotatives demonstrate that the Labovian framework is well-suited
to capturing the progression of rapidly changing discourse-pragmatic variables in
synchronic dialect data (e.g. Buchstaller & D’Arcy 2009; Ito & Tagliamonte 2003;
Tagliamonte 2005; Tagliamonte & D’Arcy 2004, 2007). Yet, as shown in Part II of
 Chapter 8.  Challenges for the future 233

this book and elsewhere (e.g. Pichler & Levey 2011; Raumolin-Brunberg & Nurmi
2001: 262), discourse-pragmatic change is not by definition quick and observable
in the shallow time-span covered by synchronic corpora. The changes associated
with grammaticalisation may take several decades, even centuries, to unfold and
may involve long periods of stable variation (Bybee 2009: 350, 353; Traugott &
Trousdale 2010: 28). A further complicating factor in the analysis of discourse-
pragmatic change is the fact that the internal conditioning of discourse-pragmatic
variation, which is key to observing grammaticalisation in cross-generational syn-
chronic data, is under-researched (see Section 8.4 above). Without easy access to
multiple stages in the diachronic development of discourse-pragmatic variables
and without detailed knowledge of their linguistic conditioning, the existence, let
alone the trajectory, of change cannot easily be inferred from synchronic data.
It comes as no surprise, then, that the body of variationist research into (ongo-
ing) discourse-pragmatic change is negligible compared to the body of research
into discourse-pragmatic variation, and that many questions surrounding the na-
ture and mechanisms of (ongoing) discourse-pragmatic change remain virtually
unexplored. Without access to a solid database of studies exploring diverse aspects
of discourse-pragmatic change, we may not be able to provide satisfactory expla-
nations of how synchronic structures of discourse-pragmatic variation emerge.
Information about the social embedding of discourse-pragmatic change can-
not be gleaned from extant diachronic studies of language variation and change.
Because they tend to assume that grammaticalisation changes are driven by lan-
guage-internal factors, scholars of grammaticalisation are not generally concerned
with examining the role of extra-linguistic factors when studying the diachronic
development of discourse-pragmatic features (see, however, Beeching 2005, 2007;
Bromhead 2009). Where extra-linguistic factors such as gender, socio-economic
standing or speaker roles and relationships are considered in the investigation of
diachronic data, these are generally correlated with usage patterns rather than
usage changes (e.g. Culpeper & Kytö 2010; Jucker 2002; Lutzky 2008). Historical
sociolinguistics, which is equipped to investigate the sociolinguistic embed-
ding of changes at earlier stages of the language, has largely focused on analys-
ing ­morpho-syntactic variables (see, however, Nevalainen 2008; Palander-Collin
1999). As a result, we lack comprehensive diachronic accounts of the social embed-
ding of discourse-pragmatic change which could serve as a comparative baseline
for synchronic variation studies.
Absent such data, discourse variationists may need to expand their synchronic
databases into the past and exploit the increasing availability of diachronic corpora
such as the Corpus of Early English Correspondence (Nevalainen & Raumolin-
Brunberg 1996), the Helsinki Corpus of English Texts (Kytö 1996), the Corpus of
English Dialogues (Kytö & Walker 2006), or the Corpus of Late Modern English
234 The Structure of Discourse-Pragmatic Variation

Texts (extended version) (De Smet 2005). These collections of written documents
have obvious limitations for variationist analysis as regards their representative-
ness and authenticity, and they present methodological challenges as regards the
reconstruction of social information and recovery of discourse functions (see
Hernández-Campoy & Schilling-Estes 2012; Montgomery 2007 for comprehen-
sive overviews of potential pitfalls). Nonetheless, if conducted with great care and
acute awareness of these limitations (what Labov [1994: 11] calls “the art of making
the best use of bad data”), variationist analyses of diachronic texts that represent
language close to speech (e.g. trials, letters, conversations in novels, plays) will
allow scholars to add historical depth to their synchronic analyses and probe the
sociolinguistic dimension of discourse-pragmatic changes more extensively. In
addition, the analysis of discourse-pragmatic change will benefit from the fol-
lowing: the construction of monitor corpora such as the Diachronic Corpus of
Tyneside English (Buchstaller 2011), which consists of three corpora collected in
three decades and at the time of writing spans five decades of speech recordings;
and the availability of corpora such as the Origins of New Zealand English Archive
(Gordon et al. 2007), a repository of longitudinal spoken language data that cov-
ers the history of New Zealand English from 1850 onwards. With their speakers
born roughly a century apart, these corpora provide a fruitful resource for tracking
slow or longer-term changes in discourse-pragmatics, as shown in recent work by
Barnfield & Buchstaller (2010), Buchstaller (2011) and D’Arcy (2012). Exploring
diachronic and real-time data has the added advantage that it allows scholars to
verify hypotheses about discourse-pragmatic change which were formulated on
the basis of synchronic analyses, as was the case in the present study.

8.6 Investigating the range of discourse-pragmatic variables

The gradual rise in the number of discourse variation studies produced in recent
decades has not been accompanied by a meaningful broadening of the research
focus to a wider and more diverse range of variables. Quotatives and intensifiers
(and increasingly general extenders) continue to be the variables which feature
most prominently on discourse variationists’ all too narrow research agendas. To
a great extent, the intense interest in quotatives and intensifiers is generated by
their capacity to undergo rapid change and radical innovation within relatively
short time-spans. The resultant co-existence of old and new usage patterns in
synchronic cross-generational data makes it possible to study grammaticalisation
in progress and investigate the often elusive actuation problem in language change
(see, in particular, Tagliamonte & D’Arcy 2007, 2009). Other discourse-pragmatic
 Chapter 8.  Challenges for the future 235

variables have received far less quantitative attention and have been studied in far
fewer varieties, not only because they may undergo change less quickly and are
therefore less suitable to probing the mechanisms underlying discourse-pragmatic
change in synchronic datasets (see Section 8.4 above) but also because they occur
with lower frequency (see Section 8.2 above) and are less straightforward than
quotatives and intensifiers to be conceptualised as discourse-pragmatic variables
(see Chapter 2.3.2).
Variationist analyses of quotatives and intensifiers have addressed important
questions about the nature and mechanisms of language variation and change.
Yet they cannot in and of themselves constitute the basis for a theory of dis-
course-pragmatic variation and change. As uncovered by the comparable analy-
ses in Part II of this book, the structure of discourse-pragmatic variation varies
across variables, even those that are compositionally very similar, and is at least
in part determined by variables’ and variants’ functional profiles. It is therefore
unlikely that the female lead reported inter alia by Ferrara & Bell (1995), Ito &
Tagliamonte (2003) and Tagliamonte & Hudson (1999) for changes in the quota-
tive and intensifier systems extends without exception to all discourse-pragmatic
variables, or that the morpho-syntactic expansion reported inter alia by Barnfield
& Buchstaller (2010) and Ito & Tagliamonte (2003) for changes in intensification
is a key diagnostic of increasing grammaticalisation across the board of variables
(see Chapter 6 for counter-evidence to this hypothesis). In other words, we must
not assume that findings produced by studies of quotatives and intensifiers will
generalise to the range of discourse-pragmatic variables. A much wider range
of functionally and syntactically diverse discourse-pragmatic features must be
investigated before robust generalisations about discourse-pragmatic variation
and change can be made.
A broadening of the field to include a more diverse range of discourse-
pragmatic­ features in the quantitative analysis of linguistic variation and change
is contingent on successful resolution of some of the problems highlighted in the
preceding sections. The increasing construction and availability of mega-corpora
and exploitation of web-based and media data will enable us to identify a more
diverse set of discourse-pragmatic features that are highly variable, rapidly inno-
vating or interesting in other ways. Also, easy access to diachronic and real-time
data will allow us to explore the trajectories of variables which change less rapidly
and rigorously than quotatives and intensifiers. The challenges we face are numer-
ous but not insurmountable.
236 The Structure of Discourse-Pragmatic Variation

8.7 Conclusion

This chapter has by no means exhausted the challenges scholars need to over-
come in order to advance the field of variationist discourse studies beyond its
current state. The fact remains that we still have a long way to go before we can
even contemplate formulating a general theory of discourse-pragmatic variation
and change. Our journey to theoretically grounded analyses and to a fuller un-
derstanding of the labyrinthine intricacies of discourse-pragmatic variation and
change may be less arduous if we collaborate more closely with scholars working
in other research paradigms. Terkourafi’s (2011) conceptualisation of higher-level
variables in terms of stable inference processes shows how developments in prag-
matics can provide a theoretical underpinning for variationist research tools (see
Chapter 2.3.2). This study (and others before it) has benefited from the insights
furnished by grammaticalisation studies in uncovering and explaining the struc-
ture of discourse-pragmatic variation. The current rudimentary state of discourse
variation analysis can be overcome if we commit ourselves to increased cross-
disciplinary dialogue.
 References

Adger, David & Jennifer Smith. 2005. “Variation and the minimalist programme”. Syntax and
Variation. Reconciling the Biological and the Social ed. by Leonie Cornips & Karen Corrigan,
149–179. Amsterdam: John Benjamins.
Aijmer, Karin. 1997. “I think – an English modal particle”. Modality in Germanic Languages.
Historical and Comparative Perspectives ed. by Toril Swan & Olaf Jansen Westvik, 1–47.
Berlin: Mouton de Gruyter.
Aijmer, Karin. 2002. English Discourse Particles. Evidence from a Corpus. Amsterdam: John
Benjamins.
Aijmer, Karin. 2009. “‘So er I just sort I dunno I think it’s just because …’: A corpus study of
I don’t know and I dunno in learners’ spoken English”. Corpora. Pragmatics and Discourse ed.
by Andreas H. Jucker, Schreier Daniel & Marianne Hundt, 151–168. Amsterdam: Rodopi.
Aijmer, Karin & Anne-Marie Simon-Vandenbergen. 2011. “Pragmatic markers”. Discursive Prag-
matics ed. by Jan Zienkowski, Jan-Ola Östman & Jef Verschueren, 223–247. Amsterdam:
John Benjamins.
Aitken, A. J. 1979. “Scottish speech: A historical view, with special reference to the Standard
English of Scotland”. Languages of Scotland ed. by A. J. Aitken & Tom McArthur, 85–118.
Edinburgh: Chambers.
Aitken, A. J. 1984. “Scots and English in Scotland”. Language in the British Isles ed. by Peter
Trudgill, 517–532. Cambridge: Cambridge University Press.
Akatsuka, Noriko. 1985. “Conditionals and the epistemic scale”. Language 61: 3.625–639.
Algeo, John. 1988. “The tag question in British English: It’s different, i’n’it?”. English World-Wide
9: 2.171–191.
Algeo, John. 1990. “It’s a myth, innit? Politeness and the English tag question”. The State of the
Language ed. by Christopher Ricks & Leonard Michaels, 443–450. London: Faber & Faber.
Andersen, Elaine S., Maquea Briznela, Beatrice Du Puy & Laura Gonnerman. 1999. “Cross-
linguistic­evidence for the early acquisition of discourse markers as register variables”. Jour-
nal of Pragmatics 31: 11.1339–1357.
Andersen, Gisle. 1998. Are tag questions questions? Evidence from spoken data. Paper presented
at the 19th ICAME Conference, May, Belfast.
Andersen, Gisle. 2001. Pragmatic Markers and Sociolinguistic Variation. A Relevance-Theoretic
Approach to the Language of Adolescents. Amsterdam: John Benjamins.
Anderwald, Lieselotte. 2002. Negation in Non-Standard British English. Gaps, Regularizations
and Asymmetries. London: Routledge.
Axelsson, Karin. 2011. Tag questions in fiction dialogue. PhD diss., University of Gothenburg.
Baayen, R. H. 2008. Analysing Linguistic Data. A Practical Introduction to Statistics Using R.
Cambridge: Cambridge University Press.
Bailey, Guy. 2002. “Real and apparent time”. The Handbook of Language Variation and Change
ed. by J. K. Chambers, Peter Trudgill & Natalie Schilling-Estes, 312–332. Oxford: Blackwell.
238 The Structure of Discourse-Pragmatic Variation

Bakht, Maryam M. 2010. Lexical variation and the negotiation of linguistic style in a Long Island
middle school. PhD diss., New York University.
Barbieri, Federica. 2008. “Patterns of age-based linguistic variation in American English”. Journal
of Sociolinguistics 12: 1.58–88.
Barnfield, Kate & Isabelle Buchstaller. 2010. “Intensifiers in Tyneside: Longitudinal develop-
ments and new trends”. English World-Wide 31: 3.252–287.
Barron, Anne. 2008. “Contrasting requests in Inner Circle Englishes: A study in variational
pragmatics”. Developing Contrastive Pragmatics. Interlanguage and Cross-Cultural Perspec-
tives ed. by Martin Pütz & JoAnne Neff-van Aertselaer, 355–402. Berlin: Mouton de Gruyter.
Baumgarten, Nicole & Juliane House. 2010. “I think and I don’t know in English as lingua franca
and native English discourse”. Journal of Pragmatics 42: 4.1184–1200.
Bayley, Robert. 2002. “The quantitative paradigm”. The Handbook of Language Variation and
Change ed. by J. K. Chambers, Peter Trudgill & Natalie Schilling-Estes, 117–141. Oxford:
Blackwell.
Bazzanella, Carla. 2006. “Discourse markers in Italian: Towards a ‘compositional’ meaning”.
Approaches to Discourse Particles ed. by Kerstin Fischer, 449–464. Amsterdam: Elsevier.
Beach, Wayne A. & Terri R. Metzger. 1997. “Claiming insufficient knowledge”. Human Com-
munication Research 23: 4.562–588.
Beal, Joan. 1993. “The grammar of Tyneside and Northumbrian English”. Real English. The Gram-
mar of British English Dialects ed. by James Milroy & Lesley Milroy, 187–213. London:
Longman.
Beal, Joan. 1997. “Syntax and morphology”. The Edinburgh History of the Scots Language ed. by
Charles Jones, 335–377. Edinburgh: Edinburgh University Press.
Beal, Joan. 2004. “English dialects in the North of England: Morphology and syntax”. A Hand-
book of Varieties of English. A Multimedia Reference Tool. Vol. 2: Morphology and Syntax ed.
by Bernd Kortmann, Kate Burridge, Rajend Mesthrie, Edgar W. Schneider & Clive Upton,
114–141. Berlin: Mouton de Gruyter.
Beal, Joan C. & Karen P. Corrigan. 2005. “No, nay, never: Negation in Tyneside English”. Aspects
of English Negation ed. by Yoko Iyeiri, 139–156. Amsterdam: John Benjamins.
Beeching, Kate. 2005. “Politeness-induced semantic change”. Language Variation and Change
17: 2.155–180.
Beeching, Kate. 2007. “A politeness-theoretic approach to pragmatico-semantic change”. Journal
of Historical Pragmatics 8: 1.69–108.
Bell, Allan & Gary Johnson. 1997. “Towards a sociolinguistics of style”. University of Pennsylvania
Working Papers in Linguistics 4: 1.1–21.
Berkley, Susan. 2002. “How to cure the ‘verbal virus’: A five-step treatment plan”, <http://www.
school-for-champions.com/speaking/verbalvirus.htm> (01 March 2011).
Biber, Douglas, Stig Johansson, Geoffrey Leech, Edward Finnegan & Susan Conrad. 1999. The
Longman Grammar of Spoken and Written English. London: Longman.
Bisang, Walter. 2004. “Grammaticalization without co-evolution of form and meaning: The
case of tense-aspect-modality in East and Mainland Southeast Asia”. What Makes Gram-
maticalization? A Look from Its Fringes and Components ed. by Walter Bisang, Nikolaus
Himmelmann & Bjoern Wiemer, 109–138. Berlin: Mouton de Gruyter.
Blakemore, Diane. 1987. Semantic Constraints on Relevance. Oxford: Blackwell.
Blakemore, Diane. 1988. “So as a constraint on relevance”. Mental Representations. The Interface
between Language and Reality ed. by Ruth M. Kempson, 183–195. Cambridge: Cambridge
University Press.
 References 239

