A Sociolinguistic Survey of (t,d) deletion, (t) glottaling,

and their intersection in East Anglian English

Carmen Ciancia

A thesis submitted for the degree of Doctor of Philosophy

Department of Language and Linguistics

University of Essex

November 2020
To my beloved grandmother, Angelina, to whom I promised
to dedicate this thesis before she left this world.

You will always be in my heart,

in my mind.
Table of Contents
List of Tables…………………………………………………………………………vii

List of Figures………………………………………………………………………...ix

Abstract……………………………………………………………………………….xi

Acknowledgements…………………………………………………………………xiii

Structure of the thesis……………………………………………………...…………xv

Chapter 1 – INTRODUCTION……………………………………………………...17
1.1 Motivations to examine (t,d) and (t) in East Anglian English ......................17

1.2 East Anglian history and borders ..................................................................18

1.3 Urban areas and speech communities ...........................................................22
Colchester ..............................................................................................23
Ipswich ...................................................................................................25
Norwich .................................................................................................28
1.4 Broad research questions ..............................................................................29
1.5 Summary .......................................................................................................32
Chapter 2 - FIELDWORK AND METHODS .......................................................33
2.1 The sample ....................................................................................................33
2.2 The Fieldworker “away from home” ............................................................36
2.3 Entering the communities .............................................................................37
2.4 The sociolinguistic interview ........................................................................42
2.5 Early field techniques....................................................................................45
2.6 Data analysis instruments..............................................................................47
2.7 Dependent variables ......................................................................................49
(t,d) Deletion ..........................................................................................50
(t) Glottaling ..........................................................................................51
2.8 Independent variables ...................................................................................52
2.9 Summary .......................................................................................................61
Chapter 3 - (t,d) DELETION: LITERATURE REVIEW ..........................................62
3.1 Articulatory description of the plosives /t, d/ .....................................................62
3.2 A brief note on terminology ...............................................................................63
3.3 Description of the (t,d) variable ........................................................................63

iii
 3.4 Neutral contexts .................................................................................................65
3.5 (t,d) in Sociolinguistic theory.............................................................................67
3.6 A brief diachronic development of (t,d)............................................................68
3.7 Empirical findings from North American studies ..............................................70
3.7.1 (t,d) in AAVE ........................................................................................71
3.7.2 (t,d) in US English dialects..........................................................................74
3.7.3 (t,d) in varieties influenced by Spanish: Chicano English and Tejano
English…………....................................................................................................78
3.7.4 (t,d) in Creole varieties: mesolectal Jamaican Creole and Standard
Jamaican Creole...................................................................................................79
3.7.5 (t,d) explored on theoretical grounds .....................................................82
3.8 (t,d) in British English dialects .....................................................................90
3.8.3 York .......................................................................................................90
3.8.4 Manchester.............................................................................................92
3.8.5 Northeast of England .............................................................................94
3.8.6 Mersea Island .........................................................................................95
3.9 (t,d) in Standard British English ...................................................................96
3.10 Summary .......................................................................................................97
Chapter 4 - (t,d) DELETION: RESULTS AND DISCUSSION ..............................100
4.1 Analytical procedure - linguistic constraints ..............................................100
4.2 Analytical procedures – Rbrul ....................................................................104
4.3 Overall Results – model with glottal variants included in the (t,d) dataset 106
4.4 Overall Results – model with no glottal variants included in the (t,d)
dataset….................................................................................................................109
4.4.1 Following environment ........................................................................113
4.4.2 Voicing agreement ...............................................................................117
4.4.3 Style-shifting........................................................................................119
4.4.4 Morphological Class ............................................................................121
4.4.5 Preceding segment ...............................................................................129
4.4.6 Non-significant predictors – Overall Results ......................................133
4.5 Treating Colchester, Ipswich and Norwich separately ...............................135
4.5.1 Results from Colchester .......................................................................135
4.5.2 Results from Ipswich ...........................................................................140
4.5.3 Results from Norwich ..........................................................................143

iv
 4.5.4 Interaction between locality and other predictors ................................147
4.6 Summary .....................................................................................................149
Chapter 5 - (t) GLOTTALING: LITERATURE REVIEW ....................................151
5.1 Articulatory description of the plosive /t/ and its variant [Ɂ] ......................151
5.2 T-Glottalisation in Phonological Theory ....................................................155
5.3 T-Glottalisation in Variationist Sociolinguistic Theory..............................159
5.4 Diachronic development of [Ɂ] ...................................................................162
5.5 T-Glottalisation in Southern England .........................................................166
5.5.1 East Anglia (1889 - 2007)....................................................................167
5.5.2 London vernacular (1938 - 2013) ........................................................172
5.6 T-Glottaling in Cardiff ................................................................................175
5.7 T-Glottaling in Northern England ...............................................................176
5.8 T-Glottaling in Scotland..............................................................................179
5.9 T-Glottaling beyond the British Isles. .........................................................181
5.10 Summary .....................................................................................................182
Chapter 6 - (t) GLOTTALING: RESULTS AND DISCUSSION .......................185
6.1 Analytical procedure – linguistic constraints ...................................................185
6.2 Analytical procedure – Rbrul ......................................................................188
6.3 Overall Results and Individual Variation for word-final /t/ ........................189
6.3.1 Preceding environment - word-final /t/................................................191
6.3.2 Style-shifting in word-final /t/ .............................................................195
6.3.3 Following environment in word-final /t/ .............................................198
6.3.4 Non-significant factors – overall findings for word-final /t/ ...............201
6.4 Treating the three localities separately........................................................204
6.4.1 Results from Colchester – word-final /t/ .............................................205
6.4.2 Results from Ipswich – word-final /t/ ..................................................207
6.4.3 Results from Norwich – word-final /t/.................................................209
6.5 Overall results and individual variation for word-medial /t/.......................212
6.5.1 Non-significant factors – word-medial /t/............................................226
6.6 Treating Colchester, Ipswich and Norwich separately. ..............................226
6.6.1 Results from Colchester – word-medial /t/ ..........................................226
6.6.2 Results from Ipswich – word-medial /t/ ..............................................228
6.6.3 Results from Norwich – word-medial /t/ .............................................230

v
 6.7 Summary .....................................................................................................232
Chapter 7 - INTERSECTION BETWEEN (t) DELETION AND (t) GLOTTALING
IN WORD-FINAL CONSONANT CLUSTERS: RESULTS AND DISCUSSION 234
7.1 Terminological Remarks .............................................................................234
7.2 Status of the two phonological variables in East Anglia ............................235
7.2.1 Research questions for intersection .....................................................236
7.3 Covariation between linguistic variables ....................................................236
7.4 A brief account of lenition ..........................................................................239
7.5 A brief account on rule ordering – feeding and bleeding ...........................243
7.6 Analytical procedure ...................................................................................245
7.7 Overall Results and Individual Variation....................................................249
7.7.1 Preceding phonetic segment ................................................................251
7.7.2 Following phonetic segment ................................................................253
7.7.3 Stress on the following syllable ...........................................................255
7.7.4 Style-shifting........................................................................................257
7.7.5 Sex .......................................................................................................258
7.7.6 Syllable stress ......................................................................................259
7.8 Summary .....................................................................................................260
Chapter 8 - CONCLUDING REMARKS AND FUTURE WORK.........................261
REFERENCES ..........................................................................................................268
Appendix I .................................................................................................................294
Appendix II ................................................................................................................296

vi
List of Tables
Table 1.3.1 Census 2011 - Employment and unemployment data in Colchester, England
and Wales (Nomis, 2011). ...........................................................................................25
Table 1.3.2 Census 2011 - Employment and unemployment data in Ipswich, England
and Wales (Nomis, 2011). ...........................................................................................27
Table 1.3.3 Census 2011 - Employment and unemployment data in Norwich, England
and Wales (Nomis, 2011). ...........................................................................................29
Table 2.1.1 . Fieldwork sample design employed in the East Anglian study. .............35
Table 2.8.1 A comparison between Zipf values and fpmw, adapted from SUBTLEX-
UK website (van Heuven et al., 2014).........................................................................57
Table 2.8.2 Collapsing ESeC classes. Adapted from Rose et al. (2010). ....................58
Table 3.10.1 (t,d) deletion: Comparison of seven studies.. ........................................97
Table 4.2.1 Constraints of the (t,d) variable in the East Anglian dataset, with glottals
included......................................................................................................................105
Table 4.3.1 Multivariate analysis of (t,d) in East Anglia, including glottal variants.107
Table 4.4.1 Constraints of the (t,d) variable in the East Anglian dataset, without
glottals........................................................................................................................110
Table 4.4.2 Multivariate analysis of (t,d) in the whole dataset..................................112
Table 4.4.3 Regression analysis of morphological class. ..........................................122
Table 4.4.4 Estimates of the value of pr in the East Anglian dataset. .......................127
Table 4.5.1 Multivariate analysis of (t,d) deletion in Colchester. .............................138
Table 4.5.2 Multivariate analysis of (t,d) deletion in Ipswich. ..................................142
Table 4.5.3 Multivariate analysis of (t,d) deletion in Norwich. ................................145
Table 5.1.1 Phonetic variants and glottal variants of /t/. ...........................................155
Table 5.3.1 Most frequent and least frequent environments for t-glottaling to occur.
...................................................................................................................................161
Table 6.2.1 Constraints on the (t) variable for the WF and WM dataset. ..................189
Table 6.3.1 Multivariate analysis of (t) glottaling in the whole dataset. ...................191
Table 6.4.1 Multivariate analysis of WF /t/ glottaling in Colchester. .......................206
Table 6.4.2 Multivariate analysis of word-final /t/ glottaling in Ipswich. .................209
Table 6.4.3 Multivariate analysis of word-final /t/ glottaling in Norwich.................210
Table 6.5.1 Multivariate analysis of (t) glottaling in the whole WM dataset. ...........217
Table 6.6.1 Multivariate analysis of WM /t/ in Colchester. ......................................227

vii
Table 6.6.2 Multivariate analysis of WM /t/ in Ipswich. ...........................................229
Table 6.6.3 Multivariate analysis of WM /t/ in Norwich. .........................................231
Table 7.6.1 Constraints on the intersection between (t) deletion and (t) glottaling. ..248
Table 7.7.1 Multivariate analysis of /t/ deletion vs. /t/ glottaling. .............................250

viii
List of Figures

Figure 1.1.1 Geographical location of Colchester, Ipswich, and Norwich. .................18

Figure 1.2.1 Map representing linguistic East Anglia and the transition zone. Adapted
from Trudgill (2001, 2018). .........................................................................................22
Figure 2.8.1 /t/ deletion after the alveolar /n/ and before a following pause. ..............54
Figure 2.8.2 Example of /t/ deletion: transition to the following phonetic segment. ..55
Figure 2.8.3 Example of /t/ glottaling followed by a vowel. .......................................55
Figure 3.8.1 Word frequency from SUBLEX-UK in Manchester (Baranowski &
Turton, 2020). ..............................................................................................................94
Figure 4.4.1 Overall results in the East Anglian dataset............................................109
Figure 4.4.2 /h/ in hammer. Adapted from Trudgill (1974). .....................................114
Figure 4.4.3 Rates of deletion before following [h]. .................................................115
Figure 4.4.4 Rates of deletion before dropped /h/. ....................................................115
Figure 4.4.5 Heterovoiced tokens in the preceding segment. ....................................118
Figure 4.4.6 Rates of deletion across different speech styles. ...................................119
Figure 4.4.7 (t,d) deletion by style and morphological class. ....................................120
Figure 4.4.8 Deletion rates by morphological class………………………………. .121
Figure 4.4.9 Rates of deletion by morphological class in East Anglia, York and
Manchester.................................................................................................................124
Figure 4.4.10 Rates of retention by morphological class and word-frequency. ........127
Figure 4.4.11 Morphological class by individual speakers' age. ...............................128
Figure 4.4.12 Interaction between morphological class and age ...............................129
Figure 4.4.13 Preceding environment in East Anglia: comparison between different
datasets.......................................................................................................................130
Figure 4.4.14 Rates of deletion according to the preceding environment in East Anglia,
York, and Manchester................................................................................................132
Figure 4.4.15 Word frequency from SUBTLEX-UK and percentages of deletion. ..134
Figure 4.5.1 Interaction between preceding environment and locality. ....................147
Figure 4.5.2 Probabilities for the following environment in Colchester. ..................148
Figure 4.5.3 Probabilities for the following environment in Ipswich........................148
Figure 4.5.4 Probabilities for the following environment in Norwich. .....................148
Figure 4.5.5 Morphological class across the three East Anglian localities. ..............149

ix
Figure 5.4.1 Evidence of glottaling in the Linguistic Atlas of England. Adapted from
Orton et al. (1978)......................................................................................................164
Figure 5.5.1 Distribution of glottal variants. Adapted from Trudgill (1974). ...........169
Figure 6.3.1 Overall results for word-final /t/ glottaling. ..........................................190
Figure 6.3.2 Probability of WF /t/ glottaling between EA and RP. ...........................193
Figure 6.3.3 Rates of /t/ glottaling for the preceding environment. ..........................195
Figure 6.3.4 Rates of (t) glottaling by three speech styles.........................................196
Figure 6.3.5 WF (t) glottaling by preceding environment in casual speech ..............197
Figure 6.3.6 Rates of (t) glottaling by following environment. .................................199
Figure 6.3.7 Rates of (t) glottaling by following environment in casual speech. ......201
Figure 6.3.8 Rates of WF /t/ glottaling by individuals' age. ......................................202
Figure 6.3.9 Rates of word-final /t/ glottaling by class and age. ...............................203
Figure 6.4.1 Restrictions on the application of (t) glottaling. Adapted from Trudgill
(1974). ........................................................................................................................212
Figure 6.5.1 Overall results for word-medial /t/ glottaling. .......................................213
Figure 6.5.2 Rates of (t) glottaling by syllable type. .................................................214
Figure 6.5.3 Rates of WM /t/ glottaling by preceding environment and sex, ............218
Figure 6.5.4 Rates of WM /t/ glottaling by preceding environment and age. ...........219
Figure 6.5.5 Rates of /t/-glottaling by following environment and sex .....................222
Figure 6.5.6 Rates of WM /t/-glottaling by following environment and age ............222
Figure 6.5.7 Rates of WM /t/- glottaling by sex and age...........................................224
Figure 6.5.8 Percentages of WM /t/ by individual speaker. ......................................225
Figure 7.7.1 Distribution of /t/ across the lenition scale……………………………249
Figure 7.7.2 Probability of deletion according to the preceding phonetic segment..251
Figure 7.7.3 Deletion by preceding phonetic segment and sex…………………….252
Figure 7.7.4 Probability of deletion according to the following phonetic segment..253
Figure 7.7.5 Deletion by following phonetic segment and sex…………………….254
Figure 7.7.6 Probability of deletion by stress on the following syllable…………...256
Figure 7.7.7 Probability of deletion in the C(C)t linguistic context, across style…..257
Figure 7.7.8 Probability of deletion by sex of participants…………………………258
Figure 7.7.9 Rates of deletion by style and sex…………………………………………259
Figure 7.7.10 Probability of deletion by syllable stress…………………………….260

x
Abstract

This thesis examines two well-studied phonological features - (t,d) deletion and (t)

glottaling – in East Anglian English by maintaining the structuralist roots of the

variationist paradigm (e.g. Wolfram 1993; Patrick 1999). It also investigates, for the

first time, the covariation between the two linguistic variables by exploring the

intersection of (t) deletion and (t) glottaling in word-final consonant clusters (e.g.

different). (t,d) deletion has been largely investigated in US English dialects, yet it has

received comparatively little attention in the UK. (t) glottaling has been widely

examined as a change in progress in England (including Norwich, Trudgill, 1974,

1988) and Scotland, yet little research on this variable has been carried out in Ipswich

(Straw & Patrick, 2007) or Colchester.

Data was gathered in three East Anglian speech communities: Colchester, Ipswich

and Norwich, where 36 participants, equally distributed, have been recorded by means

of sociolinguistic interviews, reading passages and word lists. Mixed-effects Rbrul

regression analysis was carried out.

(t,d) results are in line with previous US studies showing that (t,d) absence is

primarily conditioned by linguistic factors and its profile is that of a stable variable. A

more fine-grained analysis is suggested for the following phonological environment.

For (t) glottaling, this thesis also proposes a closer inspection of the following

phonological environment. The preceding phonological context - little explored in

previous studies - plays a notable role. While word-final /t/ glottaling has completed its

social change and is spreading in phonological space even in environments where it

used to be blocked, word-medial /t/ is both phonetically and socially conditioned.

xi
 The covariation between (t) glottaling and (t) deletion shows that the transition

glottaling → deletion, in the lenition scale, is in feeding order and is mostly linguistically

driven. In this analysis, women exhibit a higher use of glottal variants, whereas males

promote deletion – the last stage of the lenition scale

xii
Acknowledgements

I would like to gratefully acknowledge the invaluable assistance that the following

people provided during my PhD. First and foremost, I wish to express my deepest

gratitude to my supervisor, Professor Peter Patrick, for his invaluable advice and

support throughout the whole process. I am very grateful for his guidance and

assistance at every step of this research - I have learned a lot from his vast expertise.

Thanks to my examiners, Dr. Maciej Baranowski (University of Manchester) and Dr.

Ella Jeffries (University of Essex). I would like to thank Prof. Peter Trudgill, Prof.

Gregory Guy and Prof. Josef Fruehwald for their suggestions. Thanks also to the

attendees of ICLaVE-10, UKLVC-12 and NWAV-48 for their thoughtful comments. I

am also grateful to all of the Department faculty members for their help offered at

different stages of this research: Prof. Monika Schmid, Prof. Enam Al-Wer, Dr. Ella

Jeffries, Dr. Laurel Lawyer, Dr. Yuni Kim and Dr. Rebecca Clift. I am particularly

grateful for the assistance given by Dr. Uri Horesh with Rbrul. I would like to thank

Kerri Butcher and her cousin, Hannah Buckle and David Provan, for their assistance

with the collection of part of my data. I must also gratefully acknowledge all the East

Anglian people who volunteered to participate in this project. My greatest thanks must

go to my parents, Mariolina and Antonio, for putting me through the best education

possible and for their constant encouragement and support throughout the completion

of this work. My special thanks also go to my brother, Giuseppe, whose immense love

has helped me to get through the difficult times. To my fiancé, Francesco, for his

constant and invaluable support, for his patience and understanding, and for his

unconditional love. I am truly grateful to my grandparents, Angelina, Carmela,

Giuseppe and Pietro, for always believing in me and for their love and encouragement.

xiii
Thanks to my aunt Titina, aunt Lina; to my mother-in-law Maria, and to my cousins

Rosamaria, Tuglio, Costantino. Finally, my life during PhD would not have been

memorable without my special friends and colleagues: Amanda Cole and Siham

Rouabah, who have also helped me throughout this journey. And thanks to my lovely

flatmate Paloma Carretero Garcia.

xiv
Structure of the thesis

This thesis begins with an introductory Chapter which outlines the purpose of this

work; it explains the motivations to study East Anglian English, starting with a

discussion which aims at defining what is meant by linguistic East Anglia. It provides

an insight into Colchester, Ipswich and Norwich from which the data has been collected

and addresses the research questions of the present survey. Finally, this Chapter briefly

summarises the origin of the sociolinguistic paradigm and reviews the use of social

variables within the sociolinguistic theory.

Chapter 2 focuses on fieldwork and the methods employed in the present study, with

considerations on the coding procedure. It provides a brief outline of the two dependent

variables - (t,d) deletion and (t) glottaling – as well as a brief outline of the internal and

external constraints employed in the present research.

Chapter 3 presents a phonetic, phonological, and sociolinguistic description of the (t,d)

variable tracing also its diachronic profile. It selectively reviews the empirical findings

from North American (t,d) studies along with systematic research carried out in the

UK, and it reviews (t,d) deletion on theoretical grounds.

Chapter 4 presents and discusses results of (t,d) deletion in East Anglian English.

Firstly, the overall results from the three localities (Colchester, Ipswich and Norwich)

are examined together; secondly, the three urban areas are explored separately in order

to investigate the behaviour of the independent variables and their related patterns in

each locality.

Chapter 5 reviews the literature of (t) glottaling and its diachronic development, starting

with a phonetic and phonological description of /t/. This variable is explored within

xv
phonological and sociolinguistic theories, with a summary of empirical findings in the

UK and in other English-speaking areas (e.g. in the USA).

Chapter 6 presents and discusses the findings of (t) glottaling in East Anglia. Initially,

the overall results (the three localities together) are investigated; secondly, the three

localities are analysed separately.

Chapter 7 explores (t) deletion and (t) glottaling at their intersection in word-final

consonant cluster. It reviews the concept of co-variation between phonological

variables as well as the notion of lenition. Finally, it presents and discusses the overall

results from the three urban areas.

Chapter 8 summarises the findings, draws conclusions and summarises the research

questions which are answered throughout this work.

xvi
Chapter 1 - INTRODUCTION

This work is a synchronic survey of urban East Anglian English and it is rooted in the

sociolinguistic paradigm. Urban sociolinguistic studies set out to describe language as

a social phenomenon, attempting to establish relationships between language and

society, and “pursuing the complementary questions of what language contributes to

making community possible and how communities shape their languages by using

them” (Coulmas, 1997: 2). The two variables selected for this survey are two well-

studied phonological features: (t,d) deletion in word final consonant clusters, which has

received wide attention in many US English dialects, and (t) glottaling, which has been

largely explored in the UK. Through the employment of quantitative methods, this

study attempts not to reach conclusions based only on single details, but also takes into

account patterns of variation distinguishing “the shape of the forest through all the

trees” as Guy (2014: 196) suggests.

This research project, which is carried out in the South East of England –

precisely in East Anglia - is largely inspired by the Labovian framework, which laid

the basis for variationist studies (e.g. Trudgill (1974) in Norwich).

1.1 Motivations to examine (t,d) and (t) in East Anglian English

The East Anglian English variety has been widely investigated by Trudgill (1974, 1988,

1999, 2003) over the years. While most of his systematic work is devoted to a particular

region – Norfolk1 - little systematic research has been carried out in Suffolk (Kokeritz

1
It should be noted that the borders of Norfolk, or more precisely of Norfolk dialect as intended here,
do not line up with political borders. Indeed, linguistically, its definition also includes some northern
Suffolk towns, such as Southwold and Lowestoft (Trudgill, 2004a).

17
1932; Potter, 2018) or Essex (Altendorf, 2003). Britain (2020: 14) has recently claimed

that despite a few multilocality studies, “no research has been able to provide a picture

of the state of the traditional dialect across the whole [East Anglian] region”. This

study provides a contribution in this respect as it investigates (t,d) deletion and (t)

glottaling in the speech communities of Colchester (Essex), Ipswich (Suffolk), and

Norwich (Norfolk), as illustrated in figure 1.1.1. Before reviewing the concept of

‘speech community’, let us firstly define East Anglia in both geographical and

linguistic terms.

Figure 1.1.1 Geographical location of Colchester, Ipswich, and Norwich.

1.2 East Anglian history and borders

The constitution of East Anglia as we know it today seems to date back to the fifth

century when Angles, Saxons and Jutes, from the north-west European continent,

began their settlement in the British Isles. On their arrival, the Angles who occupied

the east part of Britain mainly settled in Norfolk and Suffolk, including the bordered

areas to the south and to the west; whereas, “Essex and Cambridgeshire – were distinct

from the very beginning by reason of being mainly Saxon” (Trudgill, 2003: 23). During

the Roman invasion, the north Essex town of Colchester was the capital of Britain and
18
the queen of the British Celtic Iceni tribe was Boudica (Finchman, 1976). The Iceni, or

the Brittonic tribe of eastern Britain, included the current county of Norfolk, parts of

Suffolk and Cambridgeshire as well as neighbouring areas such as Corieltauvi to the

west (the current East Midlands), Catuvellauni (the current Bedfordshire,

Buckinghamshire, and Hertfordshire) and Trinovantes (the current north Essex and

south Suffolk). The first significant battle in East Anglian history was documented

around AD 61 when the uprising of the East Anglian Iceni tribe against the Romans

was taking place - the latter ended with the death of the queen Boudica (Finchman,

1976).