Boye, Kasper & Peter Harder. 2007. “Complement-taking predicates: Usage and linguistic struc-
ture”. Studies in Language 31: 3.569–606.
Brinton, Laurel J. 1996. Pragmatic Markers in English. Grammaticalization and Discourse Func-
tion. Berlin: Mouton de Gruyter.
Brinton, Laurel J. 2001. “Historical discourse analysis”. The Handbook of Discourse Analysis ed.
by Deborah Schiffrin, Deborah Tannen & Heidi E. Hamilton, 138–160. Oxford: Blackwell.
Brinton, Laurel J. 2005. “Processes underlying the development of pragmatic markers: The case of
(I) say”. Opening Windows on Texts and Discourses of the Past ed. by Janne Skaffari, 279–299.
Amsterdam: John Benjamins.
Brinton, Laurel J. 2006. “Pathways in the development of pragmatic markers in English”. The
Handbook of the History of English ed. by Ans van Kemenade & Bettelou Los, 307–334.
London: Blackwell.
Brinton, Laurel J. 2007. “The development of I mean: Implications for the study of historical prag-
matics”. Methods in Historical Pragmatics ed. by Susan Fitzmaurice & Irma Taavitsainen,
37–79. Berlin: Mouton de Gruyter.
Brinton, Laurel J. 2008. The Comment Clause in English. Syntactic Origins and Pragmatic Devel-
opment. Cambridge: Cambridge University Press.
Britain, David. 2002. “Space and spatial diffusion”. The Handbook of Language Variation and
Change ed. by J. K. Chambers, Peter Trudgill & Natalie Schilling-Estes, 603–637. Oxford:
Blackwell.
Britain, David, ed. 2007. Language in the British Isles. Cambridge: Cambridge University Press.
Britton, Derek. 2002. Northern fronting and the north Lincolnshire merger of the reflexes of
ME /u:/ and ME /o:/. Language Sciences 24: 3-4.221–229.
Bromhead, Helen. 2009. The Reign of Truth and Faith. Epistemic Expressions in 16th and 17th
Century English. Berlin: Mouton de Gruyter.
Brown, Gillian. 1977. Listening to Spoken English. London: Longman.
Brown, Gillian & George Yule. 1983. Discourse Analysis. Cambridge: Cambridge University Press.
Brown, Keith & Martin Millar. 1980. “Auxiliary verbs in Edinburgh speech”. Transactions of the
Philological Society 78: 1.81–133.
Brown, Penelope & Stephen C. Levinson. 1987. Politeness. Some Universals in Language Usage.
Cambridge: Cambridge University Press.
Bublitz, Wolfram. 1992. “Transferred negation and modality”. Journal of Pragmatics
18: 5-6.551–577.
Buchstaller, Isabelle. 2004. The sociolinguistic constraints on the quotative system – British
English and US English compared. PhD diss., University of Edinburgh.
Buchstaller, Isabelle. 2006a. “Diagnostics of age-graded linguistic behaviour: The case of the
quotative system”. Journal of Sociolinguistics 10: 1.3–30.
Buchstaller, Isabelle. 2006b. “Social stereotypes, personality traits and regional perception
displaced: Attitudes towards the ‘new’ quotatives in the U.K.”. Journal of Sociolinguistics
10: 3.362–381.
Buchstaller, Isabelle. 2008. “The localization of global linguistic variants”. English World-Wide
29: 1.15–44.
Buchstaller, Isabelle. 2011. “Quotations across the generations: A multivariate analysis of speech
and thought introducers across 5 decades of Tyneside speech”. Corpus Linguistics and Lin-
guistic Theory 7: 1.59–92.
Buchstaller, Isabelle & Alexandra D’Arcy. 2009. “Localised globalization: A multi-local, multi-
variate investigation of quotative be like”. Journal of Sociolinguistics 13: 3.291–331.
240 The Structure of Discourse-Pragmatic Variation

Buchstaller, Isabelle, John R. Rickford, Elizabeth Closs Traugott, Thomas Wasow & Arnold
Zwicky. 2010. “The sociolinguistics of a short-lived innovation: Tracking the develop-
ment of quotative all across spoken and internet corpora”. Language Variation and Change
22: 2.191–219.
Burnard, Lou, ed. 2000. “The British National Corpus users reference guide”, http://www.natcorp.
ox.ac.uk/docs/userManual (21 January 2010).
Bybee, Joan. 2001. Phonology and Language Use. Cambridge: Cambridge University Press.
Bybee, Joan. 2003. “Mechanisms of change in grammaticization: The role of frequency”. The
Handbook of Historical Linguistics ed. by Brian D. Joseph & Richard D. Janda, 602–623.
London: Blackwell.
Bybee, Joan. 2006. “From usage to grammar: The mind’s response to repetition”. Language
82: 4.711–733.
Bybee, Joan. 2009. “Grammaticization: Implications for a theory of language”. Crosslinguistic
Approaches to the Psychology of Language. Research in the Tradition of Isaac Slobin ed. by
Jiansheng Guo, Eelena Lieven, Nanchy Budwig, Susan Ervin-Tripp, Keiko Nakamura &
Seyda Őzçaliskan, 345–355. Hove: Psychology Press.
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: Cambridge University Press.
Bybee, Joan. 2011. “Usage-based theory and grammaticalization”. The Oxford Handbook of Gram-
maticalization ed. by Heiko Narrog & Bernd Heine, 69–78. Oxford: Oxford University Press.
Bybee, Joan & Paul Hopper. 2001. “Introduction to frequency and the emergence of linguistic
structure”. Frequency and the Emergence of Linguistic Structure ed. by Joan Bybee & Paul
Hopper, 1–24. Amsterdam: John Benjamins.
Bybee, Joan & Joanne Scheibman. 1999. “The effect of usage on degrees of constituency: The
reduction of don’t in English”. Linguistics 37: 4.575–596.
Cameron, Deborah. 2001. Working with Spoken Discourse. London: Sage.
Cameron, Deborah. 2005. “Language, gender, and sexuality: Current issues and new directions”.
Applied Linguistics 26: 4.482–502.
Cameron, Deborah, Fiona McAlinden & Kathy O’Leary. 1988. “Lakoff in context: The social and
linguistic functions of tag questions”. Women in Their Speech Communities ed. by Jennifer
Coates & Deborah Cameron, 74–93. London: Longman.
Campbell, Lyle. 2001. “What’s wrong with grammaticalization?”. Language Sciences 23:
2-3.113–161.
Cattell, Ray. 1973. “Negative transportation and tag questions”. Language 49: 3.612–639.
Cheshire, Jenny. 1981. “Variation in the use of ain’t in an urban British English dialect”. Language
in Society 10: 2.365–381.
Cheshire, Jenny. 1982. Variation in an English Dialect. A Sociolinguistic Study. Cambridge: Cam-
bridge University Press.
Cheshire, Jenny. 1987. “Syntactic variation, the linguistic variable, and sociolinguistic theory”.
Linguistics 25: 2.257–282.
Cheshire, Jenny. 1997. “Involvement in ‘standard’ and ‘nonstandard’ English”. Taming the Ver-
nacular. From Dialect to Written Standard Language ed. by Jenny Cheshire & Dieter Stein,
68–82. Harlow: Longman.
Cheshire, Jenny. 1998. “English negation from an interactional perspective”. Negation in the
History of English ed. by Ingrid Tieken-Boon van Ostade, Gunnel Tottie & Wim van der
Wurff, 29–53. Berlin: Mouton de Gruyter.
Cheshire, Jenny. 2005. “Syntactic variation and beyond: Gender and social class variation in the
use of discourse-new markers”. Journal of Sociolinguistics 9: 4.479–508.
 References 241

Cheshire, Jenny. 2007. “Discourse variation, grammaticalisation and stuff like that”. Journal of
Sociolinguistics 11: 2.155–193.
Cheshire, Jenny, Paul Kerswill, Sue Fox & Eivind Torgersen. 2011. “Contact, the feature pool
and the speech community: The emergence of Multicultural London English”. Journal of
Sociolinguistics 15: 2.151–196.
Cheshire, Jenny, Paul Kerswill & Ann Williams. 2005. “Phonology, grammar, and discourse in
dialect convergence”. Dialect Change. Convergence and Divergence in European Languages
ed. by Peter Auer, Frans Hinskens & Paul Kerswill, 135–167. Cambridge: Cambridge Uni-
versity Press.
Coates, Jennifer. 1987. “Epistemic modality and spoken discourse”. Transactions of the Philologi-
cal Society 85: 1.110–131.
Coates, Jennifer. 1996. Women Talk. Conversations between Women Friends. Oxford: Blackwell.
Coates, Jennifer. 2004. Women, Men and Language. A Sociolinguistic Account of Gender Differ-
ences in Language. 3rd ed. London: Pearson.
Corbett, John, J. Derrick McClure & Jane Stuart-Smith. 2003. “A brief history of Scots”. The
­Edinburgh Companion to Scots ed. by John Corbett, J. Derrick McClure & Jane Stuart-Smith,
1–16. Edinburgh: Edinburgh University Press.
Couper-Kuhlen, Elizabeth & Margret Selting. 1996. “Towards an interactional perspective on
prosody and a prosodic perspective on interaction”. Prosody in Conversation ed. by Eliza-
beth Couper-Kuhlen & Margret Selting, 11–56. Cambridge: Cambridge University Press.
Coupland, Nikolas. 1983. “Patterns of encounter management: Further arguments for discourse
variables”. Language in Society 12: 3.459–476.
Craigie, William, Adam J. Aitken, James A. C. Stevenson, Harry D. Watson, Margaret G. Dareau
& K. Lorna Pike, eds. 1937–2002. A Dictionary of the Older Scottish Tongue. Aberdeen:
Aberdeen University Press.
Croft, William. 1990. Typology and Universals. Cambridge: Cambridge University Press.
Crosby, Faye & Linda Nyquist. 1977. “The female register: An empirical study of Lakoff ’s hy-
potheses”. Language in Society 6: 3.313–322.
Culpeper, Jonathan & Merja Kytö. 2010. Early Modern English Dialogues. Spoken Interaction as
Writing. Cambridge: Cambridge University Press.
Dailey-O’Cain, Jennifer. 2000. “The sociolinguistic distribution of and attitudes toward focuser
like and quotative like”. Journal of Sociolinguistics 4: 1.60–80.
D’Arcy, Alexandra. 2005. Like: Syntax and development. PhD diss., University of Toronto.
D’Arcy, Alexandra. 2007. “Like and language ideology: Disentangling fact from fiction”. American
Speech 82: 4.386–419.
D’Arcy, Alexandra. 2011. “Corpora: Capturing language in use”. Analysing Variation in English
ed. by Warren Maguire & April McMahon, 49–71. Cambridge: Cambridge University Press.
D’Arcy, Alexandra. 2012. “The diachrony of quotation: Evidence from New Zealand English”.
Language Variation and Change 24: 3.343–369.
Degand, Liesbeth & Anne-Marie Simon-Vandenbergen, eds. 2011. Grammaticalization, Prag-
maticalization and/or (Inter)Subjectification: Methodological Issues for the Study of Dis-
course Markers. Thematic issue: Linguistics 49: 2.
Dehé, Nicole & Anne Wichmann. 2010. “The multifunctionality of epistemic parentheticals
in discourse: Prosodic cues to the semantic-pragmatic boundary”. Functions of Language
17: 1.1–25.
Denis, Derek. 2011. “Innovators & innovation: Tracking the innovators of and stuff in York
English”. University of Pennsylvania Working Papers in Linguistics 17: 2.60–70.
242 The Structure of Discourse-Pragmatic Variation

Denis, Derek. MS. Grammaticalization of general extenders in York English. Unpublished manu-
script, University of Toronto.
De Smet, Henrik. 2005. “A corpus of Late Modern English texts”. International Computer Archive
of Modern and Medieval English 29.69–82.
Diani, Giuliana. 2004. “The discourse functions of I don’t know in English conversation”. Dis-
course Patterns in Spoken and Written Corpora ed. by Karin Aijmer & Anna-Brita ­Stenström,
157–171. Amsterdam: John Benjamins.
Diewald, Gabriele. 2006. “Discourse particles and modal particles as grammatical elements”.
Approaches to Discourse Particles ed. by Kerstin Fischer, 403–425. Amsterdam: Elsevier.
Dines, Elizabeth R. 1980. “Variation in discourse – ‘and stuff like that’”. Language in Society
9: 1.13–31.
Douglas, Fiona M. 2003. “Scottish Corpus of Text and Speech: Problems of corpus design”. Liter-
ary and Linguistic Computing 18: 1.23–37.
Douglas-Cowie, Ellen. 1978. “Linguistic code-switching in a Northern Irish village: Social inter-
action and social ambition”. Sociolinguistic Patterns in British English ed. by Peter Trudgill,
37–51. London: Edward Arnold.
Drager, Katie. 2011. “Sociophonetic variation and the lemma”. Journal of Phonetics 39: 4.694–707.
Drew, Paul. 1992. “Contested evidence in courtroom cross-examination: The case of a trial for
rape”. Talk at Work. Interaction in Institutional Settings ed. by Paul Drew & John Heritage,
470–520. Cambridge: Cambridge University Press.
Dubois, Betty Lou & Isabel Crouch. 1975. “The question of tag questions in women’s speech:
They don’t really use more of them, do they?”. Language in Society 4: 3.289–294.
Du Bois, John W., Stephan Schuetze-Coburn, Susanna Cumming & Danae Paolino. 1993. “Out-
line of discourse transcription”. Talking Data. Transcription and Coding in Discourse Re-
search ed. by Jane A. Edwards & Martin D. Lampert, 45–89. Hillsdale: Lawrence Erlbaum.
Eckardt, Regine. 2012. “Grammaticalization and semantic reanalysis”. Semantics. An Interna-
tional Handbook of Natural Language Meaning. Vol. 3. ed. by Claudia Maienborn, Klaus
von Heusinger & Paul Portner, 2675–2702. Berlin: Mouton de Gruyter.
Eckert, Penelope. 1997. “Age as a sociolinguistic variable”. The Handbook of Sociolinguistics ed.
by Florian Coulmas, 151–167. Oxford: Blackwell.
Eckert, Penelope & Sally McConnell-Ginet. 1992. “Think practically and look locally: Language
and gender as community-based practice”. Annual Review of Anthropology 21.461–490.
Edelsky, Carole. 1993. “Who’s got the floor?”. Gender and Conversational Interaction ed. by
­Deborah Tannen, 189–227. Oxford: Oxford University Press.
Erman, Britt. 1987. Pragmatic Expressions in English. A Study of you know, you see and I mean
in Face-to-Face Conversation. Stockholm: Minab/Gotab.
Erman, Britt. 1992. “Female and male usage of pragmatic expressions in same-gender and
mixed-gender interaction”. Language Variation and Change 4: 2.217–234.
Erman, Britt. 1998. “‘Just wear the wig innit!’ From identifying and proposition-oriented to
intensifying and speaker-oriented: Grammaticalization in progress”. Papers from the 16th
Scandinavian Conference of Linguistics ed. by Timo Hankioja, 87–100. University of Turku:
Department of Finnish and General Linguistics.
Erman, Britt. 2001. “Pragmatic markers revisited with a focus on you know in adult and adoles-
cent talk”. Journal of Pragmatics 32: 9.1337–1359.
Erman, Britt & Ulla-Britt Kotsinas. 1993. “Pragmaticalization: The case of ba’ and you know”.
Acta Universitatis Stockhomiensis, 76–93. Stockholm: Almqvist and Wiksell.
 References 243