Martin (1999) reports that the names “Suffolk” and “Norfolk” were first

recorded in the 1040s (cf Dymond & Northeast 1985). These two place names, which

referred to the self-governing area in the British Isles, literally mean “the north/south

folk of the Kingdom of East Anglia”, yet it is unknown when the two current counties

became separate. It is assumed that this might have happened during the 5th - 7th

centuries when the Germanic tribes settled in Britain. The dialectal similarities between

Suffolk and Norfolk could be explained by the fact that, originally, the Kingdom of

East Anglia was considered one cultural and linguistic area (Trudgill, 2003). However,

Trudgill (2003) adds that cultural and dialectal differences between these two entities

subsequently surfaced owing to difficulties to traverse the land around the River

Waveney – a barrier of communication between the ‘north folk’ and ‘south folk’ which

limited the dialectal contact. In his commentary on boundaries of East Anglia, Trudgill

(2004a: 163) reveals that the borders of this area are still a debated issue:

“As a modern topographical and cultural term, East Anglia refers to an area
with no official status. Like similar terms such as “The Midlands” or “The
Midwest”, the term is widely understood but stands for an area which has
no clear boundaries. Most people would agree that the English counties of
Norfolk and Suffolk are prototypically East Anglian, although even here
the status of the Fenland areas of western Norfolk and north western

19
 Suffolk is ambiguous: the Fens were for the most part uninhabited until the
17th century, and the cultural orientations of this area are therefore less
clear. The main issue, however, has to do with the extent to which the
neighbouring counties, notably Cambridgeshire and Essex, are East
Anglian or not.”

As he explains, there seems to be no hesitation in considering Norfolk and Suffolk a

part of East Anglia - two counties which embrace all the traditional East Anglian

features (see footnote 4) and meet the North Sea. The issue is how to interpret the

western and southern borders (Martin, 1999). As pointed out by Britain (2001) the Fens

were largely unpopulated until lately, hence its status made it onerous to assess.

Trudgill (2001) adds that if the Fenland were not to be included in East Anglia, even

the Fens which are part of Norfolk will have an ambivalent status. Moreover, he

explains that if the label “East Anglia” comprises the Fens, then Lincolnshire,

Cambridgeshire, Huntingdonshire2, and the Soke of Peterborough3 should also be

included. Other historical sources (e.g. Wilson 1977) also include Essex, Bedfordshire

and Hertfordshire; this means that the status of these counties should also be taken into

account (Trudgill, 2001).

To shed light on which neighbouring counties belong to this area, Trudgill

(2001a) draws isoglosses which illustrate the distribution of seven linguistic features4

by using data from The Survey of English Dialects (SED). His map (p. 11) demonstrates

that the geographical extent of what he calls ‘linguistic East Anglia’ can be divided into

2
Currently, this area is part of Cambridgeshire (Trudgill, 2001).
3
Currently, this area is part of Cambridgeshire, but historically belonged to Northamptonshire (Trudgill,
2001).
4
The seven features illustrated include (1) verbal -s marking (also referred to as third person singular
zero); (2) lack of /h/ dropping; (3) presence of glottaling and glottalisation; (4) the realisation of the
BATH vowel with the open front [a:], as in bath; (5) the realisation of the NURSE vowel as /ɐ/, in words
like church; (6) short /ɪ/ realised with the schwa, as in suet; and (7) the realisation of the GOAT vowel
with /ʊ/ in road and both. With respect to the GOAT vowel, Trudgill (2004a) notes that words like rowed
are realised as [ʌu] in the northern region, whereas words like road are realised as [uː~ʊ]; in the southern
region road and rowed are both commonly articulated as [ʌu]. See Butcher (2019) for recent research
on the GOAT vowel in Suffolk.

20
a core zone surrounded by nearly all Norfolk and Suffolk, whose boundaries are

delimited to south and west by a transition zone. The latter is partly marked by the

Fens5 and represents both a geographical and linguistic transition. Britain (2013)

observed that linguistic features which represent a transition from East Anglia to the

East and Midlands include /h/ dropping, with the dropped voiceless fricative to the west

and Midlands, and present [h] in East Anglia; /ɪ/ in unstressed syllables is articulated

as [ɪ] to the west, while to the east it is realised as a schwa; the MOUTH vowel is realised

with the open-mid form [ɛː] to the west, similarly to the Midlands, whereas to the east

is realised as [ɛu]; the typical East Anglian verbal -s marking was found not to be

present in dialects spoken in the Fens. Not only does the Fenland function as transition

area between East Anglia and the Midlands, but it sets linguistic boundaries between

the north and the south of England, such as the TRAP- BATH split, and the FOOT- STRUT

split.6 Having highlighted some of the most notable differences between west,

Midlands, and east of England, let us turn to the starting point: establishing the East

Anglia boundaries; according to Trudgill (2001a: 10):

“East Anglia, from a linguistic perspective, consists of all of Norfolk and

Suffolk apart from the Fens, and part of northeastern Essex. The
transitional area consists of the Norfolk, Suffolk and Cambridgeshire Fens
plus most of the rest of Cambridgeshire, central Essex, and a small area of
northeastern Hertfordshire.”

5
See Britain (2001, 2013) for further details on linguistic variation in the Fens.
6
See Britain (2001, 2013) for further details.

21
Figure 1.2.1 Map representing linguistic East Anglia and the transition zone. Adapted
from Trudgill (2001, 2018).

North Essex, as Trudgill (2004a, 2018) explains, includes the town of Colchester and

its neighbouring area. This survey employs the above definition of linguistic East

Anglia, despite the decrease in size of both core and transition zones due to supralocal

dialect levelling (Trudgill 2003, 2004a).

1.3 Urban areas and speech communities

This section firstly explains what is meant by “urban” area and secondly provides a

brief review of the concept of speech community. The definition of “urban” area,

provided by the UK government, is related to the size of their population. Johnston

(1994: 651) states that an urban area “exceed[s] the thresholds of population size and/or

density which are frequently used in census definitions of urban places.” Cloke et al.

(2014: 688) uphold that the concept of “urban” is sometimes taken for granted and used

“in an almost unthinking manner.” By contrast, a core concept in sociolinguistics which

is not taken for granted is that of speech community. Traditionally, the term speech

community is referred to as “all people who use a given language” (e.g. Lyons, 1970

cited in Patrick 2002: 579). However, Patrick (2002) shows that the concept of speech

22
community intersects with many issues within sociolinguistic theory and, sometimes,

has been considered onerous to explore (e.g. Fasold, 1984). The most influential

conception of speech community was provided by Labov (1972) because, as Patrick

(2002) highlights, it is “more empirically-rooted, less generalised” compared to other

theories by Hymes or Gumperz (Patrick 2002: 584)7. Indeed, Labov (1994) argues that

the speech community is not an assumption or a theory, rather it is the outcome of

empirical research. Despite being greatly influential, Labov’s definition of speech

community has been subject to criticism. Britain and Matsumoto (2005: 7), for

instance, criticise the fact that he did not include non-native speakers in his New York

sample yet, Patrick (2002: 589) points out that “the legitimacy of analytical choices [...]

depends upon selection of the research question, in addition to the site.”

Speech communities may also overlap with one another, thus it is suggested to

conceptualise their intermediate structures (Patrick, 2002). Moreover, the speech

community should not be treated as a component of social examination, as it is viewed

as social unit used for linguistic analysis (Patrick, 2002).

Having outlined the difference between urban area and speech community, let

us now explore more closely the three urban areas selected for this study.

Colchester

Colchester, a historic market town in the county of Essex, is considered Britain’s oldest

town founded by Romans. It is located 66 miles north-east of London. The borough of

Colchester encompasses an area of 125 square miles (320 km2) from Dedham Vale

(Suffolk border) to Mersea Island. The population growth started between 1914-92,

with an approximate increase of 10,000 between the two World Wars (Baggs et al.,

7
See Patrick (2002) for a more detailed account on the conception of speech community.

23
1994). The 1801-1981 census recorded a population size of 43,452 in 1911; in 1951,

the total population was 57,449, whereas in 1981 it grew by 82,227. This steady

increase was influenced by the central government legislation for housing and town

planning in 1921 as well as by the wide ancient boundaries within which Colchester

was able to expand. After the Second World War, the community increased both

southwards towards London as well as inwards from neighbouring villages and towns.

Despite the marginal bomb damage, some slum clearance and new buildings were

needed to minimise the population pressure on available housing. A total of 6,147

public and private houses were built between 1945 and 1961, housing a total population

of 65,080. Many schools and Churches were built around 1914, and the University of

Essex in 1961 (Baggs et al., 1994).

The 2011 UK Census reported a population of 173,074, with approximately

49.3% males, and 50.7% females. In terms of ethnicity, Colchester hosts 88% of White

English/ Welsh / Scottish / Northern Irish / British; 0.6% is White Irish; 0.4% is White

Gypsy or Irish Traveller; 4% belong to the Other White ethic group; the rest of the

demographic density is composed of mixed/multiple ethnic groups, such as White and

Black Caribbean (0.5%), White and Asian (0.5%), White and Black Africans (0.2%),

Other Mixed (0.4%), Indians (0.8%), Pakistani (0.2%), Bangladeshi (0.2%), Chinese

(0.9%), Other Asian (1%), Black African (1%), Black Caribbean (0.3%), Other Black

(0.1%), Arab (0.5%), and 0.4% is included in the Other ethnic group.

The 2017-2018 Colchester Annual Economic Report shows a 9.8% increase

of the demographic density since the 2011 Census, with an estimate of 190,100

inhabitants. The ONS Annual Population Survey 2017 reveals that, in this town, the

employed and self-employed outnumber the unemployed. The number of unemployed

people is lower compared with the unemployment rates on a national scale.

24
 Location Economically active Unemployed Retired

Colchester 75.6% 3.7% 13%

England and Wales 70% 4.5% 14%

Table 1.3.1 Census 2011 - Employment and unemployment data in Colchester,

England and Wales (Nomis, 2011).

With respect to business demography, there has been a 12% rise in the number of

businesses from 2010 to 2017. In terms of education, the population of Colchester is

particularly higher skilled, with 89.1% people holding an NVQ1 or a higher

qualification, while less than 4.1 % holds no qualification.

The approximated social grade provided by the UK government is divided in

four categories: AB, which is the highest grade, covers “higher and intermediate

managerial/administrative/professional occupations”; C1 relates to “supervisory,

clerical and junior managerial/administrative/professional occupations”; C2 covers

“skilled manual occupations”; DE, the lowest grade, includes “semi-skilled and

unskilled manual occupations; unemployed and lowest grade occupations” (Nomis,

2013). 23% of the population holds the highest social grade (AB); 35% consists of non-

manual workers (C1); 20% is involved in skilled manual occupations (C2); whereas

21% has the lowest social grade (DE).

Ipswich

The town of Ipswich is located about 80 miles north-east of London and it is a non-

metropolitan district. The population started growing since the early 1800s when the

number of inhabitants was 11,277, and had reached 66,630 by 1901 (Malster, 2000).

25
A significant expansion of the town was recorded after the Second World War, as

numerous houses were built owing to bomb damage in the town8. As per request of the

UK government, Londoners were also resettled in Ipswich and an approximate 70,000

people were included in the envisaged resettlement during the 1960s. Malster (2000)

states that the expansion plan included local services such as schools, the town centre,

etc. The 2011 Census shows that the Ipswich population is equal to 133,384 or 144,957

(49.8% males and 50.2% females) depending on the precise borders which surround

the urban town (Nomis, 2013). 85.4% of people who live in this town belong to the

White English/ Welsh/ Scottish/ Northern Irish / British ethnic group category; 0.5% is

White Irish; 0.1% is White Gypsy or Irish Traveller; 5% is Other Whites; the rest of

the demographic density is composed of mixed/multiple ethnic groups, such as: 0.4%

is White and Black African, 0.4% is White and Asian, 1.75% is White and Black

Caribbean; 0.7% of the population is composed by Other Mixed; 1.16% by Indians and

1.13% by Bangladeshi; 0.2% by Pakistani; 0.4% is Chinese; 0.1% is classified as ‘other

Asians’; 0.8% by Black Africans; 1% by Black Caribbean; 0.3% belongs to other

ethnic groups; 0.1% by Arabs, and 0.6% by any other ethnic group.

The Census carried out in 2011 also reports 74% of the population are

economically active; 5% are unemployed, whereas 12% are retired. Levels of

economically active population and levels of unemployment are relatively higher

compared to the overall rate for England and Wales.

8
The Chantry estate, studied by Straw (2006), is an example.

26
 Location Economically Unemployed Retired
active

Ipswich 74% 5% 12%

England and Wales 70% 4.5% 14%

Table 1.3.2 Census 2011 - Employment and unemployment data in Ipswich, England
and Wales (Nomis, 2011).

The Suffolk Haven Gateway Employment Land Review 2009 describes the changes

which occurred in the Ipswich economy over the last twenty-five years. During this

time period, there has been a notable increase in tertiary sector activities, with an

approximate rate of 80% of the total employment. According to the East of England

Forecasting Model (2016), this town has recently experienced a steady growth in job

numbers as, between 2001 and 2016, 5,330 additional jobs were registered.

The approximated social grade provided by the UK government shows that 18% of the

Ipswich population has the highest social grade (AB); 28% is included in the C1

category; 25% is included C2, whilst 30% holds the lowest social grade (DE).

82.5% of the Ipswich population holds NVQ1 and above, whereas 8.4% holds no

qualifications.

Ipswich is also mentioned in several works of literature; it has been used as a

setting by Charles Dickens for his novel The Pickwick Papers, and it is also mentioned

in the Canterbury Tales where Chaucer criticises the Ipswich merchants.

27
 Norwich

The city of Norwich is located in the county of Norfolk, along the River Wensum, and

lies about 100 miles from London. The city lays north of the A47 – the principal

connection which links Norwich with Great Yarmouth to the east and Kings Lynn to

the west. It is well-connected to London Liverpool Street via the National railway

station, which provides an hourly service, and it is well-linked to the Midlands.

Norwich is deemed to be the capital of East Anglia as, for centuries, it was the second

largest city in England (after London). This suggests that it is “of considerable cultural

and commercial importance for the surrounding area of Norfolk and indeed for East

Anglia as a whole” (Trudgill, 1974: 6). The Second World War brought about notable

damage in large parts of the city, with the highest number of casualties in the East of

England. In 1945, the City of Norwich Plan was developed to supply a massive

redevelopment of the area.

According to the Census carried out in 2011, the population of Norwich had

132,512 inhabitants, with 49% males and 51% females. With respect to ethnicity, the

Norwich economic assessment 2018 reports an approximate rate of 84.7% of white

English /Welsh /Scottish /Northern Irish /British; 0.7% is White Irish; 0.1% is White

Gypsy or Irish Traveller; 5.4% of the population represents the Other White ethnic

group category, where the largest growth (2.7% points) has taken place9. The rest of

the demographic density is composed of mixed/multiple ethnic groups: White and

Black Caribbean (0.5 %), White and Black African (0.5%), White Asian (0.7%); Other

Mixed groups (0.6%), Indian (1.3%), Pakistani (0.2%), Bangladeshi (0.4%), Chinese

(1.3%), Other Asian (1.3%), Black African (1.3%), Black Caribbean (0.2%), Other

Black (0.1%), Arab (0.5%), any other ethnic group (0.4%).

9
It is believed that this rise is prompted by immigration from Eastern Europe.

28
 68% of the population is economically active – a rate which is slightly lower

compared to that of England and Wales, as illustrated in table below:

Location Economically active Unemployed Retired

Norwich 68% 5% 11%

England and Wales 70% 4.5% 14%

Table 1.3.3 Census 2011 - Employment and unemployment data in Norwich, England
and Wales (Nomis, 2011).

83.5% of the Norwich population holds NVQ1 and above, whilst 10.6% holds no

qualifications. The approximated social grade provided by the UK government shows

that 20% of the Norwich population holds the highest social grade (AB); 31% consists

of non-manual workers (C1); 18% consists of skilled workers (C2); whereas 30% holds

the lowest social grade (DE).

1.4 Broad research questions

Having described the settings, the demographic density, and the economic profile of

Colchester, Ipswich and Norwich, let us turn the attention to the research questions

which aim at exploring the language use of these three East Anglian localities from a

variationist sociolinguistic perspective. As mentioned earlier, the two linguistic

variables selected for this study include (t,d) deletion and (t) glottaling. But why study

(t,d) deletion and (t) glottaling in East Anglia? (t,d) deletion has been largely

investigated in many North American studies, where the profile of (t,d) is that of a

stable variable; comparably (t,d) has received little attention in the UK, where (t,d) was

mainly researched in Northern varieties of British English, as in York (Tagliamonte &

29
Temple, 2005), Manchester (Baranowski & Turton, 2020) and in Tyneside English

(Woolford, 2018). Despite slight differences in constraint ranking, phonological and

morphological constraints were found to be largely uniform in North American

studies; conversely, opposite results were found in the UK: in York (Tagliamonte &

Temple, 2005) (see section 3.8.1) morphological class did not reach statistical

significance, whereas data from Manchester (Baranowski & Turton, 2020) (see section

3.8.2) and Tyneside English (Woolford, 2018) (see section 3.8.3) exhibit the usual

robust morphological effect in line with US findings (see Chapter 3 for further details).

In view of these conflicting findings, the present survey set out to shed light on the

unsolved problem of morphological constraints for (t,d) in British English dialects. In

doing so, (t,d) will be treated as a single variable following the vast majority of studies.

In order to provide a thorough description of (t,d) deletion in word-final consonant

clusters (e.g. want), we should necessarily investigate (t) glottaling - a typical

phonological feature of East Anglia – which intersects with (t) deletion, as in the

following lenition scale: /want/ → [wɒnt] → [wɒnɁ] → [wɒnØ].

(t) glottaling has been widely investigated in this area (e.g. Norwich) in both apparent

time and real time (Trudgill 1974; 1988), yet little systematic research has been

conducted in Ipswich and Colchester. Previous work reports lack of gender and social

class effects in some communities (e.g. Manchester), accompanied by high rates, hence

the use of word-final /t/ glottaling is suggested to be an advanced change nearing

completion (Baranowski & Turton, 2015).

With respect to (t,d) deletion, the research questions addressed in this survey broadly

ask:

1) How does (t,d) behave in East Anglia?

2) Is the morphological effect present or absent in East Anglian English?

30
 3) Is it worth providing a more fine-grained analysis of the following phonetic

environment?

4) Do external (i.e. social) factors play a notable role?10

5) Do Colchester, Ipswich and Norwich exhibit similar patterns?11

With respect to (t) glottaling, the research questions ask:

1) How does (t) glottaling behave in East Anglia?

2) Is the variability of word-final /t/ glottaling conditioned by phonological factors

only?

3) Is word-medial /t/ behind word-final /t/ glottaling in the change?

4) Do external (i.e. social) factors play a notable role?

5) Do Colchester, Ipswich and Norwich exhibit similar patterns?

A far more detailed account on research questions with descriptive and theoretical

inquiries, and hypothesis is provided in Chapter 3 (section 3.6) and Chapter 5 (section

5.10), following the respective literature reviews of (t,d) deletion and (t) glottaling.

With respect to the intersection of (t) deletion and (t) glottaling in word-final consonant

cluster, the research question asks:

1) How does (t) glottaling and (t) deletion behave when the two variables are

explored at their intersection? Do frequencies of deletion and glottaling change?

How should they be properly counted?

10
In previous (t,d) studies, as will be outlined in Chapter 3, social factors played a marginal role.
However, in Tyneside English (Woolford (2018) external factors were marked as significant constraints.
11
This survey is an urban sociolinguistic study and it is not integrated into a spatial framework. While
geographical space was found to be of importance for language variation and change (e.g. Britain, 2013),
this study does not employ the use of wave and gravity models to investigate the diffusion of these
variables across geographical space.

31
1.5 Summary

This Chapter sought to explain the motivations of investigating (t,d) deletion and (t)

glottaling in East Anglian English by presenting a summary of research questions (RQs

will be discussed in more detail in sections 3.7 and 5.10). It has also provided a

geographical, historical, economic and cultural account of the three speech

communities examined.

32
Chapter 2 - FIELDWORK AND METHODS

This Chapter primarily focuses on the fieldwork and methods employed to carry out

the present study by highlighting the quantitative paradigm of variationist research

pioneered by Labov. Before providing details of such methodology, this chapter

outlines sampling methods as well as fieldwork considerations. I will then provide a

brief overview on the early techniques adopted prior to the development of modern

sociolinguistic methods. This discussion leads us to consider the principles of sampling

as a trait d’union between research design and research goals.

2.1 The sample

The sampling procedure is a valued step that challenges any sociolinguistic researcher

in order to guarantee representativeness. The latter - a key for social scientific studies

- defines the level of accuracy of a sample enabling researchers to draw conclusions on

the larger population under investigation (Milroy & Gordon, 2003).

The first sociolinguistic surveys followed a particular protocol (sample frame) by

enumerating the relevant population, and by randomly selecting the individuals from

that particular research site. This random sampling method, pioneered by Labov

(1966), aimed at providing an equal opportunity to all members of the population to

take part in the case study by ensuring objectivity and avoiding bias. Consequently,

Labov developed a secondary random sample. Schilling (2013) argues that lists such

as telephone directory used to randomly select the informants are themselves biased.

In the case of directory lists, for instance, people with only mobile phones are excluded

and not all those randomly selected may be willing to participate.

33
Trudgill (1974)’s sample was defined as quasi-random and was taken from four ward

voter registration lists. The label quasi-random springs from the methodology adopted

which is random in the selection of names, but it is not random in the selection of the

wards.

The quota or judgment sample, employed in Labov’s work in Martha’s

Vineyard (1963), has been generally adopted in sociolinguistic research, such as in the

Veeton study (Patrick, 1999) where social network approach and a judgment sample

were combined. Participants, with this sampling technique, are chosen to fill pre-

selected cells related to social factors. In other words, “the researcher (1) identifies in

advance the types of speakers to be studied; and, (2) seeks out a quota of speakers who

fit the specified categories” (Tagliamonte, 2006: 23). Schilling (2013) argues that this

method ensures that social variable cells are filled, providing a trait d’union between

research design and research goals. Indeed, to achieve a representative account it is

necessary not to lose sight of the main research goals.

This means that if one’s aim is to describe the whole population, selecting speakers

only from a sub-group would be a clear bias. Romaine (1980) considers the validity of

sampling procedures claiming that very small samples might not ensure statistical

representativeness. Sankoff (1980a) has summarised three crucial decisions to be made

prior to starting data collection:

1) to define the sampling universe (i.e. a group or a community in which the

researcher is interested.)

2) to evaluate language variation within the population;

3) to determine the sample size.

Milroy & Gordon (2003) suggest that researchers should be aware of the way they

adopt the definition of sampling universe as it may have an influence on their results.

34
To assess the variation within the community, Trudgill (1986) shows that the definition

of ‘native speaker’ may be problematic as some people have lived in Norwich all their

lives, yet they did not acquire the Norwich accent. Similarly, Payne (1980) shows that

it is onerous for the children whose parents speak different dialects to acquire the

dialectal pattern of the community in which they were born. This decision, as shown

by Horvath (1985), may lead to relevant theoretical observations, such as the role of

ethnic minority speakers who were leading linguistic changes in the speech community

of Sydney. Another example of this kind is provided by Wolfram (1969) who focused

on African American English in Detroit - a unique ethnic dialect. Since many African

Americans speak Standard English natively, Wolfram et al. also included European

Americans in the sample to compare the standard end of the dialect continuum.

To determine the sample size by using a stratified sampling technique, Milroy

& Gordon (2003) suggest that if researchers aim to analyse informants from four social

classes, both sexes and four age groups, they will need 32 cells. To balance the social

parameters, for the present survey, I adopted a judgment sample which consists of 36

East Anglian speakers, as illustrated in table 2.1.1. For class divisions see section 2.8.

East Anglia
Working Class Middle Class
Young Middle Old Young Middle Old
(18-28) (35-50) (60+) (18-28) (35-50) (60+)
M F M F M F M F M F M F
3 3 3 3 3 3 3 3 3 3 3 3

Table 2.1.1. Fieldwork sample design employed in the East Anglian study.
.

Since the main goal of this study is to investigate East Anglian English, the boundary

to define the sampling universe is taken from Trudgill’s (2001a) definition of linguistic

35
East Anglia, as outlined in the previous Chapter. On the basis of what Trudgill defines

as linguistic East Anglia, I identified three research sites for the sample design:

Colchester, Ipswich and Norwich, where participants are proportionally distributed.

Similar research designs relate to contemporary studies carried by Watt et al. (2019)

and by Leemann et al. (2019). Watt et al. (2019) examined the correspondences

between vowel phoneme boundaries in North-East England dialects, where a total

number of 31 individuals from Newcastle, Sunderland, and Middlesbrough was

analysed. Leemann et al. (2019) investigated the FOOT - STRUT split as well as the

TRAP - BATH split in 12 localities across England (Birmingham, Bristol, Leeds,

Liverpool, London, Manchester, Newcastle-upon-Tyne, Norwich, Nottingham,

Peterborough, Sheffield, York), with an average of 10-12 speakers per locality, aged

between 18 and 30.