Escalera, Elena Andrea. 2009. “Gender differences in children’s use of discourse markers: Sepa-
rate worlds or different contexts?”. Journal of Pragmatics 41: 12.2479–2495.
Ferrara, Kathleen W. 1997. “Form and function of the discourse marker anyway: Implications
for discourse analysis”. Linguistics 35: 2.343–378.
Ferrara, Kathleen W. & Barbara Bell. 1995. “Sociolinguistic variation and discourse function of
constructed dialogue introducers: The case of be + like”. American Speech 70: 3.265–289.
Fetzer, Anita. 2011. “‘I think this is I mean perhaps this is too erm too tough a view of the
world but I often think …’: Redundancy as a contextualization device”. Language Sciences
33: 2.255–267.
Fillmore, Charles J. 1963. “The position of embedding transformation in a grammar”. Word
19.208–231.
Fischer, Kerstin. 2000. “Discourse particles, turn-taking, and the semantics-pragmatics inter-
face”. Revue de Sémantique et Pragmatique 8.111–137.
Fischer, Kerstin. 2006. “Towards an understanding of the spectrum of approaches to discourse
particles: Introduction to the volume”. Approaches to Discourse Particles ed. by Kerstin
Fischer, 1–20. Amsterdam: Elsevier.
Fischer, Olga. 1998. “On negative raising in the history of English”. Negation in the History
of English ed. by Ingrid Tieken-Boon von Ostade, Gunnel Tottie & Wim van der Wurff,
55–100. Berlin: Mouton de Gruyter.
Fishman, Pamela. 1983. “Interaction: The work women do”. Language, Gender and Society ed. by
Barrie Thorne, Cheris Kramarae & Nancy Henley, 89–101. Cambridge: Newbury House.
Fleischman, Suzanne. 1999. Pragmatic markers in comparative and historical perspective: Theo-
retical implications of a case study. Paper presented at the 14th International Conference
on Historical Linguistics, August, Vancover.
Ford, Cecilia E. 2001. “At the intersection of turn and sequence: Negation and what comes
next”. Studies in Interactional Linguistics ed. by Margret Selting, 51–79. Amsterdam: John
Benjamins.
Ford, Cecilia E. & Sandra A. Thompson. 1996. “Interactional units in conversation: Syntactic, in-
tonational, and pragmatic resources for the management of turns”. Interaction and Grammar
ed. by Elinor Ochs, Emanuel A. Schegloff & Sandra A. Thompson, 134–184. Cambridge:
Cambridge University Press.
Foulkes, Paul & Gerard J. Docherty. 1999. “Urban voices – overview”. Urban Voices. Accent
Studies in the British Isles ed. by Paul Foulkes & Gerard J. Docherty, 1–24. London: Arnold.
Fox, Barbara A., Makoto Hayashi & Robert Jasperson. 1996. “Resources and repair: A cross-
linguistic study of syntax and repair”. Interaction and Grammar ed. by Elinor Ochs, Emanuel
A. Schegloff & Sandra A. Thompson, 185–237. Cambridge: Cambridge University Press.
Fox Tree, Jean E. 2007. “Folk notions of um and uh, you know, and like”. Text & Talk 27: 3.297–314.
Fox Tree, Jean E. 2010. “Discourse markers across speakers and settings”. Language and Linguis-
tics Compass 4: 5.269–281.
Frank-Job, Barbara. 2006. “A dynamic-interactional approach to discourse markers”. Approaches
to Discourse Particles ed. by Kerstin Fischer, 359–374. Amsterdam: Elsevier.
Fraser, Bruce. 1980. “Conversational mitigation”. Journal of Pragmatics 4: 4.341–350.
Fraser, Bruce. 1990. “An approach to discourse markers”. Journal of Pragmatics 14: 3.383–395.
Fraser, Bruce. 1999. “What are discourse markers?”. Journal of Pragmatics 31: 7.931–952.
Fuller, Janet M. 2001. “The principle of pragmatic detachability in borrowing: English-origin
discourse markers in Pennsylvania German”. Linguistics 39: 2.351–369.
244 The Structure of Discourse-Pragmatic Variation

Fuller, Janet M. 2003. “The influence of speaker roles on discourse marker use”. Journal of Prag-
matics 35: 1.23–45.
Furman, Reyhan & Asli Özyürek. 2007. “Development of interactional discourse mark-
ers: Insights from Turkish children’s and adults’ oral narratives”. Journal of Pragmatics
39: 10.1742–1757.
Geluykens, Ronald. 1993. “Topic introduction in English conversation”. Transactions of the Philo-
logical Society 91: 2.181–214.
Gisborne, Nikolas & Graeme Trousdale. 2008. “Constructional approaches to language-particular­
description”. Constructional Approaches to English Grammar ed. by Graeme Trousdale &
Nikolas Gisborne, 1–4. Berlin: Mouton de Gruyter.
Givón, Talmy. 1993. English Grammar. A Function-Based Introduction. Amsterdam: John
Benjamins.
Glauser, Beat. 1974. The Scottish-English Linguistic Border. Lexical Aspects. Bern: Francke.
Goffman, Erving. 1967. Interaction Ritual. Essays on Face to Face Behavior. Garden City, New
York: Doubleday.
Goldberg, Adele E. 1995. A Construction Grammar Approach to Argument Structure. Chicago:
University of Chicago Press.
Goldberg, Julia Anna. 1980. Discourse particles: An analysis of the role of y’know, I mean, well
and actually in conversation. PhD diss., University of Cambridge.
Gordon, Elizabeth, Jennifer Hay & Margaret Maclagan. 2007. “The ONZE corpus”. Creating and
Digitizing Language Corpora ed. by Joan C. Beal, Karen P. Corrigan & Herman L. Moisl,
82–104. New York: Palgrave Macmillan.
Goss, Emily L. & Joseph C. Salmons. 2000. “The evolution of bilingual discourse marking: Modal
particles and English markers in 19th century German-American dialects”. International
Journal of Bilingualism 4: 4.469–494.
Grant, Lynn E. 2010. “A corpus comparison of the use of I don’t know by British and New Zealand
speakers”. Journal of Pragmatics 42: 9.2282–2296.
Grant, William & James M. Dixon. 1921. Manual of Modern Scots. Cambridge: Cambridge
University Press.
Grant, William & David D. Murison. 1931. The Scottish National Dictionary. Aberdeen: Aberdeen
University Press.
Gregory, Michelle, William D. Raymond, Alan Bell, E. Fosler-Lussier & Daniel Jurafsky. 1999.
“The effects of collocational strength and contextual predictability in lexical production”.
Chicago Linguistic Society 35.151–166.
Greenbaum, Sidney. 1992. “A new corpus of English: ICE”. Directions in Corpus Linguistics. Pro-
ceedings of Nobel Symposium 82, Stockholm 4–8 August 1991 ed. by Jan Svartvik, 171–183.
Berlin: Mouton de Gruyter.
Griffiths, Bill. 2005. A Dictionary of North-East Dialect. 2nd ed. Newcastle: Northumbria Uni-
versity Press.
Günthner, Susanne. 2000. “From concessive connector to discourse marker: The use of obwohl
in everyday German interaction”. Cause, Condition, Concession, Contrast ed. by Elizabeth
Couper-Kuhlen & Bernd Kortmann, 439–468. Berlin: Mouton de Gruyter.
Guy, Gregory R. 1980. “Variation in the group and the individual: The case of final stop deletion”.
Locating Language in Time and Space ed. by William Labov, 1–36. New York: Academic Press.
Guy, Gregory G. 1993. “The quantitative analysis of linguistic variation”. American Dialect Re-
search ed. by Dennis Preston, 223–249. Amsterdam: John Benjamins.
 References 245

Halliday, M. A. K. 1979. “Modes of meaning and modes of expression: Types of grammatical
structures and their determination by different semantic functions”. Function and Context
in Linguistic Analysis. A Festschrift for William Haas ed. by D. J. Allerton, Edward Carney
& David Holdcroft, 57–79. Cambridge: Cambridge University Press.
Halliday, M. A. K. 1985. Introduction to Functional Grammar. London: Arnold.
Halliday, M. A. K. & Ruqaiya Hasan. 1976. Cohesion in English. London: Longman.
Hansen, Maj-Britt Mosegaard. 2006. “A dynamic polysemy approach to the lexical semantics
of discourse markers (with an exemplary analysis of toujours)”. Approaches to Discourse
Particles ed. by Kerstin Fischer, 21–41. Amsterdam: Elsevier.
Harris, Sandra. 1984. “Questions as a mode of control in the magistrates’ courts”. International
Journal of the Sociology of Language 49: 1.5–27.
Haspelmath, Martin. 2004. “On directionality in language change with a particular reference to
grammaticalization”. Up and Down the Cline. The Nature of Grammaticalization ed. by Olga
Fischer, Muriel Norde & Harry Perridon, 17–44. Amsterdam: John Benjamins.
Hazen, Kirk. 2000. “The role of researcher identity in conducting sociolinguistic research: A re-
flective case study”. Southern Journal of Linguistics 24.103–120.
He, Agnes Weiyun. 1993. “Exploring modality in institutional interaction: Cases from academic
counselling encounters”. Text 13: 4.503–528.
Heine, Bernd. 1993. Auxiliaries. Cognitive Forces and Grammaticalization. Oxford: Oxford Uni-
versity Press.
Heine, Bernd. 2003. “Grammaticalization”. The Handbook of Historical Linguistics ed. by Brian
D. Joseph & Richard D. Janda, 575–601. London: Blackwell.
Heinemann, Trine. 2005. “Where grammar and interaction meet: The preference for matched
polarity in responsive turns in Danish”. Syntax and Lexis in Conversation. Studies on the
Use of Linguistic Resources in Talk-in-Interaction ed. by Auli Hakulinen & Margret Selting,
375–402. Amsterdam: John Benjamins.
Heritage, John. 1984. Garfinkel and Ethnomethodology. Cambridge: Polity Press.
Heritage, John & J. Maxwell Atkinson. 1984. “Introduction”. Structures of Social Action. Studies
in Conversation Analysis ed. by J. Maxwell Atkinson & John Heritage, 1–15. Cambridge:
Cambridge University Press.
Hernández-Campoy, Juan Manuel & Natalie Schilling-Estes. 2012. “The application of the quan-
titative paradigm to historical sociolinguistics: Problems with the generalisability princi-
ple”. The Handbook of Historical Sociolinguistics ed. by Juan Manuel Hernández-Campoy &
J. Camilo Conde-Silvestre, 63–79. Oxford: Wiley-Blackwell.
Hewitt, Roger. 1986. White Talk Black Talk. Inter-Racial Friendship and Communication amongst
Adolescents. Cambridge: Cambridge University Press.
Himmelmann, Nikolaus P. 2004. “Lexicalization and grammaticization: Opposite or orthogo-
nal?” What Makes Grammaticalization? A Look from Its Fringes and Its Components ed.
by Walter Bisang, Nikolaus P. Himmelmann & Björn Wiener, 21–42. Berlin: Mouton de
Gruyter.
Hirschberg, Julia & Diane Litman. 1993. “Empirical studies on the disambiguation of cue
phrases”. Computational Linguistics 19: 3.501–531.
Hllavac, Jim. 2006. “Bilingual discourse markers: Evidence from Croatian-English code-
switching­”. Journal of Pragmatics 38: 11.1870–1900.
Hoffmann, Sebastian. 2006. “Tag questions in Early and Late Modern English: Historical descrip-
tion and theoretical implications”. Anglistik 17: 2.35–55.
246 The Structure of Discourse-Pragmatic Variation

Hoffmann, Sebastian, Anne-Katrin Blass & Joybrato Mukherjee. forthc. “Canonical tag questions
in Asian Englishes: Forms, functions and frequencies in Hong Kong English, Indian English
and Singapore English”. The Oxford Handbook of World Englishes ed. by Markku Filppula,
Juhani Klemola & Devyani Sharma. Oxford: Oxford University Press.
Holmes, Janet. 1982. “The functions of tag questions”. English Language Research Journal 4.40–65.
Holmes, Janet. 1984a. “Hedging your bets and sitting on the fence: Some evidence for hedges as
support structures”. Te Reo 27.47–62.
Holmes, Janet. 1984b. “Modifying illocutionary force”. Journal of Pragmatics 8: 3.345–365.
Holmes, Janet. 1986. “Functions of you know in women’s and men’s speech”. Language in Society
15: 1.1–22.
Holmes, Janet. 1987. “Hedging, fencing and other conversational gambits: An analysis of gender
differences in New Zealand speech”. Women and Language in Australian and New Zealand
Society ed. by Anne Pauwels, 47–62. Sydney: Australian Professional Publications.
Holmes, Janet. 1990. “Hedges and boosters in women’s and men’s speech”. Language and Com-
munication 10: 3.185–205.
Holmes, Janet. 1995. Women, Men and Politeness. London: Longman.
Hooper, Joan B. 1975. “On assertive predicates”. Syntax and Semantics ed. by John P. Kimball,
91–124. London: Academic Press.
Hopper, Paul J. 1991. “On some principles of grammaticization”. Approaches to Grammaticaliza-
tion. Vol. 1: Focus on Theoretical and Methodological Issues ed. by Elizabeth Closs Traugott
& Bernd Heine, 17–35. Amsterdam: John Benjamins.
Hopper, Paul J. & Elizabeth Closs Traugott. 2003. Grammaticalization. 2nd ed. Cambridge:
Cambridge University Press.
Horn, Laurence R. 1978. “Remarks on neg-raising”. Syntax and Semantics. Pragmatics ed. by
Peter Cole, 129–220. London: Academic Press.
Horn, Laurence R. 2001. A Natural History of Negation. Stanford: CSLI Publications.
Huddleston, Rodney. 2002. “The verb”. The Cambridge Grammar of the English Language ed. by
Rodney Huddleston & Geoffrey K. Pullum, 71–212. Cambridge: Cambridge University Press.
Hudson, Richard A. 1975. “The meaning of questions”. Language 51: 1.1–31.
Hughes, Arthur, Peter Trudgill & Dominic Watt. 2012. English Accents and Dialects. An Introduc-
tion to the Social and Regional Varieties of English in the British Isles. 5th ed. London: Arnold.
Hutchby, Ian. 2002. “Resisting the incitement to talk in child counselling: Aspects of the utter-
ance ‘I don’t know’”. Discourse Studies 4: 2.147–168.
Hutchby, Ian & Robin Wooffitt. 1998. Conversation Analysis. Principles, Practices and Applica-
tions. Cambridge: Polity.
Ito, Rika & Sali Tagliamonte. 2003. “Well weird, right dodgy, very strange, really cool: Layering
and recycling in English intensifiers”. Language in Society 32: 2.257–279.
Janda, Richard. 2001. “Beyond ‘pathways’ and ‘unidirectionality’: On the discontinuity of
language transmission and the counterability of grammaticalization”. Language Sciences
23: 2-3.265–340.
Jefferson, Gail. 2002. “Is ‘no’ an acknowledgment token? Comparing American and British uses
of (+)/(−) tokens”. Journal of Pragmatics 34: 10–11.1345–1383.
Jespersen, Otto. 1917. Negation in English and Other Languages. Copenhagen: Bianco Lunos
Bogtrykkeri.
Johnson, Daniel Ezra. 2009. “Getting off the GoldVarb standard: Introducing Rbrul for mixed-
effects variable rule analysis”. Language and Linguistics Compass 3: 1.359–383.
 References 247

Johnstone, Barbara. 1990. “Variation in discourse: Midwestern narrative style”. American Speech
65: 3.195–214.
Jucker, Andreas H. 1993. “The discourse marker well: A relevance-theoretical account”. Journal
of Pragmatics 19: 5.435–452.
Jucker, Andreas H. 2002. “Discourse markers in Early Modern English”. Alternative Histories of
English ed. by Peter Trudgill & Richard J. Watts, 210–230. London: Routledge.
Jucker, Andreas H. & Sara W. Smith. 1998. “And people just you know like ‘wow’: Discourse
markers as negotiating strategies”. Discourse Markers. Descriptions and Theory ed. by
­Andreas H. Jucker & Yael Ziv, 171–201. Amsterdam: John Benjamin.
Jucker, Andreas H. & Irma Taavitsainen. 2012. “Pragmatic variables”. The Handbook of Histori-
cal Sociolinguistics ed. by Juan Manuel Hernández-Campoy & J. Camilo Condes-Silvestre,
293–306. Oxford: Wiley-Blackwell
Jucker, Andreas H. & Yael Ziv. 1998. “Discourse markers: Introduction”. Discourse Markers.
Descriptions and Theory ed. by Andreas H. Jucker & Yael Ziv, 1–12. Amsterdam: John
Benjamins.
Jurafsky, Daniel, Alan Bell, Michelle Gregory & William D. Raymond. 2001. “Probabilistic rela-
tions between words: Evidence from reduction in lexical production”. Frequency and the
Emergence of Linguistic Structure ed. by Joan Bybee & Paul Hopper, 229–254. Amsterdam:
John Benjamins.
Kaltenböck, Gunther. 2005. “Charting boundaries of syntax: A taxonomy of spoken parenthetical
clauses”. Vienna English Working Papers 14: 1.21–53.
Kaltenböck, Gunther, Bernd Heine & Tania Kuteva. 2011. “On thetical grammar”. Studies in
Language 35: 4.852–897.
Kärkkäinen, Elise. 2003. Epistemic Stance in English Conversation. A Description of Its Interac-
tional Functions with a Focus on I think. Amsterdam: John Benjamins.
Kärkkäinen, Elise. 2007. “The role of I guess in conversational stancetaking”. Stancetaking in
Discourse ed. by Robert Englebretson, 183–219. Amsterdam: John Benjamins.
Kay, Paul & Charles J. Fillmore. 1999. “Grammatical constructions and linguistic generalizations:
The What’s X doing Y construction”. Language 75: 1.1–33.
Kearns, Kate. 2007. “Epistemic verbs and zero complementizer”. English Language and Linguistics
11: 3.475–505.
Keevallik, Leelo. 2006. “From discourse pattern to epistemic marker: Estonian (ei) tea ‘don’t
know’”. Nordic Journal of Linguistics 29: 2.173–200.
Kerswill, Paul. 2003. “Dialect levelling and geographical diffusion in British English”. Social
Dialectology. In Honour of Peter Trudgill ed. by David Britain & Jenny Cheshire, 223–243.
Amsterdam: John Benjamins.
Kerswill, Paul, Jenny Cheshire, Sue Fox & Eivind Torgersen. 2007. Linguistic Innovators: The
English of Adolescents in London: Full Research Report. ESRC End of Award Report,
RES-000-23-0680. Swindon: ESRC.
Kerswill, Paul, Jenny Cheshire, Sue Fox & Eivind Torgersen. 2011. Multicultural London English:
The Emergence, Acquisition and Diffusion of a New Variety. ESRC End of Award Report,
RES-062-23-0814. Swindon: ESRC.
Kimps, Ditte. 2007. “Declarative constant polarity tag questions: A data-driven analysis of their
form, meaning and attitudinal use”. Journal of Pragmatics 39: 2.270–291.
Kjellmer, Göran. 2003. “Hesitation. In defence of er and erm”. English Studies 2: 2.170–198.
248 The Structure of Discourse-Pragmatic Variation

Kortmann, Bernd, Kate Burridge, Rajend Mesthrie, Edgar W. Schneider & Clive Upton, eds.
2004. A Handbook of Varieties of English. A Multimedia Reference Tool. Vol. 2: Morphology
and Syntax. Berlin: Mouton de Gruyter.
Kotthoff, Helga. 1993. “Disagreement and concession in disputes: On the context sensitivity of
preference structures”. Language in Society 22: 2.193–216.
Kristiansen, Tore. 2011. “Attitudes, ideology and awareness”. The SAGE Handbook of Sociolin-
guistics ed. by Ruth Wodak, Barbara Johnstone & Paul Kerswill, 265–278. Los Angeles: Sage.
Krug, Manfred. 1998. “British English is developing a new discourse marker, innit? A study in
lexicalisation based on social, regional and stylistic variation”. Arbeiten aus Anglistik und
Amerikanistik 23: 2.146–197.
Kytö, Merja. 1996. Manual to the Diachronic Part of the Helsinki Corpus of English Texts. Coding
Conventions and Lists of Sources. University of Helsinki: English Department.
Kytö, Merja & Terry Walker. 2006. Guide to A Corpus of English Dialogues 1560–1760 (Studia
Anglistica Upsaliensia 130). Uppsala: Acta Universitatis Upsaliensis.
Labov, William. 1963. “The social motivation of a sound change”. Word 19.273–309.
Labov, William. 1966. The Social Stratification of English in New York City. Washington, DC:
Center for Applied Linguistics.
Labov, William. 1972. Sociolinguistic Patterns. Oxford: Blackwell.
Labov, William. 1982. “Building on empirical foundations”. Perspectives on Historical Linguistics
ed. by Winfred P. Lehmann & Yakov Malkiel, 17–92. Amsterdam: John Benjamins.
Labov, William. 1994. Principles of Linguistic Change. Vol. 1: Internal Factors. Oxford: Blackwell.
Labov, William. 1998. “The intersection of sex and social class in the course of linguistic change”.
The Sociolinguistics Reader. Vol. 2: Gender and Discourse ed. by Jenny Cheshire & Peter
Trudgill, 7–52. Oxford: Oxford University Press.
Labov, William. 2001. Principles of Linguistic Change. Vol. 2: Social Factors. Oxford: Blackwell.
Lakoff, Robin. 1973. “Language and woman’s place”. Language in Society 2: 1.45–81.
Lam, Phoenix W. Y. 2009. “The effect of text type on the use of so as a discourse particle”. Dis-
course Studies 11: 3.353–372.
Lavandera, Beatriz R. 1978. “Where does the sociolinguistic variable stop?”. Language in Society
7: 2.171–182.
Leech, Geoffrey. 1983. Principles of Pragmatics. London: Longman.
Lehmann, Christian. 1995 [1982]. Thoughts on Grammaticalization. Munich: Lincom Europa.
Lenk, Ute. 1998. Marking Discourse Coherence. Functions of Discourse Markers in Spoken English.
Tübingen: Gunter Narr.
Levey, Stephen. 2003. “He’s like ‘Do it now!’ and I’m like ‘No!’”. English Today 19: 1.24–32.
Levey, Stephen. 2006. “The sociolinguistic distribution of discourse marker like in preadolescent
speech”. Multilingua 25: 4.413–441.
Levey, Stephen. 2012. “General extenders and grammaticalization: Insights from London pre-
adolescents”. Applied Linguistics 33: 3.257–281.
Levey, Stephen, Karine Groulx & Joseph Roy. forthc. “A variationist perspective on discourse-
pragmatic change in a contact setting”. Language Variation and Change.
Levinson, Stephen C. 1988. “Conceptual problems in the study of regional and cultural style”.
The Sociolinguistics of Urban Vernaculars ed. by Norbert Dittmar & P. Schlobinski, 161–190.
Berlin: Mouton de Gruyter.
Liao, Silvie. 2009. “Variation in the use of discourse markers by Chinese teaching assistants in
the US”. Journal of Pragmatics 41: 7.1313–1328.
 References 249

Lindemann, Stephanie & Anna Mauranen. 2001. “‘It’s just real messy’: The occurrence and func-
tion of just in a corpus of academic speech”. English for Specific Purposes 20.459–475.
Llamas, Carmen. 2001. Language variation and innovation in Teesside English. PhD diss., Uni-
versity of Leeds.
Llamas, Carmen. 2007a. “A new methodology: Data elicitation for regional and social language
variation studies”. York Papers in Linguistics 8.138–163.
Llamas, Carmen. 2007b. “‘A place between places’: Language and identities in a border town”.
Language in Society 36: 4.579–604.
Llamas, Carmen, Dominic Watt & Daniel Ezra Johnson. 2009. “Linguistic accommodation and
the salience of national identity markers in a border town”. Journal of Language and Social
Psychology 28: 4.381–407.
Lutzky, Ursula. 2008. “The discourse marker marry – a sociopragmatic analysis”. Vienna English
Working Papers 17: 2.3–20.
Macafee, Caroline. 1992. “Characteristics of non-standard grammar in Scotland”. <http://www.
abdn.ac.uk/~enl038/grammar.htm> (08 September 2007).
Macaulay, Ronald K. S. 1991. Locating Dialect in Discourse. The Language of Honest Men and
Bonnie Lasses in Ayr. Oxford: Oxford University Press.
Macaulay, Ronald K. S. 1995. “The adverbs of authority”. English World-Wide 16: 1.37–60.
Macaulay, Ronald. 2001. “You’re like ‘why not?’: The quotative expressions of Glasgow adoles-
cents”. Journal of Sociolinguistics 5: 1.3–21.
Macaulay, Ronald. 2002a. “Discourse variation”. The Handbook of Language Variation and Change
ed. by J. K. Chambers, Peter Trudgill & Natalie Schilling-Estes, 283–305. London: Blackwell.
Macaulay, Ronald. 2002b. “Extremely interesting, very interesting, or only quite interesting?
Adverbs and social class”. Journal of Sociolinguistics 6: 3.398–417.
Macaulay, Ronald. 2002c. “You know, it depends”. Journal of Pragmatics 34: 6.749–767.
Macaulay, Ronald K. S. 2005. Talk That Counts. Age, Gender, and Social Class Differences in
Discourse. Oxford: Oxford University Press.
Macaulay, Ronald. 2006. “Pure grammaticalization: The development of a teenage intensifier”.
Language Variation and Change 18: 3.267–283.
Macaulay, Ronald K. S. 2009. “Adolescents and identity”. Intercultural Pragmatics 6: 4.597–612.
McGregor, William. 1995. “The English ‘tag question’: A new analysis, is(n’t) it?”. On Subject and
Theme. A Discourse Functional Perspective ed. by Ruqaiya Hasan, 91–121. Amsterdam: John
Benjamins.
Meehan, Teresa. 1991. “It’s like, ‘What’s happening in the evolution of like?’: A theory of gram-
maticalization”. Kansas Working Papers in Linguistics 16.37–51.
Meillet, Antoine. 1912. “L’évolution des formes grammaticales”. Scientia 26: 6.130–148.
Meyerhoff, Miriam. 1994. “Sounds pretty ethnic, eh?: A pragmatic particle in New Zealand
English”. Language in Society 23: 3.367–388.
Millar, Martin & Keith Brown. 1979. “Tag questions in Edinburgh speech”. Linguistische Berichte
60.24–45.
Millar, Robert McColl. 2007. Northern and Insular Scots. Edinburgh: Edinburgh University Press.
Miller, Jim. 1993. “The grammar of Scottish English”. Real English. The Grammar of English
Dialects in the British Isles ed. by James Milroy & Lesley Milroy, 99–138. London: Longman.
Miller, Jim. 2003. “Syntax and discourse in Modern Scots”. The Edinburgh Companion to Scots ed.
by John Corbett, J. Derrick McClure & Jane Stuart-Smith, 72–109. Edinburgh: Edinburgh
University Press.
250 The Structure of Discourse-Pragmatic Variation

Miller, Jim. 2004. “Scottish English: Morphology and syntax”. A Handbook of Varieties of English.
A Multimedia Reference Tool. Vol. 2: Morphology and Syntax ed. by Bernd Kortmann, Kate
Burridge, Rajend Mesthrie, Edgar W. Schneider, & Clive Upton, 47–72. Berlin: Mouton
de Gruyter.
Miller, Jim & Keith Brown. 1982. “Aspects of Scottish English syntax”. English World-Wide
3: 1.3–17.
Miller, Jim & Regina Weinert. 1995. “The function of LIKE in dialogue”. Journal of Pragmatics
23: 4.365–393.
Milroy, James & Lesley Milroy. 1997. “Varieties and variation”. The Handbook of Sociolinguistics
ed. by Florian Coulmas, 47–64. Oxford: Blackwell.
Milroy, James & Lesley Milroy. 1998. “Mechanisms of change in urban dialects: The role of class,
social network and gender”. The Sociolinguistics Reader. Vol. 1: Variation and Multilingual-
ism ed. by Peter Trudgill & Jenny Cheshire, 179–195. London: Arnold.
Milroy, James, Lesley Milroy & Sue Hartley. 1994. “Local and supra-local change in British
English: The case of glottalisation”. English World-Wide 15: 1.1–33.
Montgomery, Michael. 2007. “Variation and historical linguistics”. Sociolinguistic Variation. The-
ories, Methods, and Applications ed. by Robert Bayley & Ceil Lucas, 110–132. Cambridge:
Cambridge University Press.
Moore, Emma & Robert Podesva. 2009. “Style, indexicality, and the social meaning of tag ques-
tions”. Language in Society 38: 4.447–485.
Müller, Simone. 2005. Discourse Markers in Native and Non-Native English Discourse. Amster-
dam: John Benjamins.
Murray, James A. H. 1873. The Dialect of the Southern Counties of Scotland. Its Pronunciation,
Grammar, and Historical Relations. London: Philological Society.
Narrog, Heiko. 2012. “Beyond intersubjectification: Textual uses of modality and mood in sub-
ordinate clauses as part of speech-act orientation”. English Text Construction 5: 1.29–52.
Nevalainen, Terttu. 2008. “Social variation in intensifier use: Constraint on -ly adverbialization
in the past?”. English Language and Linguistics 12: 2.289–315.
Nevalainen, Terttu & Helena Raumolin-Brunberg, eds. 1996. Sociolinguistics and Language His-
tory. Studies Based on the Corpus of Early English Correspondence. Amsterdam: Rodopi.
Newmeyer, Frederick J. 1998. Language Form and Language Function. Cambridge, MA: MIT
Press.
Norrick, Neal R. 1995. “Hunh-tags and evidentiality in conversation”. Journal of Pragmatics
23: 6.687–692.
Nuyts, Jan. 1990. “Negative-raising reconsidered: Arguments for a cognitive-pragmatic ap-
proach”. Journal of Pragmatics 14: 4.559–588.
Orton, H. & W. J. Halliday. 1963. Survey of English Dialects. Leeds: E. J. Arnold.
Östman, Jan-Ola. 1981. You know. A Discourse Functional Approach. Amsterdam: John
Benjamins.
Oxford English Dictionary. 2010. 3rd ed. online. Oxford: Oxford University Press. <http://www.
oed.com>
Palander-Collin, Minna. 1999. Grammaticalization and Social Embedding. I THINK and
METHINKS in Middle and Early Modern English. Helsinki: Société Néophilologique.
Parton, Sabrena R., Susan A. Siltanen, Lawrence A. Hosman & Jeff Langenderfer. 2002. “Em-
ployment interview outcomes and speech style effects”. Journal of Language and Social
Psychology 21: 2.144–161.
 References 251