2.2 The Fieldworker “away from home”

After selecting the communities and obtaining the ethical approval of research

involving human participants, as required by the University of Essex, I started my data

collection. Carrying out successful fieldwork is a central, if not the most crucial, part

of a study. However, Schreier (2013) notes that despite its pivotal role, it is

uncontroversial that fieldwork process does not usually have a prominent place in some

sociolinguistic PhD theses, where usually the discussion on how the data was gathered

lack details. The fieldworker needs to have an in-depth knowledge of the community

that will be investigated (its setting, the population density, the social and economic

character, history of both community and speakers, work and leisure, etc.), but this is

not a great problem if researchers study the community where they grew up (e.g.

Trudgill in Norwich, Britain in the English Fens, etc.). However, Schreier (2013: 25)

36
points out that a wide number of fieldworkers, if not the majority, conducts research in

unfamiliar communities and they are known as fieldworkers who work “away from

home”- (e.g. Eckert in Detroit, Wolfram in Washington DC and North Carolina, Nagy

in Faetar (Italy), Kasstan in the Monts du Lyonnais (France) and in the Canton of Valais

(Switzerland), Lacoste in Jamaica, etc.). Clearly, this requires researchers to develop

an exhaustive knowledge of both the local culture and speech prior to data collection

activities.

My status as investigator belongs to the category of fieldworkers “away from

home.” Many theoretical linguists praise native speaker investigators considering the

fieldworker’s use of language a crucial methodological issue. However, much

linguistics research (e.g. Caribbean creole studies) is based on data collected by non-

native or even non-speakers of the creole language investigated (Patrick, 1999). The

communication with interlocutors, in these cases, could be problematic if the

informants accommodate to the investigator’s speech as a result of the convergence

process.

Living in Colchester for several years has also helped me to absorb the local

culture, political events, leisure activities, local geography and many other aspects of

daily routine that validate me as a member of the community life.

2.3. Entering the communities

Fieldwork for this project started in July 2017 and was carried out for 11 months. I

have lived in Colchester during the data collection and I would make constant trips to

Ipswich and Norwich. There is no need to deny that the first stage of my data collection,

especially in Ipswich and Norwich, was comparable to the nightmares that Eckert

(2000) described before beginning her fieldwork in Detroit.

37
For the first two months, the data was gathered in the three communities to conduct a

small-scale pilot study. The latter is a key factor for a good study design (Feagin, 2002)

which provides two salient benefits: (1) valuable insights into the variables under

investigation; (2) it “helps you see what kind of an interviewer you are” (Johnstone,

2000: 114). This self-evaluation allows researchers to see whether it is beneficial to

conduct an unstructured interview where the informants lead the topic shift, or whether

researchers manage to stick closely to the pre-designed topics being aware that in this

case the data gathered will not be representative of the everyday casual speech (see

section 2.4).

To enter the three communities, I adopted two basic strategies: (1) the friend-

of-a-friend technique (Milroy, 1980) and (2) “through persons who are centrally

located in social institutions” with an overview on the community (Labov, 1984: 31).

The first approach allows the fieldworker to be an observer and a member in the

community at the same time due to entangled obligations, which spring from the help

received in recruiting the participants (Wolfram & Schilling-Estes 1998; Tagliamonte

2006). Labov (1982: 173) states that “a linguist who has gathered data in a speech

community has an obligation to act in the interests of members of that community,

when they have need of it”. In some cases, he suggests paying back the community by

using the knowledge gained from the data for the benefit of the community,

maintaining the confidentiality of their data. To this “Principle of the debt incurred”,

Johnstone (2000) adds that sometimes this is not always practically possible, as some

participants might not be interested in the results. Hence, an additional option could be

helping the participants with their daily activities1 - as was the case in this study.

1
David Britain, for instance, brought groceries for his participants, Jenny Amos gave lifts to her
informants, others would help with children’s homework, etc.

38
The second approach I adopted to enter the communities (via institutions) should be

considered carefully as making contacts only with people who hold an official status

could bias the data towards the standard speech style, and the study will not be

representative of the whole (Tagliamonte, 2006). In this respect, Labov (1984) found

this strategy to be most effective when recruiting middle-class speakers. Conversely,

in the present survey, I employed this technique to interview mainly working class

informants, and secondarily members of the middle class in the Ipswich community.

The institution I turned to is one of the local Homecares, which provides personalised

care and support to convalescents, elderly people, etc. The tasks performed by personal

carers are related to daily routine and include assisting patients with bathing, dressing,

moving and further personal needs, and they do not take over tasks which require a

wide medical knowledge or training. When I contacted one of the supervisors in that

Homecare, she showed promptly interest in my research and, after volunteering2, she

kindly recommended me to an extensive number of employees. Thus, I had the chance

to interview many personal carers, as well as a limited number of office workers. Some

of them, however, were not Ipswich born and bred, hence they were excluded. I also

contacted by email other institutions, in both Ipswich and Norwich, such as libraries,

theatres, student unions explaining the research project and asking for volunteers who

were willing to participate, yet that proved fruitless. With some staff members,

however, I exchanged multiple emails and we even scheduled an appointment, yet

eventually it turned out that most of them were extremely busy to take part in an hour,

or at least fifty-minute, interview.

These approaches aimed mostly at recruiting speakers from the young and

middle generations, but seeing how unfruitful it was, to expand my network I signed

2
Given her office-based position, as a supervisor, she was classified as a middle class member (see
section 2.8, for more details related to the social stratification).

39
up to a Meet Up smartphone App which connects you with local people and events

allowing you to make new friends. To expand my network towards the elderly residents

in both Ipswich and Norwich I went to the respective conservative clubs, which are

typically frequented by local people yet, sadly, I was told that the approval committee

do not allow researchers to recruit participants in these clubs. Finally, some office

workers in the Ipswich Museum helped me in gathering a number of older participants.

In Norwich, I also gained access to the community thanks to a former Essex

student, who is Norwich born and bred; with the help of a PhD student at the University

of East Anglia, and with the help of a barman who kindly recommended some clients.

At the end of each interview, I would ask each participant to recommend other friends

or relatives who could be interested and willing to take part in this study.

The snowball method was also adopted in Colchester, where my social network

is wider having done my Master’s at the University of Essex and having lived in the

community for several years. I firstly contacted, via telephone, a friend who I used to

attend lectures with; she introduced me to young and middle-aged working class

speakers. I then contacted a supervisor3 of the accommodation where I used to live as

well as some employees at the University of Essex, who kindly guaranteed me to their

middle class friends and colleagues.

As claimed by Feagin (2002), self-presentation plays a crucial role in the

fieldwork: I was careful to dress appropriately wearing shirt and stockings especially

when meeting elderly informants, yet I would occasionally wear blue jeans when

interviewing young speakers. I would normally take along biscuits or chocolates to

3
The term supervisor, here, is not to be intended as “the lower salariat” of the ESeC model (Rose et
al., 2010) which includes supervisory occupations, yet it is referred to the lower supervisory category
belonging to class 6. (See section 2.8 for further details).

40
make the interview less formal, paying particular attention in the selection of biscuits

to avoid the crunch noise on the tape.

Moreover, my Italian identity proved advantageous mostly in Colchester and

Ipswich where I met several participants with a surprising fervent passion for the Italian

culture. In one particular occasion, while I was interviewing a middle class woman

from Colchester, she drew my attention to the Italian flag lapel pin she was wearing as

a strong admiration symbol for the art, fashion and food of Italy4. In Ipswich, I tape-

recorded a middle-class man who kindly asked me to teach him some Italian words, at

the end of the interview. On later visits to the community, I offered to give free Italian

lessons to his kids, who apparently were seeking for a tutor.

Before each interview was carried out, I carefully followed the ethical protocol.

Milroy & Gordon (2003) claim that a non-fulfilment of ethical guidelines may cause

extreme implications to the research institutions, such as loss of reputation, loss of

funding, etc. Each participant, prior to the recording session, was given a Consent Form

adapted from a template of the Department of Language and Linguistics at the

University of Essex, as well as a Participant Information Sheet (PIS). Johnstone (2000)

emphasizes the importance of this stage as the participants gain awareness of what is

involved in taking part in a research study, also becoming aware of their rights (e.g. the

participant’s rights to anonymity, confidentiality and withdrawal, etc.). In the PIS, I

provided a brief introduction of the topic, explained why I am doing this project; what

the participant will have to do if they agree to take part; and ensured them that the data

will remain confidential. I also explained which are the advantages of taking part in a

research project, and the disadvantages (e.g. feeling uncomfortable being recorded),

and explained to them how I was going to use my data. Trechter (2013) points out that

4
Note that all of my participants identified themselves as monolingual.

41
this moment of formality can increase the unwanted observer’s paradox (see section

2.4), while the researcher is trying to elicit casual speech. A crucial step, at this point,

requires breaking the ice without diminishing the role of the consent form that the

participant has just signed.

2.4. The sociolinguistic interview

Labov (1984) suggests that a typical sociolinguistic interview should last at least one

hour or two hours per speaker. Milroy & Gordon (2003), however, point out that

sometimes it can be onerous to be categorical about the interview length, claiming that

phonological data can be gathered in 20 or 30 minutes. In this study, the length of each

one-to-one interview is about 50 or 60 minutes - only one lasted roughly half an hour.

Milroy & Gordon (2003: 58), citing Douglas-Cowie (1978), claim that “even when

interviewed by a stranger, speakers will settle down to a pattern approximating to their

everyday interactional style after about the first hour.” This is one of the principal

reasons I would meet with my participants before interviewing them. Specifically, I

would usually meet with my older participants a couple of hours before the actual

interview started, while I would normally hang out for a couple of days with young

speakers before the interview.

To be fruitful and successful, an interview should be cautiously planned. As

outlined earlier in the Chapter, a sociolinguistic interview is not synonymous with free

speech (e.g. conversations which arise among friends) due to the presence of a

recording device. This may give rise to the observer’s paradox which clashes with the

purpose of sociolinguistic research to access the vernacular (i.e. a linguistic variety

used by a speaker or a community for every day and home interaction) in order to “find

42
out how people talk when they are not being systematically observed; yet we can only

obtain these data by systematic observation” (Labov, 1972: 209).

A structured sociolinguistic interview could potentially reduce the observer’s paradox,

if structured by specific modules. These interview modules, proposed by Labov (1984),

include a number of questions which focus on a particular topic, such as children’s

games, the “danger of death”, etc. Carrying out an unplanned interview can turn out in

long-lasting questions, can be ambiguous and may lead to spontaneous speech

phenomena such as hesitations and false starts. Labov suggests that it should take less

than five seconds to deliver planned questions, and that questions should be formulated

from an outsider’s viewpoint. Moreover, it is recommended that the interviewer acts as

a learner - an undemanding task for me, as an interviewer - especially in the

communities of Ipswich and Norwich.

Topics which Labov describes as conversational networks are selected based

on (a) previous successful subjects which engaged the participants in the conversation,

and (b) topics which may yield information on neighbourhood norms and on more

general ones. Each module, in my interviews, began with general questions, such as

‘Did you go to one of the schools in the neighbourhood?’ (Tagliamonte, 2006: 38) to

measure a participant’s level of interest in that topic. If a speaker is not particularly

engaged with that topic, the interviewer moves on to further memorised questions

which are more personal experience-based, such as “Did you ever get blamed for what

you didn’t do?” (Labov, 1984: 34). The order of topics such as childhood5, games,

family, dreams, work, marriage, Brexit, etc., was not always strictly followed as I had

planned. This means that a shift of topic or a topic initiated from the participant has

always been accepted to allow the speaker to feel relaxed and produce speech

5
Such as, “Getting back when you were a kid, was there anyone you didn’t like?” etc.

43
representative of their everyday language. Semi-structured approaches, therefore, also

proved productive.

If the participants no longer answer questions, the interview is deemed a failure.

Even if the informants cooperate, the interview should not be governed by “the

cooperative principle” according to which they answer the questions briefly (Levinson,

1983: 100). Indeed, a crucial part of the interview is “the additional material that the

speakers provide, beyond the initial question” (Labov, 1984: 38). Since it is onerous to

elicit casual speech from the sociolinguistic interview, researchers typically ask

questions that involve emotional reactions; in this case, the interviewees are more

focused on what they say rather than how they say it, therefore their speech is likely to

be less controlled. A key-question, in this case, is the so-called “danger of death”.

However, this technique has not always proven successful. Trudgill, for instance,

reports a weak effect in Norwich, suggesting that perhaps the eventful lives of people

who live in this community differ from those who live in New York City.

From my personal experience, the “danger of death” question was successful

among old participants who remembered, with tears in their eyes, the difficult times

they went through during World War II when some areas of East Anglia were bombed

by Germans. This question evoked different reactions in the middle generation, who

would usually remember the loss of their loved ones (rather than fearing for their own

lives) and how they were coping with this traumatic time. This technique, however,

was found unsuccessful among young speakers from the three communities, perhaps

for the same reasons suggested by Trudgill. Some researchers when faced with this

difficulty would hit upon a topic that will engage the speakers in a similar way, such as

asking the informants to tell ghost stories (Herman, 1999). Instead of using this strategy

with young participants, I replaced the unsuccessful ‘danger of death’ question and the

44
‘ghost’ story with the question ‘Which was the worst day of your life?’, as some of

them might not be familiar with ghost stories, but it is likely that all of them have had

a worst day that might have marked their lives. Some of them talked about the divorce

of their parents and how difficult it was to be raised up by a stepmother; others talked

about particular memories from their childhood (e.g. how they were bullied at school),

etc.

It was not extremely difficult to make the informants feel at ease and to let them

speak - perhaps my friendly and warm personality played a paramount role. I always

respected their privacy and adapted to local customs. Some elderly people shared with

me personal memories before the famous “danger of death” or “the worst day of your

life” questions were asked; some welcomed me in their houses, some offered to pick

me up at the train station in both Ipswich and Norwich to drive me to their places,

others showed me photographs and paintings. I have become friends with several others

and periodically we meet to attend social events. Usually, recording sessions were

scheduled in advance, but I did not attempt to record the informants on every visit, as

mentioned above, although I always took recording equipment along6. Before turning

to the use of instrumental techniques employed for data analysis, let us take a brief step

back to selectively review early field methods.

2.5. Early field techniques

Traditional dialectologists had no recording equipment to preserve the informants’

speech for later analysis, until the advent of the tape-recorder in 1960s, and field

methods were restricted to a small portion of population such as non-mobile older rural

6
The interviews were conducted by using a PMD661MKII recorder, and recordings were made in
.wav files.

45
males (NORMs) as called by Chambers and Trudgill (1998). In this section, I will

selectively review early techniques adopted in three main studies, namely On Early

English Pronunciation, The English Dialect Grammar and The Survey of English

Dialects.

On Early English Pronunciation is a noteworthy English survey carried out by

Ellis (1889) throughout the 19th century. In this corpus of British, Welsh, Irish and

Scottish dialects, a three-part methodology was employed resulting in: Comparative

Specimen, which consists of a fifty line reading passage with the purpose to obtain

“dialect renderings” of daily words; Dialect Test, aimed at the identification of vowel

sounds as well as features of speech through a further short reading passage; whilst the

Classified Word List technique involved a large-scale of tokens, the purpose of which

was to elicit phonetic-phonological features of the words listed.

The English Dialect Grammar conducted by Wright (1905), sets out the most

striking features of all British dialects, classified following Ellis’ (1889) suggestions,

as mentioned above. In order to carry out his research, he composed a 2,431-word list,

including both the literary language and the spoken dialect. Two main procedures were

used to arrange the grammar: the first involved lists concerning Old English phonemes,

the second concerned variants grouped in relation to individual tokens.

The Survey of English Dialects, conducted under the direction of Orton between

1950s and 1960s at the University of Leeds, is considered a striking archive. The survey

includes 313 localities from the Isle of Man, Wales and England7 and is composed of

a three-part elicitation method: (a) the use of questionnaires consisted of more than

1000 words intended to elicit phonological, lexical, morphological and syntactical

7
London and West Yorkshire which, conversely, were included in The Atlas of English Dialects (Upton
& Widdowson, 1999).

46
traits; (b) diagrams and pictures were used to facilitate the respondents’ identification

of objects concerning local information; whilst, (c) spontaneous speech entailed the

informants’ opinions, occupational details, the farm, social activities, etc.

An accurate insight into language variation entails including young speakers,

women and those living in urban areas; in addition, a detailed investigation of complex

speech communities requires quantitative analytical methods to enhance a thorough

description of linguistic variation.

2.6. Data analysis instruments

With technological advancements sociolinguists have turned to various quantitative

methods which saw the employment of statistics in data analysis. The use of

quantitative analysis allows sociolinguists to make generalisations, going beyond

single cases with the attempt to develop general rules or patterns. Fasold (1972: 33),

for instance, adopted the chi-square test not only to provide a correlation between

linguistics and social factors, but also to establish the role of variability within the

linguistic analysis. Therefore, he proposed a distinction between “truly random

variability” and “conditioned variability.”

The introduction of more rigorous statistical processes such as the Cedergren-

Sankoff Varbrul program, where databases grouped speakers all together rather than

analysing them as individuals, had a great impact on early sociolinguistic studies. Guy

(1980) suggested to find out whether the variation within the speech community is due

to diversity within the group or whether it is present with the same structure in each

individual, treating speaker as random effect. This means that it is modelled as a

random variable with a mean of zero and unknown variance (Baayen, 2008), allowing

47
the intercept term to vary across subjects (Barr et al., 2014). This way we can observe

which individuals contribute most or least strongly to the variation under discussion.

The succeeding analysis of variance, well-known as ANOVA, was inadequate

to handle variationist analysis due to the impossibility to deal with a large number of

potential combinations (independent variable and categories) and due to the

inapplicable algorithm calculation which typically entails a homogeneous number of

tokens per cell (Bayley, 2013).

The program extensively used and designed for linguistic variation is known as

Varbrul (Cedergren and Sankoff 1974; Sankoff 1988), an early realisation of binary

logistic regression. The latter is a validated theory which allows linguists to model,

statistically, “the effect of one or several predictors on a binary response variable”

(Speelman, 2014: 487). This methodological instrument enabled researchers to amend

hypotheses and re-examine the data. Multivariate analysis was then conducted by

Goldvarb 2.0 (Rand & Sankoff, 1990) and Goldvarb X (Sankoff, Tagliamonte and

Smith 2005), also known as the variable rule program and an extremely productive tool

until a few years ago.

Rbrul, introduced by Johnson (2009), is one of the current tools which provides

a wide statistical software package, including the open-source program R to examine

both individual and by-group behaviour. In designing the Rbrul program, Johnson

(2009) set some crucial goals, such as maintaining multiple logistic regression as well

as cross-tabulations like Goldvarb, but also endowing this software with specific

functions which were absent in the Goldvarb package. Among these new functions

there is the support for continuous predictors (independent variables), for continuous

responses (dependent variable) and mixed-effects models supplied with random

effects. This way we can observe which individuals contribute most or least strongly

48
to the variation under discussion. It provides the outcome of effects in log-odds as well

as factor weights supported by R-squared and can tackle knock-outs8. Unlike some

precursory statistical processes, in Rbrul, the number of factor groups is unlimited.

Since this tool is more user-friendly than R, Rbrul has been employed in this research

for an attentive data analysis and to account for the variation it describes. Data has been

subject to mixed-effects modelling, including speaker and word as random effects.

Rbrul does not report the significance of differences between levels in a factor group

as p-values refer to the whole factor group. This study follows previous work in how

factor weights are represented (e.g. Tagliamonte & Temple, 2005).

The comparison of different nested models was carried out with the log-likelihood ratio

test, which performs significant tests with mixed models by comparing the likelihood

of one model to the likelihood of another model (Winter, 2020). This is also known as

‘deviance tests.’ The Akaike Information Criterion (AIC), which provides a means of

model selection, was employed to compare non-nested models.

2.7. Dependent variables

The importance of statistical modelling moves beyond some basic points, such as the

description of significance tests, and “involves a concept of language use in which a

linguistic variable is seen as the output of a system that is influenced by a number of

distinct influences, both social and linguistic” (Guy, 2014: 209). This output, also

known as the dependent variable, is an essential unit in statistical modelling. A

dependent variable can be found at grammatical (syntax, semantics, morphology),

phonological and/or pragmatic level (Tagliamonte, 2006). The two dependent variables

of this survey are (t,d) deletion and (t) glottaling which, throughout this work, will be

8
Knock-outs refer to categorical distribution of tokens: zero or 100%.

49
both referred to as phonological variables - an umbrella term commonly adopted in

most studies (Foulkes, 2006) to include the subtle difference highlighted by Hudson

(1996) between phonetic and phonological variables. Phonetic variables, under

Hudson’s (1996: 170) terminology, are those in which “the same phonological pattern

has different phonetic realisations”, such as the articulation of the phoneme /t/ with [t],

[Ɂ], and other forms of variation like taps. Phonological variables, are those in which

“the same lexical item has alternative phonological structures”, such as /h/ dropping in

the lexical items house and happy.

Data for both variables was coded auditorily, as “it is not always possible or

practical to undertake acoustic analysis of large bodies of data” (Docherty et al., 1997:

280).

2.7.1. (t,d) Deletion

The first phonological variable examined in this survey is (t,d) deletion - the alternation

between [t,d] and Ø – also known as coronal stop deletion (CSD). The type of logistic

regression analysis carried out is binary: deletion vs. [t,d]. Even though it is common

practice to code tokens realised with glottal stops or glottalisation in the retained

category, along with [t,d], in the present study glottal realisations of word-final /t/ were

excluded from the (t,d) dataset as (t) glottaling holds social meaning in British English

and thus I believe is a different phenomenon9 (see Chapter 4 for additional excluded

cases). A total number of 4,879 tokens was coded into an excel spreadsheet,

constrained by linguistic and social factors.

To contribute to the variability of (t,d), the present survey aims at exploring

more closely those predictors which suggest that there is a universal constraint on this

9
For a similar view, see Amos et al. (2020).

50
variable and which are in line with universal phonetic and phonological properties of

segments (e.g. the following phonological environment). A detailed discussion on this

respect will be provided in Chapter 4.

2.7.2. (t) Glottaling

In British English, (t) is governed by two closely related processes: T-glottaling, the

replacement of /t/ with the glottal stop [Ɂ], as in butter [bʌʔə]; and glottal reinforcement

(often called ‘T-glottalisation’, ‘Pre-glottalisation’, ‘Post-glottalisation’), as in

mattress [mæʔtrəs]. In this analysis, (t)-glottaling, the few cases of glottal

reinforcement of /t/ with [tʔ] (n = 8) or [ʔt] (n = 24), and the few cases where a period

of creaky voice occurred (n = 4) are all coded the same. As Straw & Patrick (2007:

388) note, the term glottalization is vaguely used in the literature “to refer to one or

more such elements (e.g. now including complete stops, now excluding them),

especially when generalising across studies”. Docherty & Foulkes (1999: 54) propose

the label ‘glottal(ised)’ variants as it “covers two distinct [acoustic] types.” 10

These alternations, throughout this work, will be referred to as glottal variants,

due to the high number of tokens with alternating production of [t] and [Ɂ] in the

dataset, compared to the relatively low number of other types of glottal variation. 4,923

tokens were collected; in 3,051 cases (62%) /t/ occurred in word-final position (e.g.

that), while in 1,872 cases /t/ was found in word-medial context (e.g. nightmare).11

Each realisation of (t) was coded into an Excel spreadsheet 12, along with internal and

external independent variables. The type of logistic regression analysis conducted is

10
Further terminological details will be provided in Chapter 5.
11
The total number of tokens does not include cases realised with the alveolar tap [ɾ], as they were
excluded prior coding.
12
For the excluded cases see Chapter 6.

51
binary: glottal variants vs. /t/. To potentially contribute to the variability of (t), one of

the main purposes of this study is to also explore (t) glottaling in environments which

received little attention in the literature (e.g. the preceding phonetic segment) and in

linguistic contexts where (t) glottaling was considered to be blocked (e.g. after a non-

resonant consonant, as in project). A critical and more detailed explanation as to why

these environments have not been excluded from the analysis will be discussed in

Chapter 5 and Chapter 6.