Peitsara, Kirist & Anna-Liisa Vasko. 2002. “The Helsinki Dialect Corpus: Characteristics of
speech and aspects of variation”. Helsinki English Studies 2. http://blogs.helsinki.fi/hes-eng/
files/2011/03/HES_Vol2_Peitsara_Vasko.pdf (24 January 2013).
Persson, Gunnar. 1993. “Think in panchronic perspective”. Studia Neophilologica 65: 1.3–18.
Petersen, Andrew. 2004. “The campaign against ‘like’ ”. The Wall Street Journal 3 February.
Petyt, K. M. 1978. “Secondary contractions in West Yorkshire negatives”. Sociolinguistic Patterns
in British English ed. by Peter Trudgill, 91–100. London: Edward Arnold.
Pichler, Heike. 2009. “The functional and social reality of discourse variants in a northern English
dialect: i don’t know and i don’t think compared”. Intercultural Pragmatics 6: 4.561–596.
Pichler, Heike. 2010. “Methods in discourse variation analysis: Reflections on the way forward”.
Journal of Sociolinguistics 14: 5.581–608.
Pichler, Heike & Stephen Levey. 2011. “In search of grammaticalization in synchronic dialect data:
General extenders in north-east England”. English Language and Linguistics 15: 3.441–471.
Pichler, Heike & Eivind Torgersen. 2012. Tag questions in contemporary London English:
Current trends in invariabilisation. Paper presented at the 33rd ICAME conference, June,
Leuven.
Plug, Leendert. 2010. “Pragmatic constraints in usage-based phonology, with reference to some
Dutch phrases”. Journal of Pragmatics 42: 7.2014–2035.
Pomerantz, Anita. 1984. “Agreeing and disagreeing with assessments: Some features of preferred/
dispreferred turn shapes”. Structures of Social Action. Studies in Conversation Analysis ed.
by J. Maxwell Atkinson & John Heritage, 57–101. Cambridge: Cambridge University Press.
Pope, Jennifer, Miriam Meyerhoff & D. Robert Ladd. 2007. “Forty years of language change on
Martha’s Vineyard”. Language 83: 3.615–627.
Poplack, Shana. 1989. “The care and handling of a mega-corpus: The Ottawa-Hull French
Project”. Language Change and Variation ed. by Ralph W. Fasold & Deborah Schiffrin,
411–451. Amsterdam: John Benjamins.
Poplack, Shana. 2011. “A variationist perspective on grammaticalization”. The Oxford Handbook
of Grammaticalization ed. by Heiko Narrog & Bernd Heine, 209–224. Oxford: Oxford Uni-
versity Press.
Poplack, Shana & Sali Tagliamonte. 2001. African American English in the Diaspora. Oxford:
Blackwell.
Potter, Jonathan. 2004. “Discourse analysis as a way of analysing naturally occurring talk”. Quali-
tative Research: Theory, Method and Practice ed. by David Silverman, 200–221. London:
Sage.
Precht, Kristen. 2008. “Sex similarities and differences in stance in informal American conversa-
tion”. Journal of Sociolinguistics 12: 1.89–111.
Psathas, George & Timothy Anderson. 1990. “The ‘practices’ of transcription in conversation
analysis”. Semiotica 78: 1-2.75–99.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik. 1985. A Comprehensive
Grammar of the English Language. London: Longman.
Raumolin-Brunberg, Helena & Arja Nurmi. 2011. “Grammaticalization and language change in
the individual”. The Oxford Handbook of Grammaticalization ed. by Heike Narrog & Bernd
Heine, 251–262. Oxford: Oxford University Press.
Redeker, Gisela. 1990. “Ideational and pragmatic markers of discourse structure”. Journal of
Pragmatics 14: 3.367–381.
Redeker, Gisela. 1991. “Linguistic markers of discourse structure”. Linguistics 29: 4.1139–1172.
252 The Structure of Discourse-Pragmatic Variation

Richards, Hazel. 2008. Mechanisms, motivations and outcomes of change in Morley (Leeds)
English. PhD diss., University of York.
Rickford, John, Isabelle Buchstaller, Thomas Wasow, Arnold Zwicky & Elizabeth Traugott. 2007.
“Intensive and quotative all: Something old, something new”. American Speech 82: 1.3–31.
Romaine, Suzanne. 1984. “On the problem of syntactic variation and pragmatic meaning in
sociolinguistic theory”. Folia Linguistica 18: 3-4.409–437.
Romaine, Suzanne. 1995. “The grammaticalization of irrealis in Tok Pisin”. Modality in Gram-
mar and Discourse ed. by Joan Bybee & Suzanne Fleischman, 389–428. Amsterdam: John
Benjamins.
Romaine, Suzanne & Deborah Lange. 1991. “The use of like as a marker of reported speech and
thought: A case of grammaticalization in progress”. American Speech 66: 3.227–279.
Roth-Gordon, Jennifer. 2007. “Youth, slang, and pragmatic expressions: Examples from Brazilian
Portuguese”. Journal of Sociolinguistics 11: 3.322–345.
Rudolph, Elisabeth. 1991. “Relationship between particle occurrence and text type”. Multilingua
10: 1-2.203–223.
Russell, Brenda, Jenna Perkins & Heather Grinnel. 2008. “Interviewees’ overuse of the word ‘like’
and hesitations: Effects in simulated hiring decisions”. Psychological Reports 102: 1.111–118.
Sacks, Harvey, Emanuel A. Schegloff & Gail Jefferson. 1974. “A simplest systematics for the
organisation of turn-taking for conversation”. Language 50: 4.696–735.
Sankoff, David. 1988. “Variable rules”. Sociolinguistics. An International Handbook of the Sci-
ence of Language and Society ed. by Ulrich Ammon, Norbert Dittmar & Klaus J. Mattheier,
984–997. Berlin: Mouton de Gruyter.
Sankoff, David, Sali Tagliamonte & E. Smith. 2005. Goldvarb X. A multivariate analysis applica-
tion. Department of Linguistics, University of Toronto, and Department of Mathematics,
University of Ottawa.
Sankoff, David & Pierrette Thibault. 1981. “Weak complementarity: Tense and aspect in Mon-
treal French”. Syntactic Change ed. by B. Johns & D. Strong, 206–216. Ann Arbor: University
of Michigan.
Sankoff, Gillian. 1973. “Above and beyond phonology in variable rules”. New Ways of Analysing
Variation in English ed. by Charles-James N. Bailey & Roger W. Shuy, 44–61. Washington:
Georgetown University Press.
Sankoff, Gillian & Hélène Blondeau. 2007. “Language change across the lifespan: /r/ in Montreal
French”. Language 83: 3.560–588.
Schegloff, Emanuel A. & Harvey Sacks. 1973. “Opening up closings”. Semiotica 7: 4.289–327.
Schegloff, Emanuel A. 1982. “Discourse as an interactional achievement: Some uses of ‘uh huh’
and other things that come between sentences”. Analyzing Discourse. Text and Talk ed. by
Deborah Tannen, 71–93. Washington, D.C.: Georgetown University Press.
Schegloff, Emanuel A. 1996. “Turn organisation: One intersection of grammar and interaction”.
Interaction and Grammar ed. by Elinor Ochs, Emanuel A. Schegloff & Sandra A. Thompson,
52–133. Cambridge: Cambridge University Press.
Schegloff, Emanuel A. 2007. Sequence Organisation in Interaction. A Primer in Conversation
Analysis. Cambridge: Cambridge University Press.
Scheibman, Joanne. 2000. “I dunno: A usage-based account of the phonological reduction of
don’t in American English conversation”. Journal of Pragmatics 32: 1.105–124.
Scheibman, Joanne. 2001. “Local patterns of subjectivity in person and verb type in American
English conversation”. Frequency and the Emergence of Linguistic Structure ed. by Joan Bybee
& Paul Hopper, 61–97. Amsterdam: John Benjamins.
 References 253

Schiffrin, Deborah. 1980. “Meta-talk: Organisational and evaluative brackets in discourse”.

Sociological Inquiry 50: 3-4.199–236.
Schiffrin, Deborah. 1987. Discourse Markers. Cambridge: Cambridge University Press.
Schiffrin, Deborah. 2001. “Discourse markers: Language, meaning and context”. The Handbook
of Discourse Analysis ed. by Deborah Schiffrin, Deborah Tannen & Heidi E. Hamilton,
54–75. Oxford: Blackwell.
Schleef, Erik. 2008. “The ‘lecturer’s ok’ revisited: Changing discourse conventions and the influ-
ence of academic division”. American Speech 83: 1.62–84.
Schneider, Klaus P. & Anne Barron. 2008. “Where pragmatics and dialectology meet: Introducing
variational pragmatics”. Variational Pragmatics. A Focus on Regional Varieties in Pluricentric
Languages ed. by Klaus P. Schneider & Anne Barron, 1–32. Amsterdam: John Benjamins.
Schourup, Lawrence. 1985. Common Discourse Particles in English Conversation. New York:
Garland.
Schourup, Lawrence. 1999. “Discourse markers: Tutorial overview”. Lingua 107: 3-4.227–265.
Simon-Vandenbergen, Anne-Marie. 1998. “The modal metaphor I don’t think: System and text”.
English as a Human Language. To Honour Louis Goossens ed. by Johan van der Auwera,
Louis Goosens, Frank Durieux & Ludo Lejeune, 312–324. München: Lincom Europa.
Simon-Vandenbergen, Anne-Marie. 2000. “The functions of I think in political discourse”.
International Journal of Applied Linguistics 10: 1.41–63.
Smith, Jennifer. 2000. “‘You Ø Na Hear O’ That Kind O’ Things’: Negative do in Buckie Scots”.
English World-Wide 21: 2.231–259.
Smith, Jennifer. 2009. Obsolescence vs. stability in a Shetland dialect: Evidence from three gen-
erations of speakers: Full Research Report. ESRC End of Award Report, RES-000-22-2052.
Swindon: ESRC.
Smith, Jennifer, Mercedes Durham & Hazel Richards. forthc. “The social and linguistic in the
acquisition of sociolinguistic norms: Caregivers, children and variation”. Linguistics.
Sowa, Joseph. 2009. “‘Sweet as!’: The intensifier as in New Zealand and Australian English”.
English Today 25: 2.58–61.
Sperber, Dan & Deirdre Wilson. 1995. Relevance. 2nd ed. Oxford: Blackwell.
Starks, Donna, Laura Thompson & James Christie. 2008. “Whose discourse particles?: New
Zealand eh in the Niuean migrant community”. Journal of Pragmatics 40: 7.1279–1295.
Steele, Hazel. 2003. Caregiver and child in the transmission and acquisition of variable dialect
features. MA diss., University of York.
Stein, Dieter. 1985. “Discourse markers in Early Modern English”. Papers from the 4th Interna-
tional Conference on English Historical Linguistics ed. by R. Eaton, 283–302. Amsterdam:
John Benjamins.
Stenström, Anna-Brita. 1990. “Pauses in monologue and dialogue”. The London-Lund Corpus
of Spoken English. Description and Research ed. by Jan Svartvik, 211–252. Lund: Lund Uni-
versity Press.
Stenström, Anna-Brita. 1998. “From sentence to discourse: Cos (because) in teenage talk”. Dis-
course Markers. Descriptions and Theory ed. by Andreas H. Jucker & Yael Ziv, 127–146.
Amsterdam: John Benjamins.
Stenström, Anna-Brita, Gisle Andersen & Ingrid Kristine Hasund. 2002. Trends in Teenage Talk.
Corpus Compilation, Analysis and Findings. Amsterdam: John Benjamins.
Stubbe, Maria & Janet Holmes. 1995. “You know, eh and other exasperating expressions: An
analysis of social and stylistic variation in the use of pragmatic particles in a sample of New
Zealand English”. Language and Communication 15: 1.63–88.
254 The Structure of Discourse-Pragmatic Variation

Stubbs, Michael. 1983. Discourse Analysis. The Sociolinguistic Analysis of Natural Language.
Chicago: The University of Chicago Press.
Stubbs, Michael. 1986. “‘A matter of prolonged fieldwork’: Notes towards a modal grammar of
English”. Applied Linguistics 7: 1.1–25.
Svartvik, Jan, ed. 1990. The London-Lund Corpus of Spoken English. Description and Research.
Lund: Lund University Press.
Tagliamonte, Sali. 2002a. “Comparative sociolinguistics”. The Handbook of Language Variation
and Change ed. by J. K. Chambers, Peter Trudgill & Natalie Schilling-Estes, 729–763. Ox-
ford: Blackwell.
Tagliamonte, Sali. 2002b. Grammatical Variation and Change in British English: Perspectives
from York: Full Research Report. ESRC End of Award Report, R-000-23-8287. Swindon:
ESRC.
Tagliamonte, Sali. 2005. “So who? Like how? Just what? Discourse markers in the conversations
of young Canadians”. Journal of Pragmatics 37: 11.1896–1915.
Tagliamonte, Sali. 2008. “So different and pretty cool! Recycling intensifiers in Toronto, Canada”.
English Language and Linguistics 12: 2.361–394.
Tagliamonte, Sali & Alexandra D’Arcy. 2004. “He’s like, she’s like: The quotative system in Cana-
dian youth”. Journal of Sociolinguistics 8: 4.493–514.
Tagliamonte, Sali & Alexandra D’Arcy. 2007. “Frequency and variation in the community gram-
mar: Tracking a new change through the generations”. Language Variation and Change
19: 2.199–217.
Tagliamonte, Sali & Alexandra D’Arcy. 2009. “Peaks beyond phonology: Adolescence, incremen-
tation, and language change”. Language 85: 1.58–108.
Tagliamonte, Sali & Derek Denis. 2008. “Linguistic ruin? LOL! Instant messaging and teen
language”. American Speech 83: 1.3–34.
Tagliamonte, Sali & Derek Denis. 2010. “The stuff of change: General extenders in Toronto,
Canada”. Journal of English Linguistics 38: 4.335–368.
Tagliamonte, Sali & Rachel Hudson. 1999. “Be like et al. beyond America: The quotative system
in British and Canadian youth”. Journal of Sociolinguistics 3: 2.147–172.
Tagliamonte, Sali & Chris Roberts. 2005. “So weird; so cool; so innovative: The use of intensifiers
in the television series Friends”. American Speech 80: 3.280–300.
Tagliamonte, Sali & Jennifer Smith. 2002. “‘Either it isn’t or it’s not’: NEG/AUX contraction in
British dialects”. English World-Wide 23: 2.251–281.
Tannen, Deborah. 1993. Gender and Conversational Interaction. Oxford: Oxford University Press.
Terkourafi, Marina. 2011. “The pragmatic variable: Toward a procedural interpretation”. Lan-
guage in Society 40: 3.343–372.
Terraschke, Agnes. 2007. “Use of general extenders by German non-native speakers of English”.
International Review of Applied Linguistics 45: 2.141–160.
Thompson, Sandra A. 2002. “‘Object complements’ and conversation”. Studies in Language
26: 1.125–164.
Thompson, Sandra A. & Paul J. Hopper. 2001. “Transitivity, clause structure, and argument
structure: Evidence from conversation”. Frequency and the Emergence of Linguistic Structure
ed. by Joan L. Bybee & Paul Hopper, 27–60. Amsterdam: John Benjamins.
Thompson, Sandra A. & Anthony Mulac. 1991a. “The discourse conditions for the use of the
complementizer that in conversational English”. Journal of Pragmatics 15: 3.237–251.
 References 255