2.8. Independent variables

The linguistic factor groups included in this survey are the same for both variables,

with the exception of voicing agreement (homovoiced vs. heterovoiced) which was

only coded for (t,d). The remaining linguistic predictors are listed as follows: preceding

and following phonetic environment, and syllable stress; whereas the external

predictors include lexical frequency,13 socio-economic class, age, sex, and style. Both

preceding and following environments were coded on the phonetic surface. Following

/h/, for instance, was coded either as [h] when the voiceless glottal fricative was

consonantal, or as a vowel when /h/ was dropped. Each case of /h/ dropping was coded

as (h)-Vtype, implying that the following phonetic segment was realised with a particular

vowel type in those tokens where /h/ was underlyingly present. The same procedure

of coding on the phonetic surface was applied to the so-called ‘David variable’ – the

alternation between the ‘standard’ form [ɪ] and the ‘non-standard’ [ə], as in wanted

13
Location was initially included in the model to control for variation in geographical space yet, as
expected, this predictor turned out not to be statistically significant, therefore it was subsequently
excluded from the analysis. However, statistical details related to this factor group will be presented in
Appendix I and Appendix II along with other non-significant predictors.

52
[wɑnʔəd] rather than [wɑnʔɨd]14. Historically rhotic tokens /-rt/, /-rd/ were excluded

prior to coding in both (t,d) and (t) datasets. In the (t,d) dataset, /-lt/ /-ld/ sequences

were only coded when /l/ was phonetically consonantal. In the (t) dataset, /l/ was coded

as a vowel when the lateral-approximant was vocalised, whereas when /l/ had

consonantal features it was coded as consonant.15

Following pause is a delicate context to inspect, especially when examining the

intersection between /t/ deletion and /t/ glottaling in the C(C)t_Pause context (e.g. good

point.). Tanner et al. (2017: 8) point out, that a following pause is commonly treated as

an environment in sociolinguistic literature on CSD, similar to that of following

consonant or following vowel. However, they state that “beyond a certain pause length, it

is very unlikely that an upcoming word would be planned, and in this limiting case pause, or

rather, the null environment, is indeed a separate type of context.” 16

The criterion I adopted to measure pause is largely based on the pause duration

following Fors (2015). Along the line of Fors (2015), in the Production and Perception

of Pauses in Speech, a typical pause lasts only about a quarter to half a second. Since

the presence of a glottal indicates some reflex in the signal (creaky vocal fold pulses or

a spike in voicing offset), in this study, a pause is considered to begin after the voicing

bar either ceases abruptly or fades out. Based on this criterion, an interruption of the

voicing bar early in the pause indicates a glottal variant, whereas no evidence of /t/ or

glottal gesture early in the pause implies /t/ deletion. Figure 2.8.1 illustrates an example

14
This characteristic feature of East Anglian English is found to apply in unstressed, closed syllables
(Trudgill, 1986) and in unstressed open syllables (Potter, 2018).
15
L-vocalisation is considered a distinctive feature of London English (Wells, 1982), and it is also
present in the North, especially in Yorkshire (Ihalainen,1994). However, some areas of East Anglia
inhibit /l/ vocalisation (Johnson & Britain, 2007). It is argued that vocalisation of /l/ is more likely to
occur in dialects which have a clear distinction of ‘clear’ and ‘dark’ /l/ and, in Northern East Anglian
dialects, the use of /l/ was found to be clear until the 20th century (Johnson & Britain, 2007). Indeed,
common pronunciations of hill as [hɪl] are typical of rural East Norfolk (Trudgill, 1999).
16
Tanner et al. (2017) use pause length as a proxy for prosodic boundaries in their CSD analysis, showing
that “longer pauses gradiently decrease deletion.”

53
of /t/ deletion in the word amusement realised by an eighty-two year-old man from

Colchester: the duration of the pause is nearly half a second and both auditory analysis

and spectrogram reveal neither coronal nor glottal gesture evidence early in the pause.

Figure 2.8.1 /t/ deletion after the alveolar /n/ and before a following pause.

Filled pauses (FPs), or non-silent hesitation phenomena, such as uh, um, and em are

excluded from the analysis.17 Praat was also employed to detect /t/ deletion and /t/

glottaling in some other critical cases, as shown in figure 2.8.2 and figure 2.8.3. Figure

2.8.2 illustrates an example of /t/ deletion in the word sequence went in produced by a

working-class middle-aged man from Norwich. The spectrogram displays visible vocal

folds vibrations in the realization of /n/ which precedes the underlying oral stop /t/ and

it is immediately followed by the next sound. Hence, the absence of the voiceless

coronal /t/ is represented by the transition from the alveolar /n/ to the following

phonetic segment – represented, in this case, by the high vowel /ɪ/.

17
For a detailed account on filled pauses see Fruehwald (2016).

54
 Figure 2.8.2 Example of /t/ deletion: transition to the following phonetic segment.

Figure 2.8.3 illustrates an example of (t) glottaling in the word sequences didn’t allow

realized by a working class middle-aged man from Norwich. The spectrogram shows

vocal fold vibration of the segment /n/ which is followed by an audible glottal stop.

The latter is not defined for voice, but it allows some vibration whether the glottal

constriction is weak; whereas, a total glottal occlusion would obstruct all airflow. The

underlying representation of /t/ which, in the didn’t allow it sequence, is articulated

with the glottal stop is then followed by a vowel sound.

Figure 2.8.3 Example of /t/ glottaling followed by a vowel.

55
Syllable stress is a further predictor taken into account in the analysis of both variables,

and it was coded as a binary factor: stressed syllable or primary stress (e.g. left; sit) vs.

unstressed syllable or non-primary stress (e.g. happened; community).18

The Zipf-scaled SUBTLEX-UK corpus (van Heuven et al., 2014) was adopted

to control for word-frequency as it improved the measure of lexical effects in

psychology (Brysbaert & New 2009) and it is also becoming a popular tool in

linguistics (Tamminga 2016; Baranowski & Turton, 2020). The 201.3 million words

included in this corpus are obtained by film and TV subtitles from BBC broadcasts.

The standard measure which researchers typically adopt to account for word frequency

is based on the frequency per million words (fpmw), a measure which is usually

independent of the corpus size. In corpora made of only 1 million words, a measure of

1 seems to be the lowest value; however, van Heuven et al. (2014) point out that more

than half of a word frequency list contains items whose frequency value are lower than

1 pmw. For an easier understanding of the word-frequency effect, they suggest

employing a typical Likert rating scale, from 1 to 7, without using negative values.

In addition, van Heuven et al. (2014) uphold that “the middle of the scale should

separate the low-frequency words from the high frequency words.” Values 1-3, in the

SUBTLEX-UK corpus, represent low frequency words, whereas values 4 -7 represent

high frequency ones. A comparison between Zipf values employed in the SUBTLEX-

UK corpus and fpmw is provided in the table to follow, where words whose fpmw is .1

get a value of 2.

18
Note that stress has been included in few studies which have examined the variability of (t)
glottaling (e.g. Tollfree 1999; Roberts 2006; Eddington and Taylor 2009; Barrera 2015).

56
 Zipf values Fpmw Examples

1 .01 antifugal, bioengenieering, farsighted, harelip,

proofread
2 .1 airstream, doorkeeper, neckwear, outsized, sunshade

3 1 beanstalk, cornerstone, dumpling, insatiable,

perpetrator
4 10 dirt, fantasy, muffin, offensive, transition,
widespread
5 100 basically, bedroom, drive, issues, period, spot, worse

6 1,000 day, great, other, should, something, work, years

7 10,000 and, for, have, I on, the, this, that, you

Table 2.8.1 A comparison between Zipf values and fpmw, adapted from SUBTLEX-
UK website (van Heuven et al., 2014).

The socio-economic class is also employed for multivariate explanatory analysis in this

survey, yet it is not the main focus of examination. As Patrick (1999: 82) highlights:

“The variationist paradigm retains its structuralist roots to the extent that it
seeks first to account for variation through system-internal, linguistic
explanations as far as possible, before turning to system-external elements
such as the social characteristics of speakers.”

This study goes in this direction without downplaying the considerable role of social

factors. As outlined in the previous Chapter, the use of this factor as an indicator has

been a debated issue over the years in the social sciences. For the stratification of social

class, I adopted the European Socio-Economic Classification (ESeC) model (Rose et

al., 2010), which is mainly based on economic and division of labour criteria rather

than cultural ones, and household and family are used as a unit rather than the

individual. The 10 classes proposed by Rose et al. (2010) can be collapsed into six,

five, and three-level models, as shown in the table to follow:

57
 10 version 6 class 5 class 3 class
ESeC class class version version version
Higher salariat 1
Lower salariat 2 1+2 1+2 1+2
Higher white collar 3 3+6 3+6
Petit bourgeois 4
Small farmers 5 4+5 4+5
Higher grade blue
collar 6 3+6 3+6 3+4+5+6
Lower white collar 7 7 7
Skilled manual 8 8
Semi-/unskilled 9 9 8+9 7+8+9
Unemployed (10) (10) (10) (10)

Table 2.8.2 Collapsing ESeC classes. Adapted from Rose et al. (2010).

The benefit of using this class model in sociolinguistic research is twofold: (1) the

transparency by which classes are collapsed to a three-level model allows

sociolinguists to employ this schema in small-scale surveys; (2) since this model is

merged with the European socio-economic classification this schema could be used as

a point of comparison between studies carried out across Europe to control for the effect

of this variable.

Since the sample for the present study is stratified by working and middle class,

I adopted the three-level version by excluding class 1+2 - the combined group of higher

and lower salariat19. Hence, the speakers were stratified based on the following classes:

7+8+9 (what I refer to as working class in this study) and 3+4+5+6 (what I refer to as

middle class), as illustrated in table 2.8.2. In detail, class 7 includes “non manual

workers”, such as shop workers, care workers, etc. Class 8 includes “skilled workers”

or lower technical occupations, such as plumbers, locomotive drivers, etc. Class 9

encompasses “semi-and non-skilled workers” – a type of work which is paid by the

19
Lawyers, scientists, higher education teaching professionals, etc. belong to the 1+2 group.

58
piece or hourly paid. Cleaners, labourers, messengers, etc. are to be found in this

category. Class 3 includes “higher grade white-collar (non-manual) workers”, such as

administrative assistants, jobs which require working alongside managers, etc. Classes

4 and 5 comprise small employers and self-employed, with the latter referring to those

who neither buy nor sell labour. Class 4 includes non-professional occupations; while

class 5 covers “the self-employed and small employers in agriculture, fisheries and

forestry.” The informants were assigned to a specific occupational level based on their

whole occupational history, not merely on their most recent occupation; while students

were classified on the basis of their parents’ socioeconomic position.

A further social factor employed in this survey is the age of participants, which

is usually taken into account in most variationist studies to control for linguistic change

(Schilling, 2013). While (t) glottaling is considered a change in progress, as “the results

for older speakers […] seem to reflect the traditional stigma of T-glottalisation,

particularly in intervocalic contexts” (Tollfree, 1999: 171), the profile of (t,d) deletion

is that of a stable variable in previous research, and the age of participants is usually a

low-level constraint. This study is built on an apparent-time methodology, with

speakers stratified by three age cohorts: young (18-28), middle (35-50), old (60+). This

stratification is mainly etic (by chronological age), but it includes some emic qualities

(by cultural life stages) which mirror the British society, such as the entrance of

teenagers into adulthood at 18 years old20.

The informants were also stratified by the binary category of biological sex (males vs.

females) following the traditional variationist approach in order to compare results with

previous studies. This constraint is not explored in terms of gender as this approach is

20
See Eckert (1997) who explains that culture plays an important role.

59
beyond the purposes of this thesis; indeed, the identity of participants is not the

principal object of enquiry.21

Finally, the variation of (t,d) and (t) is explored in the dimension of style – a

pivotal construct in sociolinguistic studies (Eckert & Rickford, 2001). In this survey,

style is to be intended as attention paid to speech (Labov, 1972) with attention being

“the cognitive mechanism that links social to linguistic factors” (Eckert & Rickford,

2001: 2). In some English varieties, the use of a particular speech style is adopted to

convey the high social position of speakers, such as the Speaky-Spoky Jamaican speech

style22 (Patrick, 1997). Foulkes and Docherty (2007) emphasize that the speech is not

only modified due to self-consciousness (i.e. using more of standard variants as

formality increases), but audience, interpersonal dynamics (e.g. Eckert & Rickford

2001; Eckert 2008), cognitive and interactional factors (Sharma, 2018) play a notable

role. Trudgill (1986), in the Norwich study, found variation in his own use of the glottal

stop in relation to glottal rates adopted by his interviewees. These approaches, however,

are beyond the purposes of this survey. In the present work, stylistic variation will be

elicited through sociolinguistic interviews, reading passages and word lists. The latter

aims at observing the two phonological variables when tokens are mostly realised in

isolation.

21
It should be highlighted, though, that all my participants identified as males or females, none of them
identified themselves otherwise.
22
This style is stereotypically associated with female speech. See Patrick (1997) for further details.

60
2.9. Summary

This Chapter has focused on the present sample stratification, on the field techniques

employed to enter the three speech communities, and has discussed data analysis

instruments. It has also presented a brief description of the two dependent variables,

also providing a short discussion on analytical procedure (e.g. coding), and has

discussed the independent constraints included in the model.

Chapter 3 - (t,d) DELETION: LITERATURE REVIEW

This Chapter reviews previous studies on (t,d) deletion in both US English dialects -

where (t,d) has been a “showcase variable” (Patrick, 1999: 122) - and in British English

dialects where little research on this phonological variable has been carried out.

Firstly, I will provide a phonetic description of /t/ and /d/ as well as a description of

(t,d) as a sociolinguistic variable, followed by some terminological remarks. Secondly,

we will see how this linguistic feature has diachronically developed, and then emphasis

will be placed on quantitative and qualitative analysis carried out on (t,d) across English

varieties. Since this variable has been widely researched in numerous North American

dialects, the literature will be selectively reviewed here.

3.1 Articulatory description of the plosives /t, d/

Phonetically, /t/ is a voiceless alveolar stop, articulated by obstructing airflow in the

vocal tract. A feature matrix for /t/ is [-continuant] [-sonorant] [+coronal] [-voiced]

(Roca and Johnson, 1999)1. Similarly, /d/ is an occlusive whose distinctive features are

[-continuant] [-sonorant] [+coronal] [+voiced]. Cruttenden (2001) highlights that the

lip position of /t/ and /d/ is highly affected by that of the adjacent segment, particularly

that of following vowels or glides (e.g. /w/). Moreover, the alveolar stop is sensitive to

the conditioning of the place of articulation of a following consonant. From a

sociolinguist standpoint, both /t/ and /d/ are commonly treated as one variable due to

their similar behaviour in word-final consonant clusters (for an opposite argument see

Amos et al. 2020; Pavlík 2017).

1
See Chapter 5 for a more detailed account on /t/ phonetically.

62
3.2 A brief note on terminology

In the (t,d) literature terms employed to refer to the absence of realisation of word-

final coronal stops appear to be semantically identical, such as deletion,

simplification, reduction, removal, loss or absence. However, the term deletion can

be problematic in some English varieties such as Standard Jamaican English (SJE)

or Jamaican Creole (JC). Terms like removal or loss imply that “the basilectal

speaker has an underlying representation of the missing element” (Lacoste, 2012:

69).2 By way of contrast, in the present study these terms will be used

interchangeably as in British English they do not hold such a structural linguistic

connotation. Similarly, the terms such as retention, presence, realisation,

production of coronal stops /t/ and /d/ will be used as synonyms throughout this

work.

3.3 Description of the (t,d) variable

The sociolinguistic variable referred to as ‘(t,d) deletion’ is deemed to be a form of

lenition, which weakens by having a null realisation of apical stops /t/ and /d/ in word-

final consonant cluster /Ct/ and /Cd/ or /CCt/ and /CCd/3, as shown in (1):

(1) a. […] she was jusØ calling her friend […]

b. […] during the weekenØ we’ll usually go into town […]

c. […] she is concernØ, you see? […]

d. […] she didn’Ø like the view […]

2
However, the assumption that Creole is underlyingly simpler in terms of linguistic structure does not
hold for the mesolect from a variationist perspective (See Patrick 1999 for further details).
3
(t,d) deletion differs from the more general final stop deletion rule (see Guy, 1980 p.4). Most of the
stops which fall under this rule are mainly /t/s and /d/s, as velars and stops occur in final consonant
clusters very rarely; most of the cases as listed by Guy (1980) are: /sp, sk, lp, mp, lk, ŋk,/. Apical stops
are the only plosives that can occur in a cluster preceded by many other consonants including stops,
hence the above rule is restricted to final /t,d/ deletion. Moreover, Temple (2014) also considers it
alongside other word-boundary sequences, including simple-codas, yet this approach is beyond the scope
of this PhD.

63
It is claimed that all English speakers delete /t,d/ at least occasionally, but no one does

it categorically. This phonological variable has been widely investigated in many US

English dialects and, despite slight differences in constraint ranking, phonetic and

morphological constraints were found to be largely uniform. Patrick (1999: 124) claims

that some conditioning factors mirror speech community rules, whereas others are

shared by many dialects, and others have “a basis in phonetic, functional or

phonological universals.” The pan-dialectal constraints, as summarised by Labov

(1989: 90), show that the rule application below (2) is favoured in:

(2)

a. syllable stress: unstressed > stressed

b. cluster length: CCC > CC

c. by certain preceding phonetic segments, yielding the segmental order:

/s/ > stops > nasals > other fricatives > liquids;

d. by morphological classes, with the order:

n’t > monomorphemes > semi-weak verbs > regular past verbs

e. by certain following phonetic segments, yielding the segmental order:

obstruents > liquids > glides > vowels > pause

f. by voicing agreement: homovoiced > heterovoiced.

Although morphological class has been proven to have a robust effect on (t,d), this

feature remains a phonological variable as the same surface alteration occurs in

distinct morphological contexts4, yet in the same phonological one (i.e. in word-

final consonant cluster). Indeed, the (t.d) variable rule does not apply when the past

4
In monomorphemes (mist), semi weak verbs (left), regular past tense verbs (called).

64
tense form has epenthetic schwa (e.g. wanted) or when /t,d/ is preceded by a vowel

(e.g. played).

3.4 Neutral contexts

The above rule (2) summarised by Labov (1989) does not include some neutral

contexts in which the production or the absence of /t/ or /d/ cannot be reliably

detected. The neutral environments which are usually excluded from previous (t,d)

studies concern tokens followed by homorganic tops /t/, /d/ (e.g. you can’Ø tell; I

can’Ø drop the key); occurrences followed by post-alveolar fricatives /tʃ/, /dʒ/ (e.g.

I can’t choose), and interdental fricatives /θ/, /ð/ (e.g. my husbanØ thinks that…; I

missØ the train). Interdental fricatives are considered neutral contexts as in British

English, for instance, they may be realised as their corresponding stops (Bayley,

1994). Cases of exclusion include the approximant /r/ in the preceding environment

as, in some English dialects (e.g. many British English dialects; Chicano English),

/r/ is not consonantal, as in card [kɑːd]. In a similar vein, tokens not considered are

those where preceding /l/ occurs in a cluster, yet /l/ has vocalic features (i.e. /l/

vocalisation). The features assigned to [l] and [r] in phonological theory are [+cons;

+ voc] (Chomsky & Halle, 1968), yet, in the case of non-rhotic accents or vocalised

/l/, the lateral [l] and the approximant [r] have [-cons; +vocalic] features5. In some

American dialects, clusters of alveolar nasal plus final stop preceded by unstressed

vowels are also excluded due to the intersection with the nasal flap formation rule

(Labov, 1989), hence it is onerous to establish whether the coronal stop is realised

5
The distinctive features of [r] have been a debated issue over the years. Chomsky & Halle (1968: 302),
for instance, define [r] as consonantal due to its radical obstruction, “even if it does not make a complete
contact with the roof of the mouth”. However, in their discussion the feature of [+cons] is not assigned
to glides [y] and [w]. In this respect, Fasold (1972: 89) questions that the obstruction of English [r] is
more "radical" than that of glides.

65
or reduced. High frequency lexical items are also excluded, such as and, just (e.g.

Patrick 1999). Temple (2017) argues that there are further consonants which could

be treated as neutralising contexts, yet which seem never to be mentioned. The most

remarkable is [n], realised with apical occlusion at the alveolar ridge. She claims

that when nasality occurs within a token, a following nasal may be distinguished

from coronal stops, and other stops, especially when the latter are voiceless.

However, the nasal category is non-segmental, that is it is sporadically co-temporal

with other features of the segment. In (3), for instance, the sibilant [s] seems to be

followed by aspiration plus a devoiced nasal which is clearly audible. However,

Temple (2017) argues that it is not easy to establish whether the aspiration is a reflex

/t/ followed by nasal assimilation or whether the coronal stop is deleted and the nasal

is devoiced.

(3) They are best not.

She also suggests inspecting carefully masking effects. The term “masking” (p. 135)

is used to designate an incomplete articulatory gesture which is hidden by the

articulation

of neighbouring [+cons] features. In the event of no release, the absence or presence of

/t,d/ might be questionable as she illustrates in (3):

(4) Having this lego kept me occupied for years.

In the above example, despite lack of evidence of coronal [t] in the spectrogram and

auditorily, Temple (2017: 136) claims that it is impossible to assert a categorical

deletion as

“the relatively short duration of the closure in kept compared to the /p/ of
occupied is ascribable to a rapid deceleration of speech rate and cannot
necessarily be taken as an indication of /t/ deletion”.

66
A further aspect of masking effects brought to attention is contributed by assimilation,

especially with nasals which often tend to assimilate to the place of articulation of a

following consonant, as in different plane, sound box, combined court (Temple, 2017).

While discussing the masking effects for (t,d), she states that the glottal stop is a

potential pronunciation of (t,d), even if no masked alveolar gesture can be detected.

Along this line, Temple (2017: 145) suggests that: “it would be necessary to carry

out detailed phonetic comparisons of a number of tokens with potential sequences

of glottals to establish whether there is, for example, a pattern of variation […].”

This suggestion is mostly related to the length of the glottal stop as in worked, where

[Ɂ] and [th] may be considered sequential reflexes of /k/, versus a shorter glottal

reflex found in work, for instance. The measurement of glottal length, however, is

beyond the purpose of this survey. In terms of phonetic gradience, Purse and Turk

(2016), argue that categorical6 deletion is not frequent in their Electro-Magnetic

Articulography (EMA) data of Scottish and Southern British English speakers.

Having reviewed the features of /t,d/, described (t,d) as a sociolinguistic

variable and provided an overview of the neutral contexts, let us now supply some

terminological remarks, before turning the attention to the diachronic development of

this variable.

3.5 (t,d) in Sociolinguistic theory

It is common knowledge in the sociolinguistic field that the theoretical construct of

sociolinguistic variable refers to variants which are traditionally defined as different

ways of saying the same thing. ’Variant’ refers to the phonetic realisation of a phoneme,

6
The term categorical in this context means articulatory categorical, which differs from the variationist
use of the term (e.g. 100% of application) (Temple, 2014).

67
while the more abstract representation of the source of variation is known as variable.

(t,d) deletion is one of the oldest sociolinguistic variables investigated in English

dialects on variationist grounds, and it is a classic example of a stable marker7 in

English dialects. The beginning of stable variability resembles that of a change in

progress, yet the two or (more) surface forms may co-exist for centuries, with no

evidence that one variant is pushing out its counterpart (i.e. the new item does not

replace the old variant in underlying representation). The relationship between stable

variables and social factors mainly concerns sex and social class, as outlined in Chapter

1. The age of participant, however, typically plays a marginal role in stable variables

as there is no evidence from generational change that newer forms may become

lexicalised. Reynolds (1994), with respect to (t,d), claims that the influence of solely

social factors does not explain how the change originally moved.

3.6 A brief diachronic development of (t,d)

When reconstructing a point of origin of a variable, the question to be posed is whether

all variable rules spring from diachronically categorical ones (Romaine, 1984).

According to Labov, all rules initiate as variable ones and, over time, they gradually

spread according to the environments until they reach the final stage of completion (i.e.

to become entirely regular in its application) 8. Bailey (cited in Romaine 1984) claims

that the change from categorical rule to variable rule is unnatural and argues that the

historical process of a rule can be predicted by the weight orderings of environments

(i.e. slower vs. faster). His prediction is that heavier environments are subject to a faster

application of the rule compared to lighter environments. This kind of change is usually

7
For the definition of marker, in sociolinguistic theory, see section 5.2.
8
Reaching completion most of the time may never occur as the two variants may compete for many
generations.

68
faster and earlier while, based on Bailey’s Principle 20, little quantitative evidence

means that the application is slower and later. However, Romaine (1984: 247) argues

that “there are cases where a temporally earlier rule or environment can produce a lower

quantity of output than a later one, and in which newer rules may have typically larger

outputs than older ones.” (t,d) is a well-documented process which has been a common

practice since at least the late 14th century (Wyld, 1927). Evidence from earlier

documentation shows that bimorphemic clusters spring from the suffixation of /t,d/ to

some verb stems, a phenomenon which is also known as Germanic dental preterite 9.

Romaine (1984) shows that in the 16th century, preterite and past participle forms end

in -it, with only few items ending in a cluster as promist. Verbs which are usually

referred to as “semi-weak” in the (t,d) literature (i.e. stem vowel change and suffix past-

marking) can be found with /t/ absence in Middle English, from the beginning of the

period. An early example (e.g. slep) arose from the Peterborough Chronicle, as

reported in Romaine (1984):

He lai an slep in scip

Crist and his halenchen slep.

Wright (1905) notes that /kept/ is commonly realised as /kep/ in the Midlands as well

as in northern dialects, yet this is not due to /t/ loss, while the loss of apical stops /t/

and /d/ after a preceding /n/ as in blyn ‘blind’ in also found in the Modern period (Wyld,

1927). Wyld also provides numerous examples of /t,d/ absence word-medially followed

by consonants in the 15th-18th centuries, such as freenly ‘friendly’, hansome

‘handsome’, Chrismass ‘Christmas’10.

9
See Prokosch (1939) for further details and for debated issues.
10
Similarly, Cruttenden (1994) reports the loss of apical stops in medial position in clusters of three
consonants in Present-day English, such as exactly, handsome, windmill, handbag, friendship,
landlord, restless, landscape, etc.

69
 However, as Reynolds (1994) notes, all of the examples provided by Wyld refer

to monomophmes or compounds. During the 14th century, /t,d/ did not occur in final

consonant clusters, except for a few irregular Old English (OE) verbs, such as geseald

‘sold’, geteald ‘told’, as well as gelegd ‘laid’. Indeed, past-marking in both OE and

Middle English (ME) was /-ədə/ (e.g. worshipede) (Luick, 1921). The schwa started to

disappear in late 15th century, at the end of the ME period and this linguistic change

was fully completed after three centuries.

Milroy (2007) reports that /d/ absence after [n] appears around 1700. Its

reduction is visible in hypercorrect spelling (e.g. gownd for ‘gown’) as well as in

hypercorrections which survived over years, such as sound from ME soun < Fr soun.

Romaine (1984: 242), however, argues that there is “no reason to reconstruct a period

in the history of English in which final t/d was uniformly present in mono- and

bimorphemic clusters.” She suggests that there is no reason for doing so due to lack of

evidence of a prior state of categoricality.

3.7 Empirical findings from North American studies

(t,d) deletion as a sociolinguistic variable has raised interest among variationists since

the 1960s. Empirical findings show that it is markedly conditioned by the following

phonological context and the grammatical status of the word-final apical stop, while

the preceding phonological environment has been considered a “tertiary constraint”

(Guy, 1980: 20). This variable has been initially explored through quantitative analysis

and examined under the effect of external constraints in African American Vernacular

English (AAVE) in New York City (Labov & Cohen 1967; Labov et al.1968).

The variability of (t,d) has been then investigated in Detroit (Wolfram, 1969); in

Washington (Fasold, 1972); in New York (Guy, 1980), Philadelphia (Guy 1980;

70
Tamminga 2016); and Buckeye English (Ohio) (Tamminga & Freuhwald, 2013); also

in first language acquisition by children (Labov 1989; Roberts 1994; Lacoste 2012)

and second language acquisition (Bayley 1991, 1994); in English varieties influenced

by Spanish, such as Chicano English in Los Angeles (Santa Ana, 1991), Tejano English

in San Antonio (Bayley, 1994); in creole varieties, such as mesolectal Jamaican Creole

in Kingston (Patrick, 1999), and Standard Jamaican Creole (Lacoste, 2012).

3.7.1 (t,d) in AAVE

Earlier studies on (t,d) were conducted in non-standard “Black English”.

Labov and Cohen, (1967) interviewed 100 random speakers in the main area of

South Central Harlem, and investigated (t,d) deletion along with other variables, such

as (-ing), /r/, -s absence of copula, etc. With respect to (t,d), only two categories were

distinguished for the following phonological environment: following vowel vs. non-

following vowel, with the latter including (liquids, pauses and maybe glides). A binary

distinction also concerns the grammatical status of /t, d/, resulting in mono-morphemic

vs. bi-morphemic items. Results show that the effect of the morphological class

surfaced as a remarkable constraint, with regular past tense verbs (e.g. passed)

undergoing simplification in all age groups, yet the rates of deletion increased in

monomorphemes. Working-class speakers exhibited a stylistic shift in

monomorphemes in the non-following vowel and following vowel environments;

whilst, in the speech of middle-class members the cluster reduction increased when

/t,d/ was followed by consonants.

To provide a deeper insight into non-standard English of Negro and Puerto

Rican speakers, Labov et al. (1968) examined (t,d) in 3,359 tokens collected in Central

Harlem. The regularity of (t,d) absence, once again, was found in each group of

71
respondents, regardless of the age level or the neighbourhood they lived in. This study

has shown that some speakers had categorical deletion of /t/ when the coronal stop

followed the voiceless sibilant /s/ (i.e. –st). Also, this work has revealed that the

simplification of apical stops may be influenced by the presence or absence of voicing

in the consonant cluster, and that (t,d) is less likely to undergo deletion if the first

member of the cluster is a sonorant. With respect to the grammatical category, 74% of

past-marking presence surfaced in regular verbs; whilst, unmarked cases are believed

to be entirely conditioned by phonological factors. So far, semi-weak verbs are not

included. It is with Labov et al. (1980) that we have a three-way distinction between

monomorphemic, ambiguous (or semi-weak), and bimorphemic forms.

Wolfram’s (1969) study also focused on the cluster simplification in AAVE in

Detroit by taking into account the ethnicity, social class, age and gender of the

participants. He collected his data among 12 lower working-class speakers, 12 upper

middle-class speakers and, for comparison, 12 white upper middle-class speakers. The

number of tokens is unknown, as Patrick (1999: 162) also reports in a detailed

comparative table between (t,d) studies. In Wolfram’s study, the following phonetic

environment seemed to have a notable impact on the cluster reduction, despite the lack

of a hierarchical ordering. It has to be noted, though, that following pause was treated

as vowels. Working-class speakers exhibit the highest deletion rate in monomorphemic

tokens followed by consonants (97%), while /t,d/ in bi-morphemic words followed by

vowels, was less likely to be deleted among middle-class members. Wolfram’s (1972)

data did not provide a clear explanation on how much the age of the informants may

affect (t,d) deletion. He claims that the reduction of the second item in the consonant

72
cluster occurs if: (1) the second member is a stop; (2) the two items share the same

voicing specification11.

Along the line of Wolfram (1969), Fasold (1972) divided the following segment

into consonants vs. non-consonants showing an analogous outcome, that is following

consonants are the most preferred environment for (t,d) reduction. Initially, Fasold

(1972) distinguished following pause from following consonants, yet due to the nearly

identical result between the two categories, they were collapsed. His corpus was

collected in Washington DC to investigate tense marking in Black English. 47 working-

class informants of all ages were interviewed through one-to-one interviews, reading

passages and word games, and 883 tokens were analysed. He treated separately the

simplification of [d] after vowels, whilst he did not isolate the [st] cluster as Labov did.

With the use of chi-square tests, he aimed to provide a correlation between linguistic

and social factors, and to also establish the role of variability within the linguistic

analysis, distinguishing truly random variability from conditioned variability. His

findings show that deletion occurred the most when /t,d/ was followed by spirants and

stops, and that preceding /s/ (e.g. cast) increased deletion more than other voiceless

spirants. Voicing, however, is not a significant constraint for the Washington data.

More deletion occurred in bi-morphemic tokens followed by consonants (76%) than in

bi-morphemic ones followed by vowels (29%). Irregular verbs (i.e. left) were treated

separately, showing a 98% rate of past-marking. He also noticed that the voiced

alveolar [d] is devoiced in verbs, such as send-sent. Similar findings are reported for

Jamaican Creole (Patrick, 1999), where the devoicing processes is extended to five

verbs (see section 3.4.4 for more details).

11
[ld] clusters were excluded in the Detroit study (Wolfram, 1969).

73
Patrick et al. (1996), in the U.S. South, investigated the behaviour of (t,d) in 11 ex-

slave elders, who exhibit analogous percentages of (t,d) absence in monomorphemes

followed by consonants (56%) and monomorphemes followed by vowels (55%).

Lower reduction rates are reported for bi-morphemic words: 44% for those followed

by consonants, and 26% for those followed by vowels. The increasing of deletion was

associated to the increasing of sonority, with preceding /l/ being the most conditioning

segment.

3.7.2 (t,d) in US English dialects

Neu (1980) and Guy (1980) carried on with the distinction between monomorphemes

(i.e. mist), regular verbs (i.e. called) and ambiguous verbs (or semi-weak verbs, e.g.

told). The latter refers to those verbs which undergo a stem vowel change as well as

past-tense affixation. Neu’s (1980) findings did not show a marked deletion difference

in ambiguous verbs. She examined 2,217 tokens in the speech of 15 white informants

from the U.S. North12, and employed chi-square tests to validate the ranking constraint.

Her study shows that the lexical item and turned out to be nearly categorical, indeed,

when it was included in the monomorpheme category, deletion rates increased sharply.

She explored (t,d) in both following manner and place of articulation. The latter,

however, was not marked as a significant constraint. Her findings report a high-level

deletion in following consonants, with no consonant class differing significantly from

the others. With respect to the preceding phonetic segment, results were reported in

12
The participants were from: California (n = 5), Ohio (n = 2), Michigan (n = 2) Baltimore (n = 2),
Nebraska (n = 1), Missouri (n = 1), Massachussetts (n = 1) and New York City (n = 1). As for the age
of participants, 5 speakers were aged between 19 and 23, 8 speakers aged between 27 and 35, 2
speakers aged between 48 and 53

74
two parts, as they differ for males and females. Males exhibit the following ranking:

sibilant > nasal [excluding and] > stops; by contrast females’ hierarchy shows partially

reversed result with preceding stops leading the ranking: stops > sibilant > nasal

[excluding and].

Guy (1980) conducted his (t,d) study in the speech communities of Philadelphia

(19 speakers) and New York (4 speakers). In this study, the consonant category of the

following phonological environment is broken down into: consonants, liquids, glides,

and vowels, while for the preceding phonological segment a five-way coding was

employed: sibilants, non-sibilant fricatives, nasals, stops, and laterals13. Results show

that a following pause has the same effect as following consonants in promoting

deletion amongst New Yorkers, whilst for the Philadelphians a following pauses

behaves like a vowel in disfavouring (t,d) absence. The preceding segment is

considered a relatively weak constraint, with more reduction when /t,d/ follows /s/, and

more coronal stop retention when preceded by the lateral /l/. With respect to the

morphological class, deletion occurred more in monomorphemes than regular past

tense verbs, yielding the ranking M > A > P 14.

Guy & Boyd (1990) found that the treatment of semiweaks changes

significantly with age, showing a declining probability of deletion as age increases.

This highly significant correlation exhibits age-grading only for this class of words.

Other external factors, however, (e.g. sex, social class, geographical background,

ethnicity as well as style) did not play a salient role. Deletion in the semi-weak class

revealed a good clustering of children (aged 0 - 18), with a probability value above .75,

while younger adults (aged 19 - 44) and adults (aged 45+) exhibit a mean of .65 and

.60, respectively. These findings were summarised into three deletion patterns: pattern

13
Preceding /r/ was coded as a vowel.
14
M (monomorphemes), A (ambiguous), P (past tense of regular weak verbs).

75
I, with high probability of deletion in semiweaks; pattern II, with a probability value

between .60 and .75; pattern III, with a probability lower than .60. From a

morphological acquisition standpoint, children exhibit nearly categorical /t,d/ reduction

as, for them, final coronal stops are not underlyingly present. This means that the past

tense is only marked by the stem vowel change15. The differentiation of semiweaks

from strong verbs begins in the second stage of acquisition (pattern II), where speakers

notice /t,d/ in the underlying forms, yet those verbs are still treated as morphologically

uninflected. In the last stage (pattern III), speakers can finally distinguish between the

apical stops in the semi-weak class and those in regular past tense verbs (e.g. -ed).

This pattern was also found by Labov (1989) in a study of a single family, where

a 7-year-old child was found to exhibit the same deletion rates as their parents in all

morphological classes, yet in the semiweak class the child had a greater rate of

reduction.

Roberts (1997) examined (t,d) in language acquisition to find out how this

variable behaves in child language and when it is acquired. She interviewed 16 children

(aged 3-4) as well as 8 Philadelphian mothers. The children were recorded in their

nursery school, located in a working to lower middle-class area of South Philadelphia.

Data was coded by morphological status and following phonological environment 16,

employing the use of GoldVarb 2.0 to carry out the linguistic analysis. The data was

firstly analysed as a group, and then by individual. 13 out 16 children were individually

analysed in relation to the following environment, whilst all the 16 children were

included in the individual analysis for grammatical contexts. Overall, her results reveal

that each child had acquired the following phonetic constraint on (t,d), exhibiting

15
According to Guy & Boyd (1990) children, in their mental lexicon, only distinguish between weak
verbs and strong verbs.
16
The breakdown of the following environment is: obstruent, liquid, glide, vowel, or pause.

76
retention of apical stops when followed by a pause. This finding, linked to the fact that

deletion before following pause varies geographically (Guy, 1980) and that

Philadelphians take seriously the origin of their dialect, suggest that “it is a dialect that

is being learned rather than a universal process being applied” (Roberts, ibid: 362) as

children’s speech pattern, by the age of three, resemble their parents even in the

retention of /t,d/ before a pause.

With respect to the morphological class, while Guy & Boyd (1990) show a

categorical deletion in semi-weak verbs among the youngest children, in Roberts’

findings /t,d/ reduction in the semi-weak class among children equally occurred as in

monomorphemes. Since there seems to be a deviation from the adult pattern, she

suggests that the children, as a part of a rule-learning process, do not conform to

universal tendencies. The children were divided into groups: ten 3-year-olds and six 4-

year-olds, yet no significant difference was detected between them. Finally, (t,d) was

found to be firstly acquired in terms of grammatical and phonological constraints rather

than by style.

The correlation between grammatical and phonological constraints has been

recently explored by Tamminga (2018). Specifically, she proposes a modulation of the

following segment effect on coronal stop deletion (CSD) by syntactic boundaries. She

analysed 118 interviews of the Philadelphia Neighbourhood Corpus (PNC) (Labov &

Rosenfelder, 2011). The corpus used is composed of white upper working class and

lower working class participants from Philadelphia containing 15,874 tokens for (t,d).

Since in white Philadelphian English there are no /rt/ and /rd/ clusters, preceding /r/

was excluded before coding. The syntactic boundaries coded are of two types: strong

vs. weak. The former includes Matrix CP + Matrix CP; Matrix CP + Conjunction + Matrix

CP (e.g. And then I make my crust ‖ and I fill it up) sequences referred to the position of

77
apical stops between two independent (matrix) clauses; High adjunct + Matrix CP (e.g. When

you get old, ‖ everything bothers you) refers to the presence of a preposed adjunct (e.g.

temporal or adverbial phrase); Matrix CP + High adjunct (e.g. I thought he was a good friend

‖ until that point) is a right-adjoined adjunct placed higher than the verb which carries the

target segment. The weak syntactic boundary refers to Verb + Direct (e.g. You can’t find ‖ a

cork today) object sequences. Results, obtained from monosyllabic target words, show that

the influence of the following segment is stronger across weak syntactic boundaries than across

strong ones – a result which is in line with the Production Planning Hypothesis (Tanner et al.

2017). While the inhibition of (t,d) before a following vowel is significantly weakened when

the apical stops occur across stronger syntactic boundaries, deletion before a following

consonant appears “more stable” regardless of the boundary type.

3.7.3 (t,d) in varieties influenced by Spanish: Chicano English and Tejano

English

This variable also received notable attention in many Hispanic varieties. Hartford

(1975), analysed (t,d) in Mexican American English of teenagers in Gary, showing that

men reduced /t,d/ clusters more than women. Galindo (1980) explored the same

variable in Austin, where the same pattern of deletion was found between Mexican

AmE and English.

Santa Ana (1991,1992) investigated (t,d) in the English of 45 Los Angeles

Chicanos revealing an effect of syllable stress, with unstressed syllables promoting

more deletion than stressed ones. He tested the existence of an exponential relationship

in the rate of coronal stop deletion. 4,857 tokens were gathered, with 3,724

monomorphemes, 297 semi-weak verbs and 83617. With respect to age, adolescents

17
Details about his findings will be provided in section 3.4.5

78
seemed more likely to adopt vernacular features than their parents, who are less likely

to reduce /-t,d/ in cluster position. Bayley (1994), examining Tejano English, explains

that although their parents are all bilingual, English is not usually adopted as their

vernacular.

3.7.4 (t,d) in Creole varieties: mesolectal Jamaican Creole and Standard

Jamaican Creole

The absence of final apical stops has also been subject of discussion for Creole

languages18 (Akers 1981; Patrick 1991, 1999; Lacoste 2012). Akers’ (1977) results

reveal a high rate of /t,d/ absence in JC word-lists (99%) in contrast to other forms

investigated (54%), while in SJE word-lists rates of deletion are lower in weak verbs

(53%), but still higher than other forms (18%).

A detailed analysis of this variable, in mesolectal JC, is provided by Patrick

(1991, 1999). He examined (t,d) at the intersection with past-marking by analysing

2,323 tokens for 10 speakers in Kingston. Internal factors include preceding19 and

following phonetic segments as well as morphological class; while external factors

encompass age (from 14 to 82), sex, social class, style. His findings are parallel to the

pan-English effects, with more deletion before following consonants (87%) 20 and less

reduction before following vowels (63%). Similar to New York (Guy, 1980), a

following pause (70%) boosted /t,d/ deletion in mesolectal JC. Patrick (1991, 1999)

also points out that the exclusion of /nt/ tokens is an onerous decision to make, as the

majority of apical stops in final cluster preceded by nasals are negative contractions

18
An early suggestion was that “Creoles […] have no initial or final consonant cluster” (Romain 1988,
cited in Patrick 1999). Patrick (1999) shows that underlying clusters are present in the mesolect and often
may be reduced, especially before following consonants.
19
JC is non-rhotic in the environment C__V (Wells, 1973, cited in Patrick 1999), hence post-vocalic /r/
was excluded from the analysis.
20
A slightly higher deletion rate was found for following rhotic (78%) compared to that of glides (74%).

79
(i.e. morpheme n’t)21 and the latter, as also claimed by Labov (1989), heavily favour

deletion. Along this line, Patrick (1999: 142) emphasises that n’t morphemes

“generally show the highest rates of (TD)-absence of any morphological class.” The

latter, indeed, exhibits 87% of /t,d/ simplification in negative contractions, despite the

low number of tokens (n = 525) compared with monomorphemes (n = 1,358). He

treated separately a small class of irregular devoicing verbs (e.g. send, spend, lend,

bend, build) which end in a cluster of sonorants, but form the past by devoicing the

voiced alveolar stop. The ranking n’t (87%) > regular verbs (79%) > monomorphemes

(71%) > semi-weak verbs (59%) > irregular devoicing (38%) reveals a very high

probability of deletion in weak verbs, which usually tend to favour /t,d/ retention.

Through the intersection with past-marking, Patrick (1991, 1999) proves that, in JC,

the high deletion rate (79%) in regular verbs (e.g. called) is prompted by morphological

absence (i.e. non-marking of past tense) rather than phonological deletion. Similarly,

in semi-weak verbs tense marking surfaced only almost 50% of the time. His study also

shows that ranking of constraints was consistent across the social and linguistic range

of informants.

Lacoste (2012) explored the acquisition of (t,d) among children in rural Jamaican

schools (Landforest, Bareton and Damont) with particular focus on performance of

SJC. Recordings were conducted within the classroom, including teacher’s speech, and

when the class was no longer engaged. The latter only involved children. Results with

respect to the preceding segment show that the fricatives + stop sequence was acquired

first by Landforest children as the findings reveal 50% cluster absence, while the lowest

rate of (t,d) deletion was found in preceding nasals in Bareton and Damont with a rate

of 64%. Interestingly, Landforest children do not show evidence of acquisition of stop

21
Only few studies have included n’t tokens in the analysis (e.g. Labov 1989, Patrick 1999). They are
usually excluded from sociolinguistic studies of (t,d) owing to possible interaction with -nt tokens.

80
+ stop sequence as deletion is categorical in this context, reaching 100% of /t,d/

absence. Overall, Lacoste (2012) claims that Bareton children adopt clusters in a

homogeneous way (76% - 93%), whilst Landforest and Damont children show a more

irregular absence rate which fluctuate between 50% - 100% for Landforest, and 68% -

100% for Damont. Teachers’ rate of absence differ from that of children, as preceding

stops are consistently realised in the teachers’ speech giving a rate of /t,d/ absence equal

to 51%.22 The closest rate of performance between teachers (69%) and children (70%)

relates to fricatives. Results for the following environment exhibit a higher /t,d/ absence

after consonants, whilst vowels disfavour deletion. Following pause behaves like

vowels and disfavours at .378. Similar results were also reported for all teachers. Rates

of /t,d/ presence were lower in homovoiced than heterovoiced tokens, yet voicing

agreement alone was not found to be a powerful constraint. Social factors were

minimised in this study as the socio-economic background of speakers is similar and

all children belong to the same age cohort.

All the above studies show considerable agreement in the influence of

environmental factors and provide precise quantitative accounts on these

conditioning effects. But why is the ordering of environments so consistent? The

section to follow provides an explanation through quantitative theoretical

predictions.

22
The reduction rate of /t,d/ in childrens’ speech is 89%.

81
3.7.5 (t,d) explored on theoretical grounds

To explain whether rates of deletion in semi-weak verbs should be closer to

monomorphemes or regular past verbs, and whether (t,d) absence in regular verbs

should be a function of deletion rates in monomorphemes, Guy (1991) adopts lexical

phonology (Kiparsky, 1982, 1985). Under the lexical phonology approach, (t,d) is

explored at a derivational level at which the final consonant cluster is acquired. Guy

(1991) applied this theory to empirical findings obtained from seven native English

speakers23, gathered by means of sociolinguistic interview. The four remarkable

features of this theoretical framework related to Guy’s (1991: 6) purpose are listed

below:

1. “multiple levels of lexical derivation;

2. interleaving of morphological and phonological processes;
3. phonological rules may apply at more than one level; and
4. bracket erasure occurs at the end of each level.”

This framework suggests that the structure of lexical derivation is systematically

arranged into levels, where all morphological processes and many phonological ones

are carried out in the lexicon. Within the lexicon there are two or more ordered levels

but, as claimed by Harris (1989), two levels are enough to account for variable patterns

in English (in Guy 1991). Hence, irregular inflections (e.g. found) occur at level one,

whereas the attachment of regular inflectional affixes occurs at level two (e.g. called).

Apical stops in monomorphemes are underlyingly present from the earliest stage of

derivation (e.g. mist); semiweaks, which are subject to stem vowel change, are treated

as undergoing affixation at level one (e.g. left); regular past verbs undergo /t,d/

23
The informants speak North AmE varieties.

82
affixation at level two. Phonological rules, instead, may not be allocated to particular

levels. With the fraction 1 – pa , or pr, Guy (1991) measures retention probability.

He argues that coronal features which were introduced early in the derivation are more

likely to undergo deletion, whilst the proportion of items where the rule has never

applied will be smaller at each level. Therefore, this theory predicts an exponential

development of nonapplication rates for variable rules which apply to different

morphological classes, with non-identical derivational histories. This predicts that if a

deletion rule operates 50% of the time, it is expected to find a 50% of pa (or retention)

for those forms in which the rule operated only once; in forms where the rule operated

twice the retention rate would be 25%; while, when the rule operated three time, rates

of retention will be 12.5%. This means that regular verbs undergo /t,d/ deletion once,

after affixation at level two and bracket erasure 24, as at earlier levels apical stops are

not morphologically present. Monomorphemes, instead, undergo the deletion rule three

times: at an earlier morphological stage, after affixation at level one and bracket

erasure, and after affixation at level two. Along this line, Guy explains that speakers

whose past-marking of semiweaks is obtained by affixation at level one are subjected

to the rule application twice: after undergoing affixation at level one and bracket

erasure as well as after affixation at level two25. Guy’s (1991) prediction of deletion in

such framework is well-supported by his empirical results exhibiting values of

retention equal to .914 for monomorphemes; .877 for semiweak verbs, and .918 for

regular verbs26 and he argues that “the rate of application in each class is not

independent of the other classes” (Guy, 1991: 19).