Thompson, Sandra A. & Anthony Mulac. 1991b. “A quantitative perspective on the grammati-
cization of epistemic parentheticals in English”. Approaches to Grammaticalization. Vol. 2:
Types of Grammatical Markers ed. by Elizabeth Closs Traugott & Bernd Heine, 313–329.
Amsterdam: John Benjamins.
Todd, Loreto & Ian Hancock. 1986. International English Usage. London: Croom Helm.
Torgersen, Eivind, Costas Gabrielatos, Sebastian Hoffmann & Susan Fox. 2011. “A corpus-based
study of pragmatic markers in London English”. Corpus Linguistics and Linguistic Theory
7: 1.93–118.
Torres Cacoullos, Rena. 2011. “Variation and grammaticalization”. The Handbook of Hispanic
Sociolinguistics ed. by Juan Manuel Hernández-Campoy & J. Camilo Condes-Silvestre,
148–167. Oxford: Wiley-Blackwell.
Torres Cacoullos, Rena & James Walker. 2009. “The present of the English future: Grammatical
variation and collocations in discourse”. Language 85: 2.321–354.
Torres Cacoullos, Rena & James Walker. 2011. “Collocations in grammaticalization and vari-
ation”. The Oxford Handbook of Grammaticalization ed. by Bernd Heine & Heiko Narrog,
225–238. Oxford: Oxford University Press.
Torres, Lourdes. 2002. “Bilingual discourse markers in Puerto Rican Spanish”. Language in So-
ciety 31: 1.65–83.
Tottie, Gunnel. 1992. “Introduction: Seven types of continuity in discourse”. Language Variation
and Change 4: 2.121–123.
Tottie, Gunnel & Sebastian Hoffmann. 2006. “Tag questions in British and American English”.
Journal of English Linguistics 34: 4.283–311.
Tovena, L. M. 2001. “Neg-raising: Negation as failure”. Perspectives on Negation and Polarity Items
ed. by Jack Hoeksema, Hotze Rullmann, Victor Sanchez-Valencia & Ton van der Wonden,
331–356. Amsterdam: John Benjamins.
Traugott, Elizabeth Closs. 1982. “From propositional to textual to expressive meanings: Some
semantic-pragmatic aspects of grammaticalization”. Perspectives on Historical Linguistics
ed. by Winnfried P. Lehmann & Yakov Malkiel, 245–271. Amsterdam: John Benjamins.
Traugott, Elizabeth Closs. 1995. “Subjectification in grammaticalization”. Subjectivity and Sub-
jectivisation ed. by Dieter Stein & Susan Wright, 31–54. Cambridge: Cambridge University
Press.
Traugott, Elizabeth Closs. 2003a. “Constructions in grammaticalization”. The Handbook of His-
torical Linguistics ed. by Brian D. Joseph & Richard D. Janda, 624–647. Oxford: Blackwell.
Traugott, Elizabeth Closs. 2003b. “From subjectification to intersubjectification”. Motives for Lan-
guage Change ed. by Raymond Hickey, 124–139. Cambridge: Cambridge University Press.
Traugott, Elizabeth Closs. 2010. “Grammaticalization”. Continuum Companion to Historical Lin-
guistics ed. Silvia Luraghi & Vit Bubenik, 269–283. London: Continuum Press.
Traugott, Elizabeth Closs. 2012. “Intersubjectification and clause periphery”. English Text Con-
struction 5: 1.7–28.
Traugott, Elizabeth Closs & Richard B. Dasher. 2002. Regularity in Semantic Change. Cambridge:
Cambridge University Press.
Traugott, Elizabeth Closs & Bernd Heine. 1991. “Introduction”. Approaches to Grammaticaliza-
tion. Vol. 2: Focus on Types of Grammatical Markers ed. by Elizabeth Closs Traugott & Bernd
Heine, 1–14. Amsterdam: John Benjamins.
256 The Structure of Discourse-Pragmatic Variation

Traugott, Elizabeth Closs & Ekkehard König. 1991. “The semantics-pragmatics of grammati-
calization revisited”. Approaches to Grammaticalization. Vol. 1: Theoretical and Methodo-
logical Issues, ed. by Elizabeth Closs Traugott & Bernd Heine, 189–218. Amsterdam: John
Benjamins.
Traugott, Elizabeth Closs & Graeme Trousdale. 2010. “Gradience, gradualness and grammati-
calization. How do they intersect?”. Gradience, Gradualness and Grammaticalization ed.
by Elizabeth Closs Traugott & Graeme Trousdale, 19–44. Amsterdam: John Benjamins.
Trester, Anna Marie. 2009. “Discourse marker ‘oh’ as a means for realizing the identity potential
of constructed dialogue in interaction”. Journal of Sociolinguistics 13: 2.147–168.
Trudgill, Peter & Jean Hannah. 1982. International English. A Guide to Varieties of Standard
English. London: Edward Arnold.
Tsui, Amy B. M. 1991. “The pragmatic functions of I don’t know”. Text 11: 4.607–622.
Upton, Clive, David Parry & J. D. A. Widdowson. 1994. Survey of English Dialects. The Dictionary
and Grammar. London: Routledge.
Urmson, J. O. 1952. “Parenthetical verbs”. Mind 61.480–496.
van Bogaert, Julie. 2010. “A constructional taxonomy of I think and related expressions:
Accounting for the variability in complement-taking mental predicates”. English Language
and Linguistics 14: 3.399–427.
van Bogaert, Julie. 2011. “I think and other complement-taking mental predicates: A case of and
for constructional grammaticalization”. Linguistics 49: 2.295–332.
van der Auwera, Johan. 2009. “The Jespersen Cycles”. Cyclical Change ed. by Elly van Gelderen,
35–71. Amsterdam: John Benjamins.
Verdonik, Darinka, Andrej Žgank & Agnes Pisanski Peterlin. 2009. “The impact of context on
discourse marker use in two conversational genres”. Discourse Studies 10: 6.759–775.
Wales, Katie. 2006. Northern English. A Social and Cultural History. Cambridge: Cambridge
University Press.
Walker, James A. 2010. Variation in the Linguistic System. London: Routledge.
Wallage, Phillip. 2008. “Jespersen’s Cyle in Middle English: Parametric variation and grammati-
cal competition”. Lingua 118: 5.643–674.
Waters, Cathleen. 2013. “Transatlantic variation in English adverb placement”. Language Vari-
ation and Change 25: 2.
Watt, Dominic & Catherine Ingham. 2000. “Durational evidence of the Scottish Vowel Length
Rule in Berwick English”. Leeds Working Papers in Linguistics and Phonetics 8.205–228.
Watt, Dominic & Lesley Milroy. 1999. “Patterns of variation and change in three Newcastle
vowels: Is this dialect levelling?”. Urban Voices. Accent Studies in the British Isles ed. by Paul
Foulkes & Gerard J. Docherty, 25–46. London: Arnold.
Watt, Dominic, Carmen Llamas & Gerard J. Docherty. 2012. Linguistic Variation and National
Identities on the Scottish/English Border: Full Research Report. ESRC End of Award Re-
port, RES-062-23-0525. Swindon: ESRC.
Watts, Richard J. 1984. “An analysis of epistemic possibility and probability”. English Studies
65: 2.129–140.
Watts, Richard J. 1989. “Taking the pitcher to the ‘well’: Native speakers’ perception of their use
of discourse markers in conversation”. Journal of Pragmatics 13: 3.203–237.
Wauchope, Mary Michele. 1993. “Discourse functions of just”. Lacus Forum 19.181–189.
Weatherall, Ann. 2011. “I don’t know as a prepositioned epistemic hedge”. Research on Language
and Social Interaction 44: 4.317–337.
 References 257

Weinreich, Uriel, William Labov & Marvin Herzog. 1968. “Empirical foundations for a theory
of language change”. Directions for Historical Linguistics ed. by Winfried Lehmann & Yakov
Malkiel, 95–188. Austin: University of Texas Press.
Whisker, Kate. 2007. Secondary contraction of clitic negatives in urban North Yorkshire. MA
diss., University of York.
Whisker-Taylor, Kate. in prep. A century of language variation and change in Huddersfield. PhD
diss. in progress, Lancaster University.
Wichmann, Anne. 2011. “Grammaticalization and prosody”. The Oxford Handbook of Grammati-
calization ed. by Heiko Narrog & Bernd Heine, 331–341. Oxford: Oxford University Press.
Wilson, Deirdre & Dan Sperber. 1993. “Linguistic form and relevance”. Lingua 90: 1.1–25.
Wilson, James. 1915. Lowland Scotch as Spoken in the Lower Strathrearn District of Perthshire.
Oxford: Oxford University Press.
Winford, Donald. 1984. “The linguistic variable and syntactic variation in creole continua”.
Lingua 62: 4.267–288.
Winford, Donald. 1996. “The problem of syntactic variation”. Sociolinguistic Variation. Selected
Papers from NWAV 23 at Stanford ed. by Jennifer Arnold, Renée Blake, Brad Davidson, Scott
Schwenter & Julie Solomon, 177–192. Stanford, CA: CSLI Publications.
Wolfson, Nessa. 1976. “Speech events and natural speech: Some implications for sociolinguistic
methodology”. Language in Society 5: 2.189–209.
Woods, Howard B. 1991. “Social differentiation in Ottawa”. English around the World. Socio-
linguistic Perspectives ed. by Jenny Cheshire, 134–149. Cambridge: Cambridge University
Press.
Woods, Nicola. 1997. “The formation and development of New Zealand English: Interaction
of gender-related variation and linguistic change”. Journal of Sociolinguistics 1: 1.95–126.
Wooffitt, Robin. 2005. Conversation Analysis and Discourse Analysis. A Comparative and Critical
Introduction. London: Sage.
Wray, Alison. 2002. Formulaic Language and the Lexicon. Cambridge: Cambridge University
Press.
Wray, Alison & Michael R. Perkins. 2000. “The functions of formulaic language: An integrated
model”. Language & Communication 20: 1.1–28.
Wright, Joseph. 1902. The English Dialect Dictionary. London: Henry Frowde.
Yaeger-Dror, Malcah. 1985. “Intonational prominence on negatives in English”. Language and
Speech 28: 3.197–230.
Yaeger-Dror, Malcah. 1997. “Contraction of negatives as evidence of variance in register-specific
interactive rules”. Language Variation and Change 9: 1.1–36.
Yang, Li-chiung. 2006. “Integrating prosodic and contextual cues in the interpretation of
discourse markers”. Approaches to Discourse Particles ed. by Kerstin Fischer, 265–297.
Amsterdam: Elsevier.
Young, Richard & Robert Bayley. 1996. “VARBRUL analysis for second language acquisition
research”. Second Language Acquisition and Linguistic Variation ed. by Robert Bayley &
Dennis R. Preston, 253–306. Amsterdam: John Benjamins.
Zilles, Ana M. S. 2005. “The development of a new pronoun: The linguistic and social embedding
of a gente in Brazilian Portuguese”. Language Variation and Change 17: 1.19–53.
appendices
Appendix 1.  Inventory of functions of unbound i don’t know in the BwE corpus
function description accompanying features (turn-)position example in
Chapter 4
INTERPERSONAL MODE
# subjective uses: – signals speakers’ uncertainty + rising/fall-rising intonation ~ any (3)
epistemic marker vis-à-vis the validity of + usually co-occurs with other ~ external to syntactic
propositions (global scope) expressions of epistemicity structures
– s ignals that the following + rising intonation ~ t urn-medially before (4)
lexical material is in some way + lexical material in its scope set off modified lexical material
problematic (local scope) prosodically from surrounding ~u  tterance-medially between
discourse obligatory constituents of
a clause
# intersubjective uses: –m
 itigates potentially + rising intonation ~ any (turn-initial pre- (5a)
mitigation device controversial propositions positioning preferred)
– softens disagreements +h
 igh pitch and/or rising intonation ~ turn-initially (5b)
– avoids unwelcome assessments + falling intonation ~ turn-initially or constituting (6)
sole turn-component
TEXTUAL MODE
# repair device – bridges transitions between + a fter cut-offs, abandoned utterances, ~ t urn-medially between (7a)
repaired and repairing false starts repaired and repairing (7b)
elements + usually level intonation elements
+w  ith discourse-pragmatic features, ~n  ear-turn-initially after
(un)filled pauses & higher pitch to false starts
signal hesitant repair
# topic-development – declines proffered topics + usually no pitch movement ~ sole turn-component (8)
device +o ccasionally descending intonation ~ after topic-proffers
contour
function description accompanying features (turn-)position example in
Chapter 4
– curtails topical sequences + usually no pitch movement ~ sole turn component or (9)
+ occasionally with descending near-turn-finally
intonation contour ~ after topic-proffer was
briefly embraced
– dismisses successful disruptions + o
 ften (moderately) rising intonation ~ turn-initially (10)
from hearers to pursue original ~ a fter disruptions of
topic speakers’ hold on the floor
# turn-taking device – launches contributions whilst + often with increased speech rate ~ turn-initially (11a)
planning is still underway + c o-occurs with unfilled pauses and
(sometimes but not necessarily other hesitation signals
in competition for the floor) + f ollowing elements sometimes
produced with reduced speech rate
+n  ot usually with rising intonation
262 The Structure of Discourse-Pragmatic Variation