24
That is, the rule application occurs postlexically.
25
However, not all individuals treat semiweaks this way due to an age-graded acquisitional pattern
(Guy & Boyd, 1990).
26
These are the estimate of pr in four adults.

83
 Santa Ana (1992) tested Guy’s approach and provides evidence of the

exponential hypothesis for 45 Los Angeles Chicanos. His findings fit tightly within the

lexical phonology framework confirming Guy’s (1991) predictions.

As regards the following phonetic segment, Guy’s (1991) hypothesis entails a

novel prediction. He argues that combining following /r/ and /l/ as sharing a common

sonority feature is not quite accurate. Indeed, he split the liquid category by treating /l/

and /r/ separately due to syllabification processes. This means that while *tl- and *dl-

onsets are prohibited in English, tr- and dr- are acceptable syllable onsets as both apical

stops can resyllabify onto the following segment. This prediction is confirmed by his

data revealing that /l/, which favours deletion at .80, patterns with obstruents in

triggering deletion, whereas a following /r/ disfavours it (.42)27.

Fruehwald (2012), with respect to the morphological class, proposes a

redevelopment of this class status. He proposes that the treatment of semiweaks in

classical (t,d) deletion studies ought to be revised as, along with regular verbs, they

should be treated as being identical in phonological deletion, yet differing in terms of

morphological realisation. The Kiparsky approach of Competing Grammars refers to

the notion of speakers possessing different grammars, or different internal grammars

based on the use or certain forms or class forms which diverge amongst speakers from

different regions, generation, socioeconomic class etc. Under a Competing Grammars

approach, Fruehwald (2012) investigates whether (t,d) in semiweak verbs is due to

morphological absence or phonological deletion, as illustrated in the figure to follow.

27
In Tejano English, however, Bayley (1994) reports a categorical realisation of /-d/ in the context of a
preceding /r/. In Chicano English /-d/, in the same phonological environment, is deleted even though at
a very low rate. The influence of Spanish, in both Tejano and Chicano English, concerns the effect of
preceding /r/. Indeed, the distinctive features of /r/ Spanish are [+consonantal, -vocalic], while in
General American English /r/ is [+consonantal, -vocalic, + central].

84
Figure 3.7.1. Competing morphological grammars. Adapted from Fruehwald (2012).

To account for morphological presence, he adopted the following formula:

Observed Semiweak (t,d) presence

Regular past (t,d) presence

showing 93% of /t/ presence in semiweak verbs in the Buckeye Corpus, with only 7%

of morphological absence. As outlined in section 3.7.4., this morphological absence

was also explored in Jamaican Creole (Patrick, 1991) whereby the highest deletion rate

occurred in regular past verbs. This tense marking surfaced only 50% of the time, hence

Patrick (1991) concluded that, in JC, the past tense is variably marked.

Similar theoretical approaches cannot be applied to the preceding and following

segment effect because, as Guy (1991) claims, they do not spring from derivational

history but are due to phonetic and phonological properties, as well as phonotactic

principles. Therefore, to theoretically explain why the most deletion occurs before a

following consonant rather than before a following vowel, and to explain why the effect

of pause varies depending on the dialect, Kiparsky (1994) employed Optimality Theory

(OT)28. The main principle of OT is that the constraint inventory is universal, or innate,

and it is part of a universal grammar (UG). A pivotal notion of this theory is the

possibility of constraint violation if constraints are at odds with other constraints,

28
Unlike the Variable Rule (VR) which is non-categorical and non-deterministic, OT is non-
categorical and deterministic.

85
leading to variable results. In this output-based model the input is retrieved in the

output. Through graphical evaluation, OT illustrates the surface form and other

potential contenders which may be logically obtained, as illustrated in the table below.

/Input/ Constraint 1 Constraint 2 Constraint 3

Candidate 1

Candidate 2

Table 3.7.1 Typical layout of an OT representation.

To answer the above questions, in relation to (t,d), Kiparsky (1994) adopted three

constraints: (1) SYLL-WF29, divided into (a) *COMPLEX (no tautosyllabic clusters); (b)

*CODA (no codas); (2) ALIGN, divided into (a) ALIGN-LEFT-WORD (no

resyllabification across word boundaries); (b) ALIGN-RIGHT-PHRASE (phrase-final

consonants are not deleted); and (3) PARSE30. As part of the theory, the constraints are

assigned a hierarchical order, and in case of contrast between them, the higher-order

constraint wins. The constraint ranking reported by Kiparsky (1994) shows that there

is no deletion if PARSE >> SYLL-WF, while if SYLL-WF is ranked higher than PARSE,

each sequence implies “a different categorical deletion pattern”, as illustrated below:

1. SYLL-WF >> ALIGN>> PARSE, yields cluster reduction in all positions;

2. ALIGN >> SYLL-WF >> PARSE, yields /t,d/ absence before consonants and vowels;

3. SYLL-WF >> PARSE >> ALIGN, yields/t,d/ deletion before consonants and pause.

To account for the preceding segment, Guy & Boberg (1997) took a generalised version

of the Obligatory Contour Principle (OCP). The OCP effect predicts that a preceding

29
SYLL-WF stands for Syllable-well-formedness (Kiparsky 1994).
30
In OT, the constraint PARSE refers to “underparsing” the input, and in other cases corresponds to a
“deletion repair strategy” (Prince & Smolensky, 2004: 48).

86
segment triggers deletion if it shares the same features with /t,d/ [+cor, -son, -cont],

creating OCP clashes. Guy & Boberg’s (1997) findings, with a total number of 1,071

tokens, exhibit more deletion after preceding sibilants [+cor, -cont], stops [-son, -cont]

and /n/ [+cor, -cont] which share two features with the target segments, than after

preceding fricatives /f, v/ [-son], /l/ [+cor], /m, ŋ/ [-cont] which share only one feature.

In regard to the preceding segment, Santa Ana (1996) argues that its effect on

the coronal stop deletion is governed by a theoretical sonority hierarchy:

stops > fricatives > sibilants > nasals > /l/

according to which less sonorous preceding segments (stops and fricatives) tend to

favour deletion, whereas more sonorous segments disfavour it. However, as Patrick

(1991) noted, there seems to be no explanation as to why the place of nasals in the

hierarchy is usually higher than that of fricatives in previous (t,d) studies (e.g. Labov

1989). Patrick’s (1999) findings, instead, reveal that, in JC, the sonority hierarchy is

re-ordered, yielding the ranking: sibilants > stops > fricatives > nasals > laterals.

Lexical frequency was also found to have an impact on the phonological

conditioning of sounds (Bybee, 2002). It is argued that even if a gradual change will be

lexically regular, it shows lexical diffusion when it is in progress. Lexical diffusion

relates to the influence a sound change has on the lexicon. A rapid sound change will

affect all words in a language in the same way; while, a gradual lexical change would

affect words at different rates or different times31. Schuchardt (1885) claims that a

sound change influences high-frequency words considerably more than low-frequency

31
The Neo-Grammarian prediction that a sound change is lexically regular is supported by the fact that
when a dialect undergoes a sound change, this change is constantly spread across all items that have the
suitable phonetic context (Bybee, 2002). Labov (1994: 542) suggested two types of sound change: (1) a
gradual regular sound change is influenced by phonetic reasons, it is not lexically conditioned and has
no social awareness; (2) lexical diffusion change occurs with a rapid replacement of one phoneme for
another in words where that phoneme is present. He also claims that this can occur often “in the late
stages of internal change that has been differentiated by lexical and grammatical conditioning”. (t,d)
deletion is treated as a “lexical diffusion” change (Labov, 1994) probably due to the rapid phonological
reduction of a phoneme.

87
ones. Bybee (2002) tested this theory on (t,d) deletion on data from AmE, arguing that

a phonetically conditioned change which affects high-frequency items before low-

frequency ones can be explained through a phonological representation model which

permits a change to be both phonetically and lexically gradual. Her assumption that

(t,d) is a change in progress contrasts with sociolinguist findings. The data employed

for this analysis include n’t morphemes, regular past tense verbs, and unstressed -nt

tokens. Her findings reveal that deletion occurred at a higher rate in words with high

frequency (54.4%) than in a number of words with low frequency (34.3%). Similar

findings were obtained by Jurafsky et al. (2001) in a Switchboard corpus 32 of 2,042

monosyllabic tokens, where word frequency was found to be one of the strongest

statistically significant constraints.

A positive correlation between (t,d) deletion and lexical frequency was also

found by Guy et al. (2008), who captured the frequency effect by using frequency

counts taken over the corpus which they analysed. Their results show interaction

between the frequency constraint and morphological class: the cluster reduction rose in

monomorphemes with the increase of frequency, whereas lexical frequency had no

effect on regular past tense verbs. Higher deletion rate was found in cliticized negative

-n’t, while the lexical item and was found to interact with the following word, when

the latter was a hesitation. Indeed, in phrases like ‘and uh’, /d/ is more likely to be

retained.

Walker (2012) revisited lexical frequency in (t,d) by examining four different

corpora, including frequency measures taken over the analysed dataset. His conclusion

is that when lexical frequency is measured externally there is no surfacing effect.

32
This corpus is made of telephone conversations between monolingual American English speakers.

88
Renwick et al. (2014) explored lexical frequency, along with morphological class,

adopting the Audio British National Corpus. Data consists of word pairs (e.g. past

eleven) with the target feature for a total of 5191 tokens (2037 test tokens). To capture

phonetic reduction, they also measured intensity (dB) each 5ms. Results reveal that the

intensity of frequent words is shorter and higher compared to that of less frequent words

whose intensity is longer and lower. Renwick et al. argue that less frequent words are

realised more slowly limiting, therefore, rates of (t,d) deletion.

Purse & Tamminga (2019) take a different stance on the measurement of

frequency. Typically, in sociolinguistics, lexical frequency is measured by Wholeword,

that is a separate frequency value is assigned to each unique affix whatever semantic,

morphological or phonological relation there is between the tokens. In other words,

flower, flowers, meat and meet obtain a different value, while right (direction) and right

(verb) obtain the same value. As Root frequency, instead, “the frequency of each item

is calculated as the sum of Wholeword frequencies sharing its stem”. This type of

lexical frequency measure best captured the (t,d) data from the Philadelphia

Neighborhood Corpus (PNC) (Labov & Rosenfelder, 2011). They also measured

Conditional frequency, that is “the proportion of an item’s parent Root that is

constituted by a particular Wholeword, and represents the frequency of a Wholeword

given the Root.” Purse & Tamminga’s (2019) results, hence, reveal that previous

hidden frequency effect surface if frequency is measured by Root.

This section has explored (t,d) on theoretical grounds through Lexical Phonology

(Guy, 1991), Competing Grammars (Fruehwald, 2012), Optimality Theory (Kiparsky,

1994), Obligatory Contour Principle (Guy & Boberg, 1997), sonority hierarchy (Santa

Ana, 1996), with a different take on the sonority hierarchy proposed by Patrick (1999),

89
and word frequency (Bybee 2002; Guy et al. 2008; Walker, 2012; Purse & Tamminga

2019).

3.8 (t,d) in British English dialects

While this variable has been extensively investigated both empirically and theoretically

in many North American dialects, it has received comparably little attention in the UK.

Indeed, it was explored only in York (Tagliamonte & Temple, 2005), Manchester

(Baranowski & Turton, 2020), in standard British English (Pavlík, 2017), in Mersea

Island (Amos et. al, fc), and in Tyneside English (Woolford, 2018). Pavlík (2017) and

Amos et al. (fc), however, distance from the usual treatment of (t,d) as one variable.

3.8.1 York

The first attempt to replicate North American (t,d) studies on a British English variety

was carried out by Tagliamonte & Temple (2005). To account for this sociolinguistic

variable the data was taken from the York English Corpus (Tagliamonte, 1998) which

includes native British English speakers living in York or nearby the city. The sample,

for this analysis, includes 38 speakers equally distributed in terms of sex, while age is

treated as a continuous variable (between 16 and 91 years of age). In addition to the

usual neutral contexts, questions, negatives, interrogative constructions and following

interdental fricatives (e.g. called them) were excluded from analysis. 1232 tokens - 125

of which were realised with glottal stop or glottalization33 – were analysed. The coding

for the morphological class differentiates between monomorphemes (e.g. mist),

33
These reflexes of /t,d/ were treated as nonapplication of the reduction rule.

90
semiweaks (e.g. kept) and past tense verbs (e.g. missed), while suppletive (e.g found),

replacive (e.g. sent) and preterit went were coded separately34.

Similar to previous US findings, results confirm the greatest conditioning effect of the

following phonological segment, with obstruents (.83), glides (.70), /r/ (.60) favouring

deletion, whereas vowels (.30) and pause (.20) disfavour it. However, the neutral

position of /l/ (.50) in York is not consistent with the re-syllabification process

advanced by Guy (1991). The preceding phonological segment, a usually weak

constraint, surfaced as the second most significant predictor in York yielding the

ranking:

sibilant > nasal > liquid > stop > non-sibilant fricative,

where sibilant is the only favouring factor. The most surprising finding from the York

data is linked to the morphological class. As outlined in previous sections, the

morphological class is one of the strongest paradigmatic constraints of (t,d) in all North

American (t,d) studies. By contrast, in York, the morphological effect failed to reach

statistical significance. Even after they reanalysed the data leaving out preterite went,

replacive verbs and strong preterites, the outcome did not change as the morphological

class was still not selected as a significant predictor. The interaction between

phonological and morphological categories revealed that the following environment

operates similarly for each morphological class; the preceding segment was not marked

as significant for the semiweak class due to the low number of tokens in this category.

Preceding /s/ was found to be high across monomorphemes, semiweaks and regular

past tense verbs, while preceding stops were higher for the semiweak class. The

argument which supports the lack of morphological effect in British English is due to

the high presence of preceding sibilants, such as /s/, in monomorphemes; hence coronal

34
Preterite went and replacive verbs were found to exhibit categorical /t,d/ retention.

91
stop deletion might not be due to the morphological class itself, but it is influenced by

the preceding phonetic environment (Temple, 2009). A similar finding was also

reported by Sonderegger et al. (2011), who claim that there is no morphological effect

in British English, as when the preceding phonological context is taken out from the

model, the morphological effect disappears. Sonderegger et al. (2011) investigated (t,d)

deletion in the speech of participants in the UK Big Brother reality TV programme. In

the light of this result, Tagliamonte & Temple (2005: 282) claim that “our findings call

into question the universality of the morpho-phonological effect and have led us to

reconsider the possibility that the conditioning of the rule is primarily

phonetic/phonological.” They concluded that (t,d) behaves variably, and that the

theoretical explanations advanced for this feature (OCP, Lexical Phonology and

resyllabification) do not hold for the York data. As regards social factors, age was not

an influential predictor, and males deleted more than females.

3.8.2 Manchester

While the data from York report no evidence for the largely reported morphological

effect found in American English dialects, data from Manchester exhibit the usual

“robust morphological effect” (Baranowski & Turton, 2020). Hence, (t,d) in the British

scenario is strongly marked by this conflicting morphological outcome.

For the Manchester study, Baranowski & Turton (2020) interviewed 93

speakers, aged 8-85, who were raised in Manchester. The sample includes 64 White

British informants, 17 Pakistani, and 12 Black Caribbean Mancunians, stratified by sex

and by socio-economic class: lower working, upper working, lower-middle, middle

middle, and upper-middle. The external factors also include ethnicity, style (casual,

careful, language, minimal pairs, word list) and word frequency from the SUBTLEX-

92
UK corpus (van Heuven et al. 2014). The dependent variable is coded as binary (i.e.

[t,d] absent or present), with glottal stop replacements coded as [t,d] presence. Besides

the usual neutral environments, they also excluded following /j/ (e.g. last year)35,

negative contractions and following /h/. The latter was excluded since the analysis was

based on orthographic transcriptions only, and the Manchester dialect exhibits a mean

of 30% of /h/ dropping (Baranowski & Turton, 2015). A total of 19,550 tokens - 18,274

of which come from spontaneous speech – was analysed through mixed-effects models

in R, with speaker and word as random effect36. The morphological effect turned out

to match the American pattern, with monomorphemes triggering deletion, whilst

regular past tense verbs disfavour it. Result from the preceding phonetic environment

yield the following ranking: sibilants > affricate > nasal > stop > fricative > lateral.

Post-sonorants /n/ and /l/ (e.g. aunt, halt) were subsequently excluded from this

environment as they were found to be obligatorily realised with glottal variants. They

argue that glottaling blocks deletion, especially among monomorphemes where the

highest rate of post-sonorant tokens occurred. The hierarchy for the following phonetic

context exhibits more deletion in consonants, whereas pause and vowel disfavour the

cluster reduction. Word frequency was marked as a significant predictor, with a higher

deletion rate occurs in more frequent tokens, as shown below, while social factors do

not exhibit significant conditioning.

35
The palatal glide was initially included in this study, yet removed later on due to onerous reliability in
distinguishing between deletion and palatalization.
36
Their final converging model was an “intercept only model for speaker” due to the little inter-speaker
variation in their dataset. Standard deviation for word was higher than for speaker.

93
 100

75

%deletion
50

25

1 2 3 4 5 6
frequency

Figure 3.8.1 Word frequency from SUBLEX-UK in Manchester (Baranowski &

Turton, 2020).

3.8.3 Northeast of England

Similar findings have been reported in other northern British varieties. Woolford

(2018), has examined 24 speakers taken from the Newcastle Electronic Corpus of

Tyneside English reporting the expected morphological effect for (t,d), and a nearly

negligible difference between monomorphemes and semiweaks (the deletion rate in

monomorphemes is slightly higher). Social factors (age, sex, and social class) were all

marked as significant predictors, with an increase of /t,d/ reduction in monomorphemes

and semiweaks in apparent time. Woolford confirms earlier suggestions by Baranowski

& Turton (2016), that is: (t,d) is advancing through the life cycle of phonological

processes (Bermùdez-Otero, 2015), according to which a phonological process begins

by applying at lower levels of the grammar, such as at the word level (e.g. effecting

missed and mist) and over time it spreads to the stem level (e.g. mist only). Thus, British

English is behind US English dialects in the stem level (Baranowski & Turton, 2016;

Woolford, 2018).

94
3.8.4 Mersea Island

Amos et al. (forthcoming), on the contrary, take a different approach in exploring (t,d):

they suggest to treat the two coronal stops separately. Johnson & Amos (2016) argue

for a separate analysis of /t/ and /d/ as they appear not to share the same distribution in

word-final consonant clusters as well as in the various morpheme forms. This suggests

that in British English /t/ is more likely to glottalise rather than delete following /n/ and

/l/; therefore, they argue that /t/ glottalization and /t/ deletion are in complementary

distribution. They also argue that in monomorphemes /t/ may only occur after /p, f, s,

k/, whilst /d/ only occurs after /l, n/.37

The sample for the Mersea Island38 (Amos et al., forthcoming) study consists

of 8 speakers, two males and two females aged 19-24, and two males and two females

aged 60-75. The internal factors taken into account are the preceding and following

phonetic segment, morpheme type and inflectional status (free and bound morphemes),

intonational boundaries, and word class distinguishing between true monomorphemes,

regular past tense verbs, semiweaks, irregular strong verbs, regular adjectival forms,

and irregular adjectival forms. Results from a total of 897 tokens39 show that deletion

rates were higher for (d) (24.1%) than for (t) (17.6%). Neither age nor gender were

statistically significant. The significance of internal factors varies depending on the

variable, that is: the following phonetic environment and intonation boundary surfaced

as significant factors in the (t) analysis, whereas the following and preceding phonetic

segments emerged for (d) deletion. The following environment for (t) yields the

following ranking:

obstruent > nasal > glide > liquid > pause > vowel

37
However, this statement does not hold for words like ant.
38
Mersea Island is located in North Essex near Colchester.
39
491 tokens for (t) and 406 tokens for (d).

95
with pause and vowels disfavouring deletion. The intonation boundary, which surfaced

for (t), exhibits a near categorical deletion rate for medial (91%) as opposed to final

(9%) position of the target word.40 As regards (d), the following context yields the

ranking:

obstruent > liquid > glide > vowel > pause

with pause and vowels disfavouring the cluster reduction, while the preceding context

shows that nasals and sibilants favour coronal stop simplification, but stops and liquids

disfavour it. Morphological class and external factors show little significant influence.

The approach of treating separately the two apical stops has been also adopted by

Pavlík (2017), who investigated this variable in standard British English.

3.9 (t,d) in Standard British English

Pavlík (2017) investigated (t) and (d) through BBC radio bulletins, in word-final

consonant clusters as well as word-internally at the morpheme boundary (e.g.,

grandfather, exactly, etc.). The coding of morpheme-final (e.g. landlord) along with

word-final /t/ and /d/ is not a common practise in the study of this variable, hence it

raises the question of encompassing “potentially contrasting linguistic environments”

(Amos et al., fc) under the same category.

The classification of /t,d/ retention includes both released [t,d] and glottal(ised)

tokens, while the commonly excluded neutral environments, such as /t/, /d/, /tʃ/, /dʒ,

/θ/, and /ð/ were also taken into account. This news-reading study does not lead to an

informal speech style, indeed, Pavlík (2017) emphasises that in such a formal context

/t,d/ was not categorically simplified in Ct#t or Cd#d sequences, or in the highly

40
Amos et al. (forthcoming) provide the following examples for medial and final position: 1) That was
the last film I saw (medial); 2) He saw my keys last (final).

96
frequent lexical item and. His findings confirm the strong effect of the following

environment; morphological class is not statistically significant, whereas the effect of

the preceding environment varies (e.g. it influences the /d/ presence, yet this does not

have an effect on /t/). The interaction between preceding and following contexts shows

that the reduction of /t/ and /d/ occurs more frequently when the neighbouring segments

clash with the coronal place of articulation of /t,d/ and with the occlusion stricture.

3.10 Summary

This chapter has reviewed (t,d) deletion in both North American and British English

studies. The pan-English effect shows a notable internal agreement among US English

dialects, while in England contrasting findings have been found with respect to the

morphological effect in those studies where (t,d) was treated as one variable.

Prec. Env. Foll. Env. Morph. Class

American studies
Labov (1967) C, V M, R
Wolfram (1969) stop, son, fric C M, R
Fasold (1972) fric, son C, V, P S, R

Creole variety
sib, stop, other fric, n't, R, M, S,
Patrick (1999) nas, lat C, R, G, P, V ID

UK studies
Tagliamonte & Temple sib, nas, liq, stop, obstr, G, /r/, /l/, non-
(2005) non-sib fric V, P significant
Baranowski & Turton sib, affric, nas, stop,
(2020) fric, lat C, P, V M, S, R
Woolford (2018) M, S, R

Table 3.10.1 (t,d) deletion: Comparison of seven studies. C = consonants, V = vowels,

M = monomorphemes, R = regular past tense verbs, P = pause, R = rhotic, G = glide,
ID = irregular devoicing.

97
In the light of the above findings, this study addresses the following descriptive and

theoretical research questions broadly presented in Chapter 1.

Descriptive aims:

1) How does (t,d) behave in East Anglia?

- The hypothesis is that if major constraints are generally parallel across dialects

of English, location should not be an essential factor in East Anglia (or Britain);

2) Since the following phonological environment strongly affects the deletion of /t,d/

across the English varieties, we may want to explore this linguistic context more

closely to gain a deeper understanding of the universality of this internal constraint.

Therefore, is the classical coding (e.g. obstruents > glides > /r/ > /l/ > vowels >

pause) sufficient to account for this weighty constraint, or do we need to break

down the obstruent category further?41

- The hypothesis is that by grouping the obstruents together some relevant

differences might be obscured.

3) This study also set out to shed light on the unsolved problem of morphological

effect in British English;

- The hypothesis is that if pan-English effects show notable agreement in

internal ordering, the morphological class should be a significant,

predictable constraint;

4) Is phonetic reduction highly conditioned by word frequency?

- The hypothesis is that deletion occurs more in high-frequency words.

Theoretical aims:

5) Is (t,d) governed by a universal sonority hierarchy in British English?

41
A similar question could be addressed for following vowels, yet they are consistently marked as a
disfavouring factor due to resyllabification processes. Obstruents, on the contrary, surface as the most
favouring predictor.

98
 - The hypothesis is that less sonorous preceding segments tend to favour

deletion, whereas more sonorous segments disfavour it (Santa Ana, 1996).