– non-hesitantly launches +u
 sually no noticeable pitch movement ~ turn-initially (11b)
pre-planned turns +w
 hen in overlap position, sometimes
with syllable lengthening
# turn-holding device – links (thematically unrelated) + no noticeable pitch movement ~ turn-medially (12a)
TCUs to maintain speakers’ + i ncreased speech rate (when produced
hold on floor in competition for the floor)
– signals speakers’ +o
 ften after transition-marking ~ turn medially or (12b)
communicative presence in discourse-pragmatic features near-turn-initially after
hesitation and planning areas +u
 sually co-occurs with hesitation discourse-pragmatic
of turns signals (discourse-pragmatic features, features or filled pauses
filled and unfilled pauses)
+m
 oderately rising pitch or no prosodic
variation


function description accompanying features (turn-)position example in

Chapter 4
# turn-yielding/-closing – yields turns + f alling tone (to signal finality and ~ (near-)turn-finally (13)
device completion) or rising tone (to appeal
for cooperation)
+ often softly spoken
– closes turns to signal intention +u
 sually descending intonation to ~ turn-finally (14a)
to yield the floor and/or desire indicate finality/completeness ~ at CTRPs and non-CTRPs (14b)
to terminate topics +o
 ccasionally with high rising pitch to
signal appeal for co-operation
INTERPERSONAL-TEXTUAL MODE
# epistemic turn-taking – launches turns and marks fol- + rising intonation ~ turn-initially (15)
device lowing propositions as uncertain
# epistemic repair or – marks transitions between +b
 etween repairing and repaired ~ turn-medially (16a)
turn-holding device repaired and repairing elements elements with rising intonation OR (16b)
and qualifies upcoming after transition-markers with hesitation
propositions OR functions as signals and high-rising pitch
turn-holder while qualifying
preceding or following
propositions
# epistemic turn-closing – qualifies preceding propositions + rising or falling intonation ~ turn-finally (17)
device and closes turns
IDEATIONAL MODE
# referential uses – communicates insufficient + know carries primary stress ~ any (18a)
knowledge +n ot usually co-occurring with other (18b)
discourse-pragmatic features or pauses
+n ot usually produced with prosodic
variation
+ frequently with really
Appendices 263
Appendix 2.  Inventory of functions of bound i don’t know in the BwE corpus
[Shading indicates functions performed by utterances containing i don’t know rather than by the construction i don’t know per se.]
function description accompanying features (turn-)position example
in Chapter 4
IDEATIONAL MODE
# referential uses – communicates lack of + accented ~ any turn-position (19a)
knowledge of or familiarity + s ometimes strong emphatic stress on (19b)
with referents in complements (almost) all elements of utterance (19c)
+ falling intonation contour
+n  ot usually co-occurs with other
discourse-pragmatic features or pauses
+ often with really
INTERPERSONAL MODE
# epistemic marker – signals speakers’ uncertainty + soft volume ~ any turn-position (20)
vis-à-vis validity of propositions + r ising pitch on know or (last element (21)
264 The Structure of Discourse-Pragmatic Variation

in following complements of) complements (22)

(local scope)
– signals speakers’ uncertainty +u
 sually co-occurs with other ~ any turn-position (23a)
vis-à-vis validity or relevance epistemic markers ~u sually before the modified (23b)
of (a) surrounding utterance(s) +u
 sually with rising intonation contour utterance, but occasionally (24)
(global scope) or high pitch after it
TEXTUAL MODE
# topic-development – declines proffered topics +n
 o pitch movement or descending ~ (near) sole turn-component (25a)
device intonation contour
+o
 ften softly spoken with slightly
raised pitch
– curtails topical sequences + any intonation contour ~ sole turn component after (25b)
lengthy discussion of topics
# turn-yielding/-closing – yields turns +h
 igh pitch on the i don’t know with ~ turn-finally (25c)
device final falling tone on complement


Appendix 3.  Inventory of functions of i don’t know + wh-word in the BwE corpus
[Shading indicates functions performed by utterances containing i don’t know rather than by the construction i don’t know per se.]
function description accompanying features (turn-)position example
in Chapter 4
INTERPERSONAL MODE
# (local or global) – signals low reliability + high pitch or rising intonation ~ any (26a)
epistemic marker (26b)
TEXTUAL MODE
# topic-development – declines proffered topics + falling intonation contour ~ (near-)sole turn-component (27a)
device ~ in response to topic-proffers
# turn-yielding/ – yields turns + at CTRPs or non-CTRPs ~ turn-finally (27b)
-closing device + descending intonation to indicate
finality & completeness
Appendices 265
Appendix 4.  Inventory of functions of i don’t think in the BwE corpus
function description accompanying features sequential positioning example
in Chapter 5
INTERPERSONAL MODE
# subjective uses: – signals speakers’ uncertainty +w  ith high pitch, low volume, ~ any turn-position (7)
subjective epistemic vis-à-vis validity decreased tempo (and ~ mostly in response to opinion (8)
modality marker of propositions moderate rise on think, don’t questions:
or the end of the utterance) mostly (near-)sole turn-
(high degree of uncertainty) component or (near-)turn-
+w  ithout prosodic variation initial positioning
and mostly with level or falling occasionally followed by
intonation (low degree of validation
uncertainty) often preceded or followed
+ s ometimes with other low by tentative no
266 The Structure of Discourse-Pragmatic Variation

modality items after polar questions often

+u  sually no stress; only with pro-form so
occasionally stress on I to ~ in response to prior assessments
emphasise subjectivity, or on ~ as part of extended opinion
think if followed by so sequences involving one or more
speakers
– signals speakers’ certainty + s tress on don’t (think) (and ~ any turn-position   (9)
vis-à-vis validity other negative polarity items ~ mostly (near-)turn-initially in (10)
of propositions in the utterance) second-pair parts of question- (11)
+ f alling or final intonation answer or assessment-assessment
contour adjacency pairs
+ s ometimes decreased speech ~ also in opinion sequences to
rate and/or increased loudness emphasise a point
+ s ometimes accompanied by
assertiveness devices


function description accompanying features sequential positioning example

in Chapter 5
# intersubjective uses (1): – mitigates potentially + level or falling intonation ~ any turn-position (12)
mitigation device controversial or face- contour
threatening propositions + s ometimes contrastive pitch,
speed and volume
+ unaccented
– softens presentation of + f requently with high pitch ~ (near-)turn-initially (13)
disaligning turns and/or reduced tempo ~ usually followed by resolution
+ s ometimes preceded by filled of rejection (explication for
and/or unfilled pauses denial or provision of alternative
+ unaccented assessment)
– avoids or reduces bluntness of + flat or falling intonation ~ (near-)turn-initially (14)
overt denials and rejections + always without noticeable ~ in response to polar questions
variation in tempo and ~ s ometimes followed by explication
loudness of denial
+ unaccented
+ often co-occurs with definitive-
sounding no
# intersubjective uses (2): – signals current speakers’ + with level or falling intonation ~ usually (near-)sole turn (15)
affiliation marker agreement with or (boosts agreement) component (including
corroboration of previous + with (moderately) rising complementation) or with no
speakers’ propositions or intonation (hedges agreement) ~ not usually followed by further
inferences talk
INTERPERSONAL-TEXTUAL FUNCTIONS
# information- & –m
 arks transitions from +u sually without prosodic ~ t urn-medially between provision (16)
discourse-structuring provisions of factual variation of facts and provision of personal
device: frame information to provisions of + falling intonation contour stance
personal assessments of facts + often with but ~ usually in extended turns
Appendices 267
function description accompanying features sequential positioning example
in Chapter 5
– marks transitions between + mostly with no marked ~u
 sually (near-)turn-finally after (17)
explications and provisions prosodic variation extended validation sequences
of final assessments/ + sometimes with increased
conclusions speech rate (neutral to assertive
conclusion)
+ infrequently with high pitch,
slow speed, low volume
(tentative conclusion)
+ often with so or but
–m
 arks onsets of opinion + usually with neutral prosody ~ mostly near-turn-initially (18)
sequences after turn-initial + s ometimes produced with ~ after yes/no or hesitation
hesitation, or elaborations increased speed (meaning of ~ points backwards and forwards
268 The Structure of Discourse-Pragmatic Variation

after turn-initial minimally un/certainty is bleached)

coherent responses + s ometimes with reduced tempo
and variation in pitch (tentative
explication)
– marks shifts in the topical +m
 ostly falling but also level ~ turn-medially when providing (19)
development of interactions intonation additional unsolicited information
+n
 ot usually accompanied by ~ very occasionally turn-initially
variation in prosodic marking after taking the floor and
providing new, unsolicited
information
– structures longer opinion + mostly neutral to assertive tone ~ any position (20)
sequences + s ometimes bleached of
epistemic content
+m  ostly with I think in extended
turns


Appendix 5.  Inventory of functions of neg-tags in the BwE corpus

function description accompanying features (turn-)position example
in Chapter 6
SUBJECTIVE USES
# epistemic marker – signals speakers’ uncertainty +u
 sually rising intonation ~ turn-finally (22)
vis-à-vis validity of preceding and/or high pitch ~ followed by turn-exchange: (23)
propositions +o
 ften co-occurs with other addressees provide minimal or
– seeks verification of expressions of epistemicity elaborate responses to verify
propositions from addressees and/or signals of hesitancy previous speakers’ propositions
– conducive in the anchor clause
# attitudinal stance – foregrounds and emphasises +u
 sually level intonation ~ turn-medially: followed by brief (24)
marker preceding propositions contour pause whilst addressee signals
– signals speaker commitment continued attention and interest, or
– non-conducive amidst flow of rapid speech
~ turn-finally: generally followed by
topic-change initiated by addressees
– underlines obviousness of + level or falling intonation ~ usually turn-finally at the end of (25)
preceding statements + neutral tone longer topic sequences
– non-conducive ~ generally followed by topic-shifts
INTERSUBJECTIVE USES
# mitigation device –m itigates face-threat of + rising or falling intonation ~ usually turn-finally (26)
dismissals, disagreements ~ usually followed by turn-exchange
and ironic/sarcastic remarks and further elaboration of topic
– conducive and non-conducive
Appendices 269
function description accompanying features (turn-)position example
in Chapter 6
# involvement inducer – draws listeners into discourse + falling intonation ~ turn-medially followed by brief (27)
– seeks corroboration of pause to allow room for minimal
propositions agreement tokens
– secures addressees’ involvement ~ turn-finally to yield the floor to
in discourse interlocutors
– conducive
# alignment signal – signals alignment with + r ising tone to signal ~ usually turn-finally (28)
previous speakers and enthusiasm ~ g enerally not followed by further (29)
involvement in interaction + level and falling intonation to talk to allow interlocutors to
– non-conducive signal confidence and emphasis continue
270 The Structure of Discourse-Pragmatic Variation
 Index

A because  80–1 Bybee, Joan  35, 37, 38, 69, 70,

acknowledgment tokens  44 Bell, Barbara  235 74–5, 80, 81, 106, 117, 118, 124,
adjacency pairs  43 Bergen Corpus of London 128–9, 130, 134, 135, 136, 150,
adjective + as  14 Teenage Language (COLT) 161, 163, 220
Aijmer, Karin  39, 73 15, 173, 175, 176, 177, 209
ain’t it  174–6 Berwick Community High C
Akatsuka, Noriko  95 School  48 CA see conversation analysis
Algeo, John  172, 189, 206 Berwick English (BwE)  17–18 Cameron, Deborah  25
all  31n general extenders  16 Campbell, Lyle  38
American English hybrid nature of  24–5 cannae see -nae
be like  14, 210–11 see also BwE verb negation cannit  54
go  210–11 system; i don’t know; distributional analysis  59–62,
i don’t know  74–5, 80 i don’t think; negative 59f, 61t, 62t
i don’t think  124, 128, 130, polarity question tags extra-linguistic variation  57
134, 135, 136, 150, 163 (neg-tags) geographical distribution 
analytical methods  230–1 Berwick upon Tweed  23, 23f 55, 56t
see also methodology; Biber, Douglas et al.  229 challenges for the future  227–36
qualitative data analysis; Blakemore, Diane  5 analytical methods  230–1
quantitative data analysis; boosters  30 intra-linguistic constraints
quantitative methods i don’t think  140, 159, 164 231–2
Andersen, Elaine S. et al.  10 I think  128 low token frequency  228–30
Andersen, Gisle  15, 16, 29, 173, Brinton, Laurel J.  8, 11, 36, 38, patterns of discourse-
175, 176, 179, 180n, 184, 200, 39, 45, 129 pragmatic change  232–4
207, 208, 209, 210, 211, 222 Britain, David  14, 210 range of discourse-pragmatic
Anderwald, Lieselotte  57, 58, 134 British National Corpus (BNC) variables  234–5
apparent-time framework  26 75, 175, 209–10, 229 Cheshire, Jenny et al.  11–12, 14,
attitudinal stance markers: Brown, Gillian  7 28, 29, 30, 172–3, 176, 177, 183,
neg-tags  188–9, 192, 200, Brown, Keith  56, 57 186, 189, 206, 207, 210, 230
202, 203, 206, 208, 210 Brown, Penelope  45, 96 clause-final tags  5–6, 10–11
Bublitz, Wolfram  126–7, 128, 166 Cognitive Prominence Principle
B Buchstaller, Isabelle  14, 30, (CPP)  159, 161, 164
Barbieri, Federica  11 210–11, 234, 235 COLT see Bergen Corpus of
Barnfield, Kate  234, 235 BwE see Berwick English London Teenage Language
Barron, Anne  6 BwE verb negation system  51–2 comment clauses  5–6
Baumgarten, Nicole  73–4 distributional analysis  59–62, conduciveness: neg-tags 
Bayley, Robert  28, 32 59f, 61t, 62t 173, 177, 188–90, 192–3, 193t,
Bazzanella, Carla  45 evolution of negation/ 200, 202–5, 206–7
be  52n negative particles  52–3 conversation analysis (CA)  41–5
be all  14 multivariate analysis  63–5, 64t choice of CA  42
be like  12, 14, 210–11 negative particles/auxiliaries key concepts  42–5
Beach, Wayne A.  72 53–6, 56t, 59–66 see also qualitative data
Beal, Joan C.  57 quantitative analysis  59–65 analysis
272 The Structure of Discourse-Pragmatic Variation

conversational implicatures  39 function-based F

corpora  22–5, 55, 58, 127, 229– conceptualisation  29–31 face needs  45, 46
30, 233–4 inference-based approaches i don’t know  71–2, 82,
Corrigan, Karen P.  57 30–1 83–4, 86, 87
Coupland, Nikolas  6 multifunctionality  47, 216–17, i don’t think  139, 141–2
CPP see Cognitive Prominence 223–4 neg-tags  189
Principle range  234–5 false starts  44
Croft, William  222 see also discourse-pragmatic Ferrara, Kathleen W.  235
features filled pauses  44
D discourse-pragmatic variation first language acquisition  10
D’Arcy, Alexandra  9, 12, 14, 16, 10–11, 12–13, 13–14, 227–8, Fischer, Kerstin  14n, 45
229, 230, 232, 234 231–2, Ford, Cecilia E.  72
Dasher, Richard B.  36–7 discussion and conclusion  formulaic constructions  5
data  22–7 215–25 Fox, Barbara A. et al.  85
decategorialisation  37, 38 implications of results  220–4 Fraser, Bruce  5, 98
Degand, Liesbeth  8 synthesis of results  216–20 Fuller, Janet M.  15
Denis, Derek  14, 230 divn’t  51, 54 functional comparability 
desemanticisation  37, 39 distributional analysis  59–62, 9, 29–30
dialect convergence  11–12 59f, 61t, 62t functional domains  45–7
dialect variation  13–14 extra-linguistic variation  57
Diani, Giuliana  72–3 geographical distribution  G
Diewald, Gabriele  8 55, 56t general extenders  14, 16, 29–30,
din: neg-tags  184, 194–5, 195t multivariate analysis  63–5, 31n
Dines, Elizabeth R.  9, 10–11, 64t geographical dialect variation
29–30, 31 in neg-tags  194–5, 195t 13–14
dinnae  51, 54 see also I divn’t knaa; I divn’t see also dialect convergence
see also I dinnae ken; think Glasgow English  14
I dinnae think do  52n, 54 go  210–11
discourse markers  5 do  52n Goffman, Erving  45
discourse-pragmatic change  don’t  129 going to  32n
11–12, 13, 232–4 see also  i don’t know; GoldVarb  32, 33
discourse-pragmatic features  28 i don’t think grammatical function  8
age-based differences  11 grammaticalisation  35–41
co-occurring features  45 E chains  36n
grammaticalisation  13, 38–40 emphasis definition  35
low token frequency  228–30 i don’t know  94 and discourse-pragmatic
procedural meanings  30 i don’t think  139–40 features  13, 38–40
qualitative discourse studies innit  188–9 as expansion  39n
15, 16 epistemic marking i don’t know  74–5, 79, 80,
quantitative analysis  10–16 i don’t know  69, 72, 83, 91, 118–20, 121, 220–2
reasons for neglect of  6–9, 14 92, 95, 97–8 i don’t think  129, 135, 163,
scope and terminology  4–6 i don’t think  123, 126, 167–8, 220–2
and variationist 127, 129, 137, 140, 153, 158, mechanisms of change  35–8
sociolinguistics  3–4, 10–11, 163–4, 166–7 neg-tags  176, 179, 185–6,
222–4 neg-tags  187–8 207, 210, 222
see also discourse-pragmatic Erman, Britt  11, 29, 45 vs. pragmaticalisation  40
variables erosion  37 as reduction  39n
discourse-pragmatic variables  ethnic minority speakers  in synchronic data  41
5, 28–32 29, 175, 210, 211 Grant, Lynn E.  75
derivation-based definition extension  37 Günthner, Susanne  40
31–2 Guy, Gregory R.  110n
 Index 273