However, given the consistent lack of finding a sonority hierarchy result in

the British context, it would not be surprising to find a similar result in East

Anglia.

6) Is the Obligatory Contour Principle (OCP) an explanatory factor?

- The hypothesis is that the more features of phonetic context are shared with

/t, d/, the more likely deletion is to occur (Guy & Boberg 1997).

7) Does an exponential relationship hold among morphological categories?

8) Exploring (t,d) under a Competing Grammars approach.

- Is variation in deletion of semi-weak verbs partly due to morphological

absence?

99
Chapter 4 - (t,d) DELETION: RESULTS AND
DISCUSSION

Before attempting to answer the questions previously raised, this Chapter briefly recaps

the analytical procedures employed in the present survey and identifies some issues in

the formerly discussed (t,d) literature (e.g. the treatment of the following phonological

environment). I will then present and discuss the overall results (all three localities

together), examined through mixed-effects Rbrul regression analysis. The three

localities are firstly combined, as they are part of what Trudgill (2001a: 10) defines as

“linguistic East Anglia”; secondly, the three urban areas will be treated separately so

that results from each community can be compared and contrasted.

4.1 Analytical procedure - linguistic constraints

In the literature, the surrounding phonetic segments (preceding and following

environments) for the (t,d) variable have been mostly analysed according to the manner

of articulation (e.g. Labov et al.1968; Wolfram 1969; Guy 1991; Patrick 1991;

Tagliamonte & Temple 2005; Baranowski & Turton 2020). Data of the present research

was initially coded according to both manner and place of articulation 1, however the

latter showed a lower proportion of the variance in the response and exhibited a small

effect size, which was measured with the standardized R2 (Winter, 2020)2. Parallel

findings were reported by Neu (1980) whose study shows that the place of articulation

1
Manner and place of articulation were not included in the same run since they are collinear, and thus
highly correlated.
2
See section 2.6 for model comparison.

100
is a non-significant constraint. The final model (see table 4.1), therefore, includes only

the manner of articulation for preceding and following phonetic segment.

In early studies on (t,d), only two categories were distinguished for the following

phonological segment: following vowel vs. non-following vowel (Labov et al, 1968)

or following consonants vs. non-consonants (Wolfram, 1969). The category of non-

following vowel included liquids, pauses and glides. Pause was then excluded from the

consonant factor group and started to be coded as a separate constraint with Fasold

(1972)3. Subsequently, linguists broke down the consonant category into: obstruents

(stops, fricatives and nasals), glides, /r/, /l/, vowels and pause (Tagliamonte & Temple

20054). This study continues to break down the consonant category further and treats

separately:

a. stops

b. sibilant fricatives

c. non-sibilant fricatives and

d. nasals (sonorants).

The reason for examining more closely this factor group is twofold:

(1) the following phonological environment has proven to be the most weighty

constraint, hence it is worth speculating on the effect that each and every factor may

have on the deletion rate. Since the manner of articulation exhibits a higher proportion

of variation for this variable, the degrees of closure of a following phone could be taken

into account5. From an articulatory viewpoint, the degree of stricture differs for

3
However, since the effect of a following consonant (38.6%) and that of a following pause (39.1%) was
nearly identical, the following phonetic segment was consequently coded as: following consonants vs.
following vowels.
4
In Tagliamonte & Temple (2005: 288) the obstruent category for the following phonetic segment
includes nasals.
5
To describe the manner of articulation, phoneticians sometimes divide the sounds based on the degree
of closure, distinguishing between: closure, close approximation and open approximation (e.g. vocoids,
in which the airflow is smooth). Further distinctions encompass whether the air flows through the nose
(nasal) or otherwise (oral).

101
plosives, fricatives and nasals. Plosives and nasals could be grouped together since they

both exhibit a complete closure. However, while plosives have a complete closure in

the vocal tract followed by an explosive release of the airflow (e.g. [p,b,k,g]) 6, for

nasals the degree of closure decreases and the airflow escapes through the nose.

Fricatives, instead, are realised through a close approximation of two articulators and

the airstream resulting in a friction, which is partially obstructed.

(2) If sounds are classified based on their noise components, only plosives and

fricatives can be included in the obstruent category7, where “the constriction impeding

the airflow through the vocal tract is sufficient to cause noise” (Cruttenden, 2014: 31).

Nasals, instead, show no noise component and are more vowel-like (Cruttendan, 2014).

However, they have also been treated as obstruents as the flow of air through the mouth

is blocked.

With respect to fricatives, I have also distinguished between sibilant fricatives

and non-sibilant fricatives in the following phonetic context. Usually, it is

recommended to combine factor groups which are correlated with the response or if

they represent the same manner of articulation to reduce the deviance (Tagliamonte,

2006). However, I believe that collapsing predictors without testing how individual

factors behave could obscure some relevant influences on the variation of the target

dependent variable. The decision of treating sibilants and non-sibilant fricatives

separately in the following phonetic context is also conceptually validated as /s/ and

other fricatives were consistently found to behave differently in environments

preceding /t,d/ (Bayley 1994; Patrick, 1999; Tagliamonte & Temple 2005). Therefore,

6
Following [t] and [d] are excluded from the analysis, therefore not mentioned here.
7
Affricates are also classified as obstruents, but they are excluded from the following segment here,
for the purposes of this study.

102
they might have a statistically distinct influence on (t,d) deletion even in the following

environment, such as in around some, at last he moved, etc.

In the (t,d) literature, following vowels (e.g. dent in) are usually grouped

together. Despite being marked as a disfavouring factor due to the resyllabification of

the apical stops /t/ and /d/ onto the following segment, I determined to break down the

vowel category into high-vowels, mid-vowels and low-vowels to examine whether

deletion is consistently disfavoured in all three cases. Initial Rbrul runs, indeed, showed

that the realisation of [t,d] is inhibited when followed by any vowel type. To obtain the

neatest and most accurate model, these three factor groups were then collapsed.

Occurrences followed by /t/, /d/, interdental fricative /θ/, /ð/, post-alveolar fricatives

/tʃ/, /dʒ/, the lexical item and, tokens where /t,d/ is preceded by the approximant /r/

(i.e. historically rhotic), as well as word-final /t/ tokens realised with a glottal stop were

excluded from the analysis8. Final clusters following /l/ (e.g. halt) were only coded

when /l/ was consonantal; whereas, cases in which /l/ was vocalised were also excluded

from the analysis as the apical stops /t/ and /d/ would no longer be in a final consonant

cluster. Glottal variants are excluded from the main (t,d) regression analysis as (t)

glottaling carries social meaning in East Anglia and in other UK speech communities,

thus I argue that (t) glottaling is a different variable even though clusters containing

glottal variants (e.g. silent) have been included in previous (t,d) studies (see

Tagliamonete & Temple 2005; Baranowski & Temple 2020). The decision of

excluding glottal variants from the (t,d) analysis also aligns with Amos et. al’s (2020)

study, according to which glottalization, in British English, is a case of lenition rather

than deletion, therefore it is a separate phenomenon.

8
In most cases, the dependent variable is categorically coded as [t,d] presence vs. deletion of /t,d/.
Unlike other English varieties, in British dialects /t/ can be frequently glottal(ised), hence UK (t,d)
studies have treated the glottal variant as a case of [t,d] presence.

103
However, as a point of comparison with Manchester (Baranowski & Turton 2020) and

York (Tagliamonte & Temple 2005), I will firstly present a model with spontaneous

speech only, where glottal variants are coded together with [t,d] presence (see section

4.3).

4.2 Analytical procedures – Rbrul

Across the 36 interviews conducted in the three cities, a total of 4,879 tokens was coded

into an Excel spreadsheet. To have sufficient data for statistical analysis, at least 30

tokens per cell are recommended (Guy, 1980)9. The large number of tokens in the East

Anglian dataset, however, is well above this statistical threshold averaging 135 tokens

per speaker, and it is evenly distributed between age, class sex, and location, as shown

in Appendix I.

As discussed in Chapter 2, the statistical tool employed to carry out this

variationist analysis is Rbrul. The initial model was run with all factors to identify the

most important predictors as well as any potential interaction. In the second step of the

analysis, I run step-up and step-down separately to test for their respective relevance

on the variation of (t,d). Finally, a step-up/step-down analysis was conducted.

The type of response in the model is binary, with deletion as application value and

individual speaker and word included as random effect. Along the line of Baranowski

& Turton (2020), the most complex model tested included by-speaker as random slope

for following phonetic segment and morphological class10. Since these constraints were

found to vary across varieties, they might also vary across speakers. Crosstabulations

9
See also Milroy (1987).
10
Besides random intercepts, mixed-effects models can also include random slopes. This means that the
model allows speakers “not only to differ in the rate at which they use a variant, but also to differ in the
size of the effect between-word constraints, such as phonological context” (Gorman & Johnson, 2013:
224).

104
between the response and each linguistic and social factor were carried out to observe

for potential knock-outs before building the final model, illustrated in section 4.4.

Constraints Factors
Preceding environment nasals (e.g. remind)
sibilants (e.g. cost)
fricatives (e.g. raft)
laterals (e. g old)
stops (e.g. strict)
Following environment nasal (e.g. soft moan)
sibilant (e.g. best source)
stops (e.g. second could)
laterals (e.g. don’t like)
glides (e.g. second world)
rhotics (e.g. planned reunion)
vowels (e.g. went on)
pause (e.g. […] was lost.)
Morphological class monomorphemes (e.g. mist)
semi-weak verbs (e.g. left)
regular verbs (e.g. called)
Voicing agreement homovoiced (e.g. bold)
heterovoiced (e.g. bolt)
Syllable stress unstressed (e.g. happened)
stressed (e.g. left)
Style Spontaneous speech
Word frequency low frequency (1-3); high frequency (4-7)
Social class working class, middle class
Age young (18-28), middle (35-50), older (60+)
Sex female, male

Table 4.2.1 Constraints of the (t,d) variable in the East Anglian dataset, with glottals
included.

105
The difference between the model illustrated in section 4.4 and the model above is that

the latter includes glottal variants, it excludes n’t tokens and it is based on spontaneous

speech only, whereas the model in section 4.4 examines (t,d) deletion across all styles

(spontaneous speech, reading passages and word lists), it includes the morpheme {n’t}

in the morphological class, and excludes glottal variants from the dataset.

4.3 Overall Results – model with glottal variants included in the (t,d)
dataset

This section briefly presents the results from the mixed-effects regression analysis with

glottal variants included in the (t,d) model, along the line of Tagliamonte & Temple

(2005) for York, and of Baranowski & Turton (2020) for Manchester. The number of

glottal variants in the East Anglian dataset equals 384. In York, glottals represented

10% of the data (N = 125), whereas in Manchester 82% of /nt/ and /lt/ clusters were

glottalled11. In line with previous UK (t,d) studies, this section excludes n’t tokens from

the dataset.

The statistical information contained in the table to follow includes: R2, which

is a measure of the ‘goodness of fit’ (Winter, 2020); log-odds, which reflect the strength

of the relationship between a predictor and the response – if log-odds are above 0, there

is a positive correlation between the variables, whereas if they are negative there is a

negative correlation between them; factor weights are relative probabilities within the

range of 0 – 1.00 and are related to log-odds; AIC “is a goodness-of-fit measure for

comparison of models with different numbers of parameters” (Levshina, 2015: 194).

11
/t/ in post-sonorant position (e.g. halt, aunt) where excluded in Manchester as they categorically
realised with glottal variants.

106
Moreover, in the logistic regression with a binary response, p-values come from a

likelihood-ratio chi-squared test.

Application value = deletion; overall proportion = 0.324

R2 = 0.362; log likelihood = -1387.24; N = 2750; AIC = 2808.481

Constraints Logodds FW % Tokens

Voicing
Agreement
p <.001
heterovoiced 0.532 0.63 54 547
homovoiced -0.532 0.37 27 2203
Preceding env.
p<.001
nasals 0.683 0.664 44 1148
sibilant fricatives 0.448 0.61 33 761
/l/ -0.03 0.493 28 204
stops -0.368 0.409 13 338
non-sibilant
fricatives -0.734 0.324 11 299

Following env.
p<.001
nasals 1.221 0.772 63 145
sibilant fricatives 1.215 0.771 67 124
stops 0.933 0.718 61 222
/l/ 0.65 0.657 55 40
glides -0.043 0.489 40 272
non-sibilant
fricatives -0.167 0.458 34 228
vowels -0.9 0.289 22 979
pause -1.091 0.251 22 715
/r/ -1.818 0.14 16 25

Morphological
class p<.001
semiweak verbs 0.452 0.611 44 329
monomorphemes 0.221 0.555 38 1762
regular verbs -0.673 0.338 13 659

Table 4.3.1 Multivariate analysis of (t,d) in East Anglia, including glottal variants.

107
Results from the multivariate analysis revealed that, in East Anglia, when only

spontaneous speech is taken into account, voicing agreement, preceding phonological

environment, following phonological environment, and morphological class are

marked as significant predictors. Thus, similar to Manchester, morphological class

reached statistical significance. Table 4.3.1, which displays results obtained from the

best step-up/step-down model, shows that heterovoiced tokens favour more deletion

than homovoiced ones – a finding which goes against previous (t,d) studies (see section

4.4.2). In the preceding environment, the only favouring predictors are nasals and

sibilant fricatives, with nasals being at the top the ranking; whereas, preceding /l/, stops

and non-sibilant fricatives disfavour deletion. While the behaviour of most predictors

is in line with Manchester and York, the high position of nasals in East Anglia contrasts

with the two northerner cities. Results from the following environment reveal that

sibilant fricatives and non-sibilant fricatives behave very differently (see section 4.4.1).

Nasals, sibilant fricatives, and /l/ favour the absence of /t,d/, whilst glides, non-sibilant

fricatives, vowels, pause and /r/ are marked as disfavouring predictors. The different

behaviour of /r/ and /l/, in East Anglia, is consistent with previous US (t,d) studies (see

section 4.4.1) whereas, in York, /l/ is in a neutral position. Vowels and pause also

disfavour /t,d/ in both Manchester and York, yet following glides favour the

simplification of /t,d/ in York. With respect to the morphological class, semi-weak

verbs appear at the top of the ranking favouring deletion at 0.611, monomorphemes

favour at 0.555, while regular verbs, as expected, disfavour it. In a further run (not

shown here) glottal variants were excluded from the above model (with spontaneous

speech only). Results revealed that the difference between the two models is not

statistically significant (see section 2 for model comparison). In other words, the model

with glottal variants is not significantly different from the model without glottal

108
variants. AIC for the model with no glottals equals 2431.785, which is lower that the

AIC displayed in table 4.3.1. This means that “the smaller the AIC, the better the fit”

(Levshina, 2015: 149). Moreover, since I contend that (t) glottaling is a different

sociolinguistic variable in British English, in the next section, glottal variants will be

excluded from the (t,d) dataset, all styles (spontaneous speech, reading passages, word

lists) will be examined and negative contractions n’t will be taken into account for a

thorough account of the (t,d) variable.

4.4 Overall Results – model with no glottal variants included in the (t,d)
dataset

Figure 4.4.1 displays the overall findings for (t,d) deletion across the East Anglian

sample where the stability of this variable surfaces. The simplification of word-final

/t,d/ appears to be equally distributed across the three age cohorts, sex and social class.

Middle class females exhibit a relatively lower deletion rate in the middle-aged (22%)

and in the old (19%) groups. On the other hand, young middle class females, along

with young middle class males, show a slightly higher (t,d) absence rate compared to

working class males and females. Further negligible differences between social factors

will be commented on later in the Chapter. Let us now turn the attention to the mixed-

effects regression analysis.

40
35
30
25 WC females
% deletion

20 WC males
15 MC females
10 MC males
5
0
young middle old

Figure 4.4.1 Overall results in the East Anglian dataset.

109
Table 4.4.1 summarises the constraints included in the final model.

Constraints Factors
Preceding environment nasals (e.g. remind)
sibilants (e.g. cost)
fricatives (e.g. raft)
laterals (e. g old)
stops (e.g. strict)
Following environment nasal (e.g. soft moan)
sibilant (e.g. best source)
stops (e.g. second could)
laterals (e.g. don’t like)
glides (e.g. second world)
rhotics (e.g. planned reunion)
vowels (e.g. went on)
pause (e.g. […] was lost.)
Morphological class monomorphemes (e.g. mist)
semi-weak verbs (e.g. left)
regular verbs (e.g. called)
negative contractions (e.g. can’t)
Voicing agreement homovoiced (e.g. bold)
heterovoiced (e.g. bolt)
Syllable stress unstressed (e.g. happened)
stressed (e.g. left)
Style spontaneous, reading styles, word lists
Word frequency low frequency (1-3); high frequency (4-7)
Social class working class, middle class
Age young (18-28), middle (35-50), older (60+)
Sex female, male

Table 4.4.1 Constraints of the (t,d) variable in the East Anglian dataset, without
glottals.

110
Results of the best model achieved in the step-up/step-down analysis are presented in

table 4.4.2. In the step-up analysis, the programme adds predictors one at a time,

beginning with those which have the greatest effect on the dependent variable. This

process is repeated until no more significant predictors can be added. In the step-down

analysis Rbrul fits the full model and removes those independent variables which are

not significant. The findings pattern with previous North American studies, as the

logistic regression shows that following environment, voicing agreement,

morphological class, style and preceding environment are statistically significant,

whilst none of the social factors (social class, age and sex) has a significant influence

on the dependent variable. Parallel to previous studies on (t,d), this research confirms

that the explanatory factors for this variable are linguistic rather than social (see section

4.4.6). The predictors which reached statistical significance in the mixed-effects

regression analysis are examined more closely in the following sections.

111
Application value = deletion; overall proportion = 0.257
R2 = 0.672; log likelihood = -1588.162; N = 4,879

Constraints Logodds FW % Tokens

Following Env.
p< .001
nasal 1.489 0.816 64 197
sibilant fricative 1.444 0.809 77 192
stops 0.798 0.69 55 424
/l/ 0.140 0.535 56 44
glides -0.305 0.424 39 298
/r/ -0.331 0.418 29 81
non-sibilant fricative -0.331 0.418 32 343
vowels -1.231 0.226 18 1405
pause -1.672 0.158 10 1895
Voicing Agreement
p< .001
heterovoiced 1.22 0.772 64 891
homovoiced -1.22 0.228 17 3988
Style
p< .001
spontaneous speech 1.461 0.812 37 2710
reading style 0.236 0.559 19 1131
word list -1.696 0.155 3 1038
Morphological class
p<.001
n't negative contractions 0.774 0.684 83 196
monomorphemes 0.182 0.545 37 2204
semi-weak verbs -0.066 0.483 21 687
regular verbs -0.889 0.291 7 1792
Preceding Env.
p<.001
nasal 0.703 0.669 40 2003
sibilant fricatives 0.473 0.616 23 1226
/l/ -0.211 0.447 17 357
stops -0.358 0.412 7 728
non-sibilant fricatives -0.607 0.353 7 565

Table 4.4.2 Multivariate analysis of (t,d) in the whole dataset.

112
4.4.1 Following environment

The extensive literature on (t,d) deletion has shown that this linguistic variable is

strongly constrained by the following phonetic segment. Early empirical (t,d) findings

exhibit the following hierarchy in many variationist studies of American Englishes:

Consonant > Liquid, Glide > Vowel

with consonants being the most favoured and vowels the least favoured predictor.

Results from the East Anglian analysis confirm that following phonetic segment has a

robust effect on (t,d) with nasals, sibilants, stops and /l/ favouring deletion, whereas

glides, /r/, non-sibilant fricatives, vowels and pause disfavour it. Nasals turned out to

trigger deletion the most at .816, along with sibilants (.809). The latter differs greatly

from non-sibilant fricatives which, by contrast, disfavour deletion at .418. This result

suggests that previous research that merged these two factors (sibilants and non-sibilant

fricatives) in the obstruent category, may have obscured this difference12. A likelihood

ratio test revealed that treating sibilants and non-sibilant fricatives separately is

statistically significant (x2 (1) = 52.24, p < .001). It is worth noting that non-sibilant

fricatives outnumber sibilants in this dataset, with 343 tokens and 192 respectively. The

low position of non-sibilants in the hierarchy, at this point, requires some more detailed

comment. Firstly, it should be noted that spirants included in the East Anglian dataset

are following [f], [v] and [h], but let us briefly go back to the classification of

obstruents. These sounds are articulated with the obstruction to the airstream in the

mouth and, consequently, the glottal [h] is not included in this category (Roca &

Johnson, 1999). Thus, it could be assumed that: (a) previous studies grouped following

12
Similarly, since laterals and rhotics have different status, Guy (1991) argued for subdivision of
the liquid category and found that the effect of /l/ is dissimilar to that of /r/.

113
/h/ with obstruents; or (b) following /h/ has been excluded from the analysis, as in

Manchester (Baranowski & Turton, 2020). Moreover, following /h/ has been recently

referred to as a problematic case to establish whether the deletion rule applies post-

lexically (Temple, 2017). In the following context, indeed, /h/ is underlyingly

consonantal, but may be phonetically a vowel especially in accents which exhibit (h)-

dropping. East Anglia, however, is one of the parts of England where /h/ is retained

especially in rural dialects, at least amongst older speakers (Trudgill, 1974):

Figure 4.4.2 /h/ in hammer. Adapted from Trudgill (1974).

In the present analysis, following /h/ was coded on the phonetic surface. Thus, when

underlyingly /h/ was phonetically consonantal it was coded as a spirant 13, when it was

phonetically a vowel, it was coded in the vowel category. Cross-tabulations between

non-sibilant fricatives and following phonetic segment revealed that before following

labio-dental [f]14 /t,d/ was retained at a rate of 45%. Before following /v/ the variable

13
The term spirant, here, refers to non-sibilant fricatives only.
14
80 out of the 147pre-[f] tokens were deleted.

114
sporadically occurred and the deletion rule did not apply. Hence, among non-sibilant

fricatives, following /h/ plays a notable role in disfavouring deletion, as 84% retention

occurred before following [h] (e.g. stand here)15, and even more retention (95%) was

found when the following /h/ was dropped (see section 4.4 for the behaviour of non-

sibilant fricatives and following /h/ in the three locations separately analysed). This

result might be due to the open glottal constriction of /h/, compared to the close oral

constriction of other following consonants which inhibit /t,d/, as also claimed by

Temple (2009), hence /h/ patterns like vowels.

100

80
% deletion

60

40

20

0
Following [h]

Figure 4.4.3 Rates of deletion before following [h].

100

80
% deletion

60

40

20

0
dropped /h/

Figure 4.4.4 Rates of deletion before dropped /h/.

15
Only 28 of the 179 tokens of pre-consonantal [h] were deleted.

115
If this finding is consistent across the three East Anglian locations, it will reveal new

phonological insights into the following environment for (t,d).

The third favouring predictor, in order or ranking, is represented by stops (.69). Even

though these factors (nasals, stops and sibilants) are treated separately in this analysis,

these findings seem to match the literature as the obstruent category has constantly

highly influenced deletion and so do nasals. So far, what stands out in this category is

that non-sibilant fricatives strongly disfavour deletion.

The approximants /l/ and /r/, in line with previous research, behave differently in the

East Anglian dataset. The lateral shows a value of deletion equal to .535, whilst the

rhotic consonant disfavours at a rate of .418. This distinction, as suggested by Guy

(1991), can be explained by the resyllabification process according to which final stops

may resyllabify onto the following segment as in went round, where final /t/ is more

likely to retain since English allows for the following syllable onsets /tr-/ and /dr-/. On

the contrary, */tl-/ and */dl-/ are prohibited in all English varieties16. However, results

from the York data (Tagliamonte & Temple, 2005) is not consistent with the

resyllabification explanation as all consonants were found to favour deletion except for

/l/ (.50), being in a neutral position. This finding might be linked to the

clearness/darkness of /l/ (further details will be provided later in the Chapter).

Glides, which divide /l/ and /r/ in the hierarchy, along with vowels and pause are

marked as disfavouring factors. As shown in the previous Chapter, the effect of

following pause is “open to dialect-differentiation” (Labov, 1989: 90) and it is usually

referred to as an arbitrary factor (Patrick, 1991). In New York City (Guy, 1980),

16
Due to the similarity of /l/ with other consonants, in making unacceptable syllable onsets following /t/
or /d/, laterals and the other consonants have been grouped in a single factor in some studies on (t,d) (e.g.
Jamaican Creole), whilst rhotics were treated separately (Patrick, 1999). In Manchester, both laterals and
rhotics have been merged in the consonant category (Baranowski & Turton, 2020).

116
Jamaican Creole (Patrick, 1991), Tejano English (Bayley, 1994) and Manchester

(Baranowski & Turton, 2020) following pause boosted /t,d/ deletion more than

following vowels17. In other southern and southwestern US dialects (Santa Ana, 1996),

in Philadelphia (Guy 1980; Tamminga 2016), York (Tagliamonte & Temple, 2005) and

in East Anglia, instead, pause behaved like a vowel in promoting retention.

In terms of sonority, Guy (1991) claims that the less sonorous a following segment is,

the higher the reduction of the coronal stop. However, in East Anglia, the sonority

effect does not surface for the following environment despite glides and vowels

showing lower rates of deletion.

4.4.2 Voicing agreement

The second most powerful constraint in the regression analysis is agreement in voicing

of the segments preceding /t,d/. According to Wolfram (1969), English is governed by

a rule by which the second member of a consonant cluster can be deleted provided that

the second member is a plosive and that both members agree in voicing. Indeed, typical

findings show more deletion in homovoiced tokens than in heterovoiced ones (Labov

1989; Bayley 1994). However, in the East Anglian dataset the reverse occurs:

heterovoiced tokens (e.g. bolt) promote deletion, yet homovoiced tokens (e.g. bold)

disfavour it. Heterovoiced tokens make up only 18% of the data but are deleted at a

much higher rate (64%), as opposed to 17% deletion for homovoiced tokens. This

scenario does not change even when glottal variants are included in the dataset, indeed

heterovoiced tokens remain the most favouring predictor (see section 4.3). The

question is: why?

17
Even though following pause boosted deletion more than vowels, they were both statistically
disfavouring predictors in Manchester.

117
To begin further investigation let us zoom into this factor group. Recall that Johnson

& Amos (2016) suggested to split the (t,d) variable in British English, since /t/ and /d/

do not share the same distribution in word-final clusters. Along this line, the East

Anglian dataset was split and /t/ and /d/ were treated separately for a careful

consideration of this unconventional finding. However, splitting the dataset is not going

to help as the only data for /d/ are two irregular devoicing tokens.

Turning to the analysis of /t/ only, heterovoiced was marked as disfavouring predictor

even though the deletion rate is much higher18. Possible linguistic explanations could

be due to (a) interaction with preceding phonetic segment or (b) interaction with

morphological class if nearly all heterovoiced tokens are monomorphemes. The /t/

dataset was reanalysed again excluding the preceding context19. Results finally show

that heterovoiced tokens strongly favour deletion. Hence, this unconventional finding

springs from interaction with the preceding environment. Crosstabulations revealed

that 70% of heterovoiced tokens are preceded by nasals, whilst 29.5% are preceded by

laterals.

80
% heterovoiced tokens

70
60
50
40 nasals
30 /l/
20
10
0
preceding segment

Figure 4.4.5 Heterovoiced tokens in the preceding segment.

18
The deletion rate of disfavouring heterovoiced is 64%, whereas it is 19% for favouring homovoiced.
Note that the Variance Inflection factor (VIF) was high in this run. VIF quantifies multicollinearity in
the regression analysis.
19
Additional analysis was then conducted including the preceding environment and excluding
morphological class, yet this run turned out not to be significant.

118
Even though the lateral’s rate of deletion is also relevant, the effect of /l/ goes in the

right direction disfavouring the word-final simplification. Thus, the prime issue for

this constraint remains preceding nasals. Further investigation shows that in the /t/-

analysis all preceding nasals (n = 794) are also included in the hetero-tokens category

(n = 887), therefore there is massive overlap and the condition of orthogonality is not

fulfilled. It is claimed that to achieve an accurate analysis, factor groups must be

‘orthogonal’ i.e. there must be minimal overlap between them (Clark & Trousdale,

2013).

4.4.3 Style-shifting

The third significant predictor in the mixed-effects regression analysis is style-shifting.

Results show consistency in the literature with more (t,d) absence in spontaneous

speech than in reading style (Labov, 1967). To account for stylistic variation in the

present study, (t,d) deletion is investigated in spontaneous speech, reading passages

and word lists. Figure 4.4.6 illustrates a linear effect for the East Anglian pattern.

40
35
30
25
% deletion

20
15
10
5
0
Informal formal word lists

Figure 4.4.6 Rates of deletion across different speech styles.

119
Typically, the literature suggests that individuals are able to differentiate

monomorphemes from regular past tense verbs only in careful speech (Roberts, 1994).

Figure 4.4.7 illustrating crosstabulations between style and morphological class,

reveals that East Anglian speakers largely delete /t,d/ in negative contractions (e.g.

don’t) in both spontaneous speech and reading styles. Comparably, in both styles, lower

deletion rates occur among monomorphemes; however, /t,d/ is categorically retained

when monomorphemes and n’t tokens are realised in isolation (i.e. word lists). More

deletion occurred among semiweaks in the spontaneous speech, whilst reading styles

and word lists show a lower incidence of (t,d) absence. Noticeably, low rates of /t,d/

absence were found among regular verbs, where 5% of deletion surfaced in word lists.

100

80
% deletion

60

40

20

0
n't M S R

informal formal word list

Figure 4.4.7 (t,d) deletion by style and morphological class.

These findings align with the classical approach to sociolinguistic style according to

which individuals eschew non-standard forms (e.g. (t,d) deletion) as their level of

awareness increases (e.g. in reading styles and word lists). Similar findings are reported

in Manchester, with more deletion in spontaneous speech than in careful style

(Baranowski & Turton, 2020). Purse’s (2019) Electro-Magnetic Articulography (EMA)

study, however, shows that speakers produced the least tongue tip raising in word list tasks.

120
4.4.4 Morphological Class

The fourth significant predictor is the morphological class (Ciancia & Patrick, 2019).

In order to shed light on the unsolved problem of morphological effect in British

English dialects a crucial comparison with previous British studies is needed. In

Chapter 3 we came across the conflicting results between York (Tagliamonte &

Temple, 2005), Manchester (Baranowski & Turton, 2020), and Tyneside English

(Woolford, 2018). In York, morphological class failed to reach statistical significance;

conversely, Manchester and Newcastle exhibit the usual ‘robust morphological effect’

with more deletion in monomorphemes (e.g. mist) than inflected forms (e.g. missed).

Group results. The following graph, matching the North American pattern, shows that

East Anglian speakers are more likely to delete /-t,d/ from negative contractions (e.g.

can’t) and monomorphemes (e.g. mist) than semi-weak (e.g. left) and regular past tense

verbs (e.g. called), yielding the following ranking:

n’t > monomorphemes > semi-weak verbs > regular past tense verbs.

100
N = 196
80
% deletion

60
N = 2204
40
N = 687
20 N = 1792

0
n't contractions monomorphemes semi-weak verbs regular verbs

Figure 4.4.8 Deletion rates by morphological class.

121
It is worth-noting that, despite the low number of tokens, negative contractions are the

most favouring predictor, similar to Jamaican Creole where /t,d/ was found to be

deleted at a rate of 87% (Patrick, 1999). Patrick (1999) emphasises that n’t clusters

generally show the highest rates of (TD)-absence of any morphological class. King of

Prussia’s informants were also found to exhibit a greater deletion rate variance across

speakers in negative {n’t} (Labov, 1989). However, few studies took into account n’t

morphemes. These tokens were also included in the Tejano English study (Bayley,

1994), yet they are only mentioned in the preceding phonological segment to evince

that a sonority hierarchy which governs (t,d) is less compelling. Labov (1986) excluded

both [nt] and [nd] tokens as considered “hard to tabulate” owing to the nasal flap

formation rule in which the stop feature is realised by a flap holding characteristics of

nasality. Negative {n’t} tokens have also been removed from both the York

(Tagliamonte & Temple, 2005) and Manchester (Baranowski & Turton, 2020) studies.

Along the line of Patrick (1999), I believe that it is worth considering this

morphological category in order to account for a thorough morphological effect and to

control for the sonority hierarchy of the preceding phonetic segment. When comparing

the East Anglian morphological effect with that of York and Manchester, n’t tokens

were excluded from the analysis, yet the monomorpheme (M) factor group still remains

a favouring predictor at .61; semi-weak verbs (S) favour at .54, whereas regular verbs

(R) resist deletion at .35.

Morphological class
Factor logodds Tokens % FW
M 0.464 2204 37 0.614
S 0.173 687 21 0.543
R -0.636 1792 7 0.346

Table 4.4.3 Regression analysis of morphological class.

122
The coding procedure for monomorphemes in previous studies includes strong

preterites (e.g. went) and replacive verbs (e.g. sent) “under the assumption that such

verbs fulfil the structural description of -t,-d deletion from the earliest lexical insertion”

(Guy, 1991: 20).

The lexical phonology model which has strengthened the theoretical framework

of the (t,d) variable rule claims that all morphological processes and many

morphological ones are carried out in the lexicon. This theory suggests that different

types of morphological processes operate at two ordered levels within the lexicon:

irregular inflections are found at level one, whilst regular inflectional affixes occur at

level two. “A form must pass through all levels before surfacing” (Guy, 1991: 6). Thus:

a. Monomorphemes

The underlying /t,d/ in monomorphemes exists from the earliest stage of

lexical insertion; e.g. raft, cold.
b. Semi-weak verbs
The semi-weak class is treated as undergoing affixation at level one
owing to the stem vowel change; e.g. kept, told.
c. Regular past tense verbs
In regular past verbs, /t,d/ are inserted at level two of the lexicon (Guy,
1991); e.g. missed, called.

In York, suppletive forms (e.g. found), replacive forms (e.g. sent) and preterite went20

were coded separately (Tagliamonte & Temple, 2005), whilst they were all coded

together with semi-weak verbs in the East Anglian study. However, the East Anglian

dataset exhibits a low number of tokens for the above categories: found (n = 12), sent

(n = 12) and went (n =25). It is unlikely that the exclusion or reclassification of these

20
In note 12, Tagliamonte & Temple (2005) note that a high number of went tokens were realised with
glottal stop or were glottalized (45%), hence they suggest that it should be treated separately. In the
East Anglian dataset, however, tokens realised with final /t/-glottaling/glottalization were excluded
from analysis.

123
inflectional forms would have an effect on the statistical significance for this constraint,

given the high number of tokens in the four morphological categories. Indeed, the

strong morphological effect does not disappear when preterite went is excluded from

the analysis.

Comparison with York and Manchester. Deletion in the monomorpheme

category is higher in all three studies, whilst semi-weak verbs nearly pattern with

regular verbs in both York and Manchester21. In East Anglia, instead, there is a strong

linear effect when all styles are taken into account.22

60

40 York
% deletion

Manchester
East Anglia - all styles
20

0
M S R

Figure 4.4.9 Rates of deletion by morphological class in East Anglia, York and
Manchester.M = monomorphemes, S = Semi-weak verbs, R = Regular past verbs.

Baranowski & Turton (2016) discussed earlier findings from Manchester under the

theory of the lifecycle of phonological processes (Bermúdez-Otero 2007, 2015). This

theory, which is associated with phonological change and models the development of

phonological features over time, suggests that a phonological process initially applies

21
Note that morphological class, in York, was not marked as a significant predictor.
22
When only spontaneous speech is taken into account (including glottal variants) rates of deletion for
monomorphemes (44%) and semi-weak verbs (37%) are slightly higher than Manchester, whereas
deletion rates for regular verbs (13%) are lower than the two northern studies.

124
at lower levels of the grammar, i.e. at the word level,23 as its access to the morphological

structure is blocked. This suggests that, over time, this process may apply to the stem

level, affecting only monomorphemes, since the stem level does not have access to the

regular past tense affixes and, consequently, they do not undergo (t,d) deletion. They

also suggest that the strong morphological effect which is commonly found across US

English varieties may be due to the maximum rates at stem level reached by US

individuals. Thus, the British pattern would suggest that Britain is behind the US in

terms of stem-level as also suggested by Woolford (2018) for Tyneside English.

Comparatively, levels of deletion in East Anglia are lower than Manchester and even

lower than York, except for monomorphemes. Viewed through the lens of the above

theory, it could be assumed that (t,d) deletion is at an early point in East Anglia,

suggesting that it has already been applying at a stem level and that, probably, it will

not go to much higher overall levels if it only applies at its maximum rate late in change.

One of the core claims of life cycle theory is that some phonological processes merge

with morpho-syntax as they age24. This approach appears to apply to changes in

progress across a speech community, as opposed to e.g. Guy & Boyd’s findings which

manifest an age-grading process across the individual lifespan. However, results from

East Anglia take us to a direction which is opposite to that of a change in progress, as

age was not marked as a significant predictor, indeed all three age groups delete /t,d/ at

the same rate.

The semi-weak class was traditionally considered an intermediate category,

even though some speakers treat them as monomorphemes, others as regular past tense

verbs (Guy & Boyd, 1990). Despite being a disfavouring predictor in East Anglia, the

23
In this case mist and missed would be affected at the same rate.
24
This links to Bybee’s (2002) assumption of a change in progress as discussed in section 3.3.5. It is
worth noting, though, that social factors do not show significant conditioning of (t,d) in Manchester
(Baranowski & Turton, 2020), but they do in Tyneside English (Woolford, 2018).

125
deletion rate of semi-weak verbs is closer to that of monomorphemes than regular

verbs. Indeed, when negative {n’t} tokens were excluded from analysis, results showed

a positive correlation between semi-weak verbs and monomorphemes. However, given

the gap between the two predictors, the semi-weak class could be considered an

intermediate category (see table 4.4.3 ). Fruehwald (2012: 85) argues that semi-weak

verbs and regular past tenses should be treated “as being identical in terms of

phonological deletion and differing in morphological realisation.” Under a Competing

Grammars approach to variation, as outlined in Chapter 3, Fruehwald (ibid) suggests

that there is morphological variation in the semi-weak class. Along this line, he claims

that there are two processes to reach the surface form [kɛp]: (i) morphological absence

of /t/; (ii) phonological deletion of /t/ which was morphologically present. Fruehwald

(ibid), in the Buckeye Corpus, found young speakers to delete near categorically in the

semi-weak category; whilst Patrick (1999) found a distinctly higher rate of (t,d) absence

in regular past verbs in JC. This result from Jamaica Creole is due to an additional

intervening process, known as morphological variation. In East Anglia, however, the

measurement of morphological presence in semiweaks according to the formula below

(as in Fruehwald, 2012):

Observed Semiweak (t,d) presence

Regular past (t,d) presence

reveals the following result: 0.786/0.929 = 0.846% of morphological presence. This

suggests a low rate of morphological absence (i.e. non-marking of past tense). To the

extent it occurs, it may be due to adult speakers who have never realised semi-weak

verbs as a distinct category and/or do not categorically inflect them.

The exponential model – an explanatory factor for the morphological effect for

some English dialects (see Guy, 1991) – does not hold for the East Anglian data. While

Guy’s (1991) values obtained from monomorphemes and regular past tense verbs are

126
very close25 and confirm his hypothesis, in East Anglia the estimated value of pr

(probability of retention) for monomorphemes is comparatively lower than that of

regular verbs, as illustrated in the table to follow:

Morphological class Rate of retention Estimated value of pr

Monomorphemic 0.629 0.856 (cube root of surface rate)

Semiweak past 0.786 0.886 (square root of surface rate)
Regular past 0.929 0.929

Table 4.4.4 Estimates of the value of pr in the East Anglian dataset.

Why do monomorphemes show a smaller value of pr? Could word-frequency

potentially affect this outcome? Usually, monomorphemes are subject to a notable

deletion rate, hence it would be expected that rates of retention are lower in both high

and low frequency words. A crosstabulation between morphological class and

frequency of words26 reveals that rates of retention are lower among frequent

monomorphemes than among infrequent ones.

Figure 4.4.10 Rates of retention by morphological class and word-frequency.

25
.914 and .918, respectively in adult speakers.
26
As mentioned in Chapter 2, values 1-3, in the SUBTLEX-UK corpus, represent low frequency words,
whereas values 4 -7 represent high frequency ones.

127
 Inter-speaker variation. If there is a pan-English effect, speaker differences

should show consistent range across categories. In East Anglia, the degree of inter-

speaker variation is clustered for monomorphemes and regular past verbs as in the

figure below, however there is inter-speaker dispersion for semi-weak verbs.

100

80
% deletion

60 M
S
40
R

20

0
18 + 87

Figure 4.4.11 Morphological class by individual speakers' age.

Figure 4.4.12 also includes individual speakers’ age, yet there seems to be no pattern

opposing young speakers to older speakers (Ciancia & Patrick, 2019).

The variability within the semi-weak class motivated Guy & Boyd (1990) to

study the effect of age demonstrating that children are all tightly clustered with high

probabilities of /t,d/ deletion in the semi weak class, but as they get older they treat

semi weak verbs as bimorphemic – a pattern consistent with age-grading. However,

Guy & Boyd’s (1990) findings cannot be replicated in the present study as the East

Anglian sample does not include any children.

128
 100

80

% deletion
60 young
middle
40
old
20

0
n't M S R

Figure 4.4.12 Interaction between morphological class and age, (p < 0.01).

4.4.5 Preceding segment

In the North American English literature, the preceding phonetic segment has usually

a weak effect on the application of (t,d) deletion, and it is commonly referred to as a

“tertiary constraint” (Guy, 1980: 20)27. However, in York (Tagliamonte & Temple,

2005) its effect is far stronger than the morphological one. Similarly, Sonderegger et

al. (2011) claim that once the preceding environment is included in the model,

morphological class falls out.

Group results. The East Anglian pattern resembles North American English dialects in

the statistical significance of this constraint and in its weak effect on the response but

differs in its phonetic conditioning. While the literature reports the typical following

ranking with more deletion after a preceding alveolar fricative and least deletion after

a preceding liquid:

/s/ > stops > nasals > other fricatives > liquids (Labov, 1989)

the East Anglian pattern exhibits more (t,d) absence after preceding nasals and less

deletion after a preceding non-sibilant fricative, yielding the hierarchy:

27
Santa Ana (1996), by contrast, reports factor weightings for the preceding environment as analogous
to the morphological class and greater than the following phonological context for Chicano English.

129
 nasal > sibilant fricatives > /l/ > stops > non-sibilant fricatives.

The high position of nasals does not line up with previous studies and does not

change even when glottal variants are included in the dataset, as illustrated in the

graph below.

Figure 4.4.13 Preceding environment in East Anglia: comparison between different

datasets

Labov et al. (1968) suggested that /t,d/ reduction after /s/ is governed by a separate rule

which could explain the high position of sibilants in (1). This correlates with the

theoretical sonority predictions, whereby the conventional expectation would be: less

sonorous preceding segments favour deletion (e.g. stops and fricatives), whilst more

sonorous preceding segments disfavour it (e.g. nasals)28 (Santa Ana, 1991). However,

the sonority hierarchy does not hold for the East Anglian pattern as the findings go in

a direction opposite to the usual claim. Indeed, nasals greatly influence the /t,d/

simplification at .669, followed by sibilants (.616), whilst /l/, stops and non-sibilant

28
Note that Santa Ana (1991) treated preceding liquids as exceptions and concluded that, in Chicano
English, preconsonantal /l/ is less resonant than General American English /l/. Hence it is attributed a
“lower sonority value to ChE /l/ less than to certain fricatives”.

130
fricatives have a negative effect and disfavour at .447, .412 and .353, respectively. The

association of increased sonority with increased deletion was also reported in Chicano

English (Santa Ana, 1991), and in African American Vernacular English (Patrick et al.,

1996) where preceding /l/ was the most highly influential phonetic segment. This

negative effect of the sonority hierarchy in East Anglia seems to hold for both York

and Manchester too, due to the high place of resonants and the low position of less

sonorous segments:

(1) York (Tagliamonte & Temple, 2005)

sibilants > liquids > nasal > stops > fricatives

(3) Manchester (Baranowski & Turton, 2020)

sibilants > affricate > nasal > stop > fricative > lateral

Thus, in the light of these findings, (t,d) seems not to be governed by a sonority

hierarchy in British English. By contrast, a closer correlation with the theoretical

sonority hierarchy was found in Jamaican Creole (Patrick, 1999), where phonetic

environments in (1) are re-ordered:

sibilants > stops > fricatives > nasals > lateral.

This exchanging of place between nasals and fricatives “unites the two natural classes

of segments, which are neighbours in the sonority hierarchy” (Patrick, 1999: 144).

The high position of nasals and sibilant fricatives in the East Anglian ranking (2) is

predicted by the Obligatory Contour Principle (OCP, Goldsmith, 1976), as discussed

by Guy & Boberg (1997), whereby preceding segments trigger the absence of the final

coronal stop if they share two or more features. Along the line of autosegmental

phonology, the features shared with the target variable differ among the nasals /m/, /ŋ/

and /n/: /m, ŋ/ share only [-cont]; /n/ shares [+cor, -cont]. Hence, a preceding /n/ is

131
more likely to favour deletion. However, in the present study, due to the low number

of tokens in the preceding /m/ environment (e.g. dreamt), all nasals were combined29.

The second most favouring segment in the hierarchy is represented by sibilant

fricatives, which share the features [+cor, -son]. What is surprising, under this

approach, is the high place of /l/ which precedes stops in the ranking. If adjacent

identical segments and features are prohibited, then preceding stops

[-son, -cont] would create OCP clashes more than preceding /l/, which shares with

/t,d/ only the [+cor] feature. Non-sibilant fricatives, as expected, are the least

environment to favour deletion being identical with the target variable only in the [-

son] feature.

Comparision with York and Manchester. Figure 4.4.14 compares results from East

Anglia with York and Manchester’s findings.

Figure 4.4.14 Rates of deletion according to the preceding environment in East

Anglia, York, and Manchester.

29
The number of tokens with preceding /m/ equals to 93, whilst tokens with preceding /n/ are equal to
1909.

132
Both stops and fricatives show low rates of deletion, and notably higher rates of

deletion after sibilants, with East Anglia resembling York. Yet, after nasals East Anglia

resembles Manchester. As discussed above, in East Anglia, the behaviour of preceding

/l/ and stops is not consistent in terms of the OCP. In York, however, not only is the

probability of deletion parallel between /l/ (.43) and stops (.43), but they are also

comparable to the behaviour of nasals (.45) 30. Temple (2009) points out that high

deletion rates with a preceding /n/ and /s/, such as in East Anglia and Manchester, might

be due to the homorganic place of articulation with the final coronal stop.

4.4.6 Non-significant predictors – Overall Results

When reporting statistical results, it is suggested that “all independent variables tested

should be reported, whether significant or not. Non-significance of a potential predictor

is an important finding” (Guy, 2018). Along this line, this section presents results from

Rbrul one-level analysis to explore additional independent variables included in the

model with the hypothesis that they might have an effect on the response, but which

were not marked as significant predictors (see Appendix I for further details).

Fasold (1972), Wolfram (1972), Labov (1989), Bayley (1994)31, and Roberts (1994)

found that (t,d) deletion is more likely to occur in unstressed syllables than in

stressed ones. In the present study, even though the findings are in the right direction

from what we would expect, this predictor failed to reach statistical significance,

similar to Tejano English (Bayley, 1994). Moreover, social factors such as sex, age

and class do not exhibit significant conditioning of (t,d) deletion.

30
These are probabilities; however, percentages differ more.
31
In this case, I refer to the results of adult speakers, where unstressed syllables undergo (t,d) deletion
at a rate of 51%, whilst in unstressed syllables the word final cluster reduction is less likely to occur
(34%). Results among young Tejanos, instead, go in the opposite direction.

133
In East Anglia, males delete slightly more than females; working class speakers do

not delete considerably more than middle class members with a rate of 27% and

25%, respectively, and all age-groups exhibit (t,d) absence at the same rate. In view

of these findings, the profile of (t,d), in East Anglia, is that of a stable variable.

Location does not surface as a conditioning predictor of /t,d/, and the effect of locality

depends on the level of other predictors as shown in section 4.5.4, where the interaction

between locality and other predictors will be discussed.

Since phonetic reduction is also highly conditioned by word frequency, this predictor

was initially included in the model with the hypothesis that deletion occurs more in

high-frequency words than in low frequency ones (Bybee 2002). However, the East

Anglian database did not reveal a significant variation between low and high frequency

words, hence this factor group was excluded post-coding. Figure 4.4.15 indicates the

non-significant trend of word-token frequency, with more frequent words exhibiting a

higher deletion rate. Overall, however, the lexical frequency of this database also shows

high retention rates in both low and high frequency tokens.

Figure 4.4.15 Word frequency from SUBTLEX-UK and percentages of deletion.

134
Purse & Tamminga (2019), as