H i don’t know  31, 65, 69–122 I doØ think  123, 134

Halliday, M. A. K.  45 bound tokens with phrasal/ distributional analysis 
Haspelmath, Martin  38 causal complements  150–5, 151f, 152f, 154f, 155f
have, have got  52n, 60n 93–7, 264 multivariate analysis  156–61,
hedging coding  80–1, 99, 99t 157t, 160f
i don’t know  72, 82, 83, 92, with dependent wh- discussion  164
95–6, 161, 163 words  97–8, 265 I dunno  69–70, 77, 79
i don’t think  123, 124, 126, distributional analysis  100–7, bound tokens  95–7
137–8, 140, 142–3, 159, 101f, 102f, 103t, 104f, 105t, distributional analysis  100–7,
161, 163 107f 101f, 102f, 104f
Heine, Bernd  36, 37, 38, 207 multivariate analysis  multivariate analysis 
Hertfordshire  176, 177, 211 108–15, 109t 108, 110–12, 111t, 117–18
Hoffmann, Sebastian  189, 207, present investigation  76–7 unbound tokens  82–3, 85–6,
287 previous research  71–5 87–90, 91, 92
Holmes, Janet  9, 47, 172, 180n, qualitative analysis  48, discussion  119, 120, 121
206, 209, 230 81–99, 99t, 116–17, 261–5 I mean  29, 38
Hopper, Paul J.  35, 36, 120, 176 quantitative analysis  100–16, I think  38, 39, 128, 167
Horn, Laurence R.  123 117 ‘I think not-p’  126, 128, 166
House, Juliane  73–4 unbound tokens  82–93, ideational mode  45–6, 99t
Hudson, Rachel  235 261–3 in + deed  39
Hudson, Richard A.  172, 173 the variable and envelope of in it  177
Hughes, Arthur et al.  14 variation  77–8 in-tags see in’t-tags
Hull  176, 210 variants in BwE  78–9, 80t information-structuring role:
Hutchby, Ian  72 discussion  116–21 i don’t think  128, 136, 144–8
see also I dinnae ken; I divn’t innit  55, 185, 188–9
I knaa; I dono; I dunno derivation  184, 207, 211
I dinnae ken  69–70, 77, 79, 100, I don’t suppose  132, 153, 166 distributional analysis 
101f i don’t think  31, 65, 123–68 194–6, 195t, 197f, 198–200,
I dinnae think  123, 135, 150, coding  135–6, 149t 199f, 201f
151f, 153 distributional analysis  150–5, multivariate analysis  203–4,
I divn’t knaa  69–70, 77, 78, 79 151f, 152f, 154f, 155f 204t
bound tokens  94–5 frequency  34 non-paradigmatic uses  175
with dependent wh-word 97–8 multivariate analysis  156–61, previous research  12, 14, 29,
distributional analysis  100–7, 157t, 160f 169, 171, 173, 174–6, 177
101f, 102f, 104f post-posed tokens  133, 153 discussion  207–11
multivariate analysis  108, 112, pre-positioned tokens  in’t it  174–6, 184
113–15, 114t 133, 153 in’t-tags  180, 184, 194–5, 195t
unbound tokens  83, 84, 87, 91 present investigation  130–1 intensifiers  30, 31n
discussion  120–1 previous research  125–30 International Corpus of English
I divn’t think  123, 135 qualitative analysis  136–49, 229
distributional analysis  149t, 266–8 interpersonal mode  45–6
150–5, 151f, 152f, 154f, 155f quantitative analysis  150–62 i don’t know  71, 72, 74,
multivariate analysis  the variable and envelope of 82–4, 91, 99t, 103–5, 106
156–61, 157t, 160f variation  131–3 i don’t think  124, 127, 129,
discussion  164–5 variants in BwE  134–5, 135t 137–41, 148, 149t
I dono  69, 78, 79 discussion  162–8 neg-tags  187–92
distributional analysis  see also I dinnae think; interrogative tags  173, 179, 187,
100–7, 101f, 102f, 104 f I divn’t think; i don’t 207
multivariate analysis  think so; I doØ think; interruptions  44
108, 112, 113t I think; ‘I think not-p’ intersubjective function
discussion  119–20 i don’t think so  133, 165 i don’t know  83
274 The Structure of Discourse-Pragmatic Variation

i don’t think  140–3, 144 Corpus (BNC); Linguistic neg-tags see negative polarity
neg-tags  189–91, 206, 208 Innovators Corpus (LIC); question tags
involvement inducers: neg-tags London-Lund Corpus negative particle/negative
190–1 London-Lund Corpus  127 auxiliary variants
isn’t it  182, 183 English  55
distributional analysis  M evolution of negation in
194–202, 195t, 197f, 198t, Macafee, Caroline  176 English  52
199f, 201f Macaulay, Ronald K. S.  3, 14, 34, extra-linguistic distribution
multivariate analysis  203 39, 56, 58, 227, 231 of variants  56–7
previous research  171, 174–6 Mauranen, Anna  80–1 geographical distribution and
Ito, Rika  230, 235 Meehan, Teresa  16 categorisation  55–6, 56t
Meillet, Antoine  35–6, 40 intra-linguistic distribution
J methodology  21–49 of variants  57–8
Janda, Richard  38 conversation analysis  41–8 previous research  56–8
Jespersen, Otto  52 data  22–7 Scottish varieties  53, 55
Johnstone, Barbara  6 discourse-pragmatic variables see also BwE verb negation
Jucker, Andreas H.  15 5, 28–32 system
just  80–1, 86 grammaticalisation  35–41 negative polarity question tags
quantitative methods  32–4 (neg-tags)  31, 65, 169–71,
K variationist sociolinguistics 178–212
Kaltenböck, Gunther et al.  27–35 coding 185–6, 193t
8, 36n see also analytic methods; conduciveness  173, 177,
Kärkkäinen, Elise  136 qualitative data analysis; 188–90, 192–3, 193t, 200,
König, Ekkehard  39 quantitative data analysis 202–5, 206–7
Kortmann, Bernd et al.  14 Metzger, Terri R.  72 distributional analysis 
Kotthoff, Helga  139 Midlands varieties  134 194–202, 195t, 197f, 198t,
Kristiansen, Tore  11n Millar, Martin  56, 67 199f, 201f, 202f
Krug, Manfred  174–5, 177, 184 Miller, Jim  15, 56, 57–8 frequency  34
Milroy, James et al.  13, 55, 120 involvement inducers  190–1
L Milton Keynes  176, 210 multivariate analysis  202–5,
Labov, William  8–9, 11, 12, 13, 15, Moore, Emma  174 204t
23, 27, 28, 224, 234 Mulac, Anthony  38 non-paradigmatic neg-tags
Lakoff, Robin  172, 174 multivariate analyses  32–3 180
Lange, Deborah  15, 16, 39 paradigmatic neg-tags 
Lavandera, Beatriz R.  9, 29, 30 N 179–80
learner speakers: i don’t know nae  53n previous research  171–7
73–4 -nae  51, 54 qualitative analysis  186–93,
Lehmann, Christian  38, 39, 40 distributional analysis  269–70
Levey, Stephen et al.  7, 16, 34n, 59–62, 59f, 61t, 62t quantitative analysis  193–205
230 extra-linguistic distribution semi-paradigmatic neg-tags
Levinson, Stephen C.  45, 96 56–7 180
like  7, 9, 12, 15, 16, 39, 232 geographical distribution  the variable and envelope of
Lindemann, Stephanie  80–1 55, 56t variation  179–81
linguistic innovation  11 intra-linguistic variation variants  182–5, 185t
Linguistic Innovators Corpus 57–8 discussion  206–11
(LIC)  177, 230 multivariate analysis  63, 64t see also question tags:
Llamas, Carmen  23, 24 -na(e)  53, 54 previous research
London  11, 30, 184 see also -nae New Zealand English  14, 234
see also Bergen Corpus of native vs. learner speakers: next-turn proof procedure  43
London Teenage Language i don’t know  73–4 no  51, 53, 54
(COLT); British National
 Index 275

distributional analysis  59–62, productive constructions  51–2, Schiffrin, Deborah  5, 45, 144
59f, 61t, 62t 59 Schleef, Erik  15
extra-linguistic variation  56 prosodic features  45, 72 Schourup, Lawrence  5
geographical distribution  Scottish Corpus of Text and
55, 56t Q Speech  55
intra-linguistic variation  qualitative data analysis  44–5, Scottish varieties  53n
57, 58 48 Ayrshire  56, 58
multivariate analysis  63, 64t conversation analysis  42–5 Buckie  56–7, 58
in neg-tags  194–5, 195t functional domains  45–7 Edinburgh  56, 57
normalised frequencies  34 quantifying multifunctionality evolution of verb negation  53
Northern Ireland Transcribed 47 I dinnae ken  79
Corpus of Speech  58 validating qualitative analyses in’t it  176, 184
Northumbrian dialect  24, 55 48 Lowland dialect  24
not, -n’t  53–4 qualitative discourse studies  Shetland  58
distributional analysis  15, 16 Ulster  55, 57, 58
59–62, 59f, 61t, 62t quantitative data analysis  10–16 semantic bleaching  37, 39
evolution of  52 quantitative methods  27–8, semantic equivalence  9, 28,
geographical distribution  32–4 29–30
55, 56t question tags: previous research sequential implicativeness  43
see also isn’t it 169–77 sex-differentiated patterns 
Nuyts, Jan  126 qualitative studies  172–3 12, 13, 29
quantitative studies  174–7 see also social factors
O see also negative polarity Simon-Vandenbergen, Anne-
Östman, Jan-Ola  5 question tags (neg-tags) Marie  8, 126, 127, 136
overlaps  44 Quirk, Randolph et al.  52n, 60n, Smith, Jennifer et al.  56–7
Oxford English Dictionary 94, 95 Smith, Sara W.  15
(OED)  79, 119, 209 quotatives  30, 31n Social Agreement Principle
(SAP)  159, 161, 164
P R social factors  10–13
paralinguistic features  45 Rbrul  32–3, 108, 156 i don’t know  70, 76, 80,
Persson, Gunnar  126 Reading  176, 177, 207, 210 101–5, 108, 110, 112–13,
Petyt, K. M.  134 recipient design  43 115–16, 120–1, 218–19
phonological attrition  37 Redeker, Gisela  144 i don’t think  129, 130,
phonological variables  12 relevance theory  30 135–6, 151–3, 155, 156–9,
Pichler, Heike  3, 16, 21, 30, 31n, repair-marking  47 164, 165, 218–19
34n, 207, 208, 231 i don’t know  85, 91–2 neg-tags  185, 196, 200, 202,
Plug, Leendert  136 repetitions  44 202f, 203–4, 204t, 208–9,
Podesva, Robert  174 Rickford, John et al.  30 219–20
politeness  45 Romaine, Suzanne  15, 16, 29, 39 negative auxiliary variation
neg-tags  191, 192, 208 Rudolph, Elisabeth  14n 56–7, 65
question tags  172 question tags  174–5, 176–7
Poplack, Shana  34 S speaker sample  25–6
Potter, Jonathan  71 Sankoff, David  27 discussion  217, 222–4
pragmatic particles  5 Sankoff, Gillian  29 sociolinguistic interviews  23, 25
pragmaticalisation  40 SAP see Social Agreement Sowa, Joseph  14
preference organisation  43 Principle speaker sample  25–6, 25t
principle of accountability  15, 28 Schegloff, Emanuel A.  72 Sperber, Dan  30
principle of multiple causes  32 Scheibman, Joanne  69, 70, 74–5, Steele, Hazel  58
principle of persistence  93, 176 80, 106, 117, 118, 124, 128–9, Stenström, Anna-Brita  80–1
procedural meaning  30 130, 134, 135, 136, 150, 163, 220 Stubbe, Maria  209, 230
276 The Structure of Discourse-Pragmatic Variation

Stubbs, Michael  7 topic-development  47 verb negation, evolution of  52–3

subjective function  39, 46 i don’t know  72, 85–7, 90–1 see also BwE verb negation
i don’t know  82–3 i don’t think  146–7 system
i don’t think  127, 137–40, topic-proffering sequences  43
144–5, 163 Torgersen, Eivind et al.  W
neg-tags  187–9, 206 177, 207, 208 Walker, James  32n
suppose see I don’t suppose Toronto English  12, 14, 16, 230, Watt, Dominic  24
Survey of Regional English  23 232 Watts, Richard J.  136
syntacticisation of word order Torres Cacoullos, Rena  32n Wauchope, Mary Michele  86
40 Tottie, Gunnel  189, 207 Weatherall, Ann  72
transition marking  47 Weinert, Regina  15
T Traugott, Elizabeth Closs  7–8, Weinreich, Uriel et al.  3
tag questions (TQs) see negative 35, 36–7, 39, 39n, 46, 120, 207 wh-words  97–8, 265
polarity question tags Trousdale, Graeme  120 will  32n, 52n, 54
(neg-tags); question tags: Tsui, Amy B. M.  71–2 Wilson, Deirdre  30
previous research turn-exchange  43, 47 winnae  54
Tagliamonte, Sali  12, 16, 34, i don’t know  72, 73, 85, Wooffitt, Robin  71
230, 235 87–92 wouldnae see -nae
temporal development of Tyneside dialect  24, 55, 57, 234
interactions  44 Y
Terkourafi, Marina  30, 31n, 236 V Yaeger-Dror, Malcah  159, 161,
textual mode  45–7 Van Bogaert, Juli  131n 164
i don’t know  72, 73, 74, 85, variationist sociolinguistics  York  14, 230
87, 96, 97, 98, 99t, 105, 117 3, 27–35 Yorkshire varieties  129, 130,
i don’t think  127–8, 144, and discourse-pragmatic 134, 164
149t features  3–4, 10–11, 222–4 you know  11, 29
textual tokens  46 discourse-pragmatic variables you see  29, 39
th-fronting  12 28–32 Young, Richard  32
thetical grammar  8, 36n premises  27–8 Yule, George  7
Thompson, Sandra A.  38, 72, 118 principal tenets  3
token frequency  228–30 quantitative methods  27–8,
32–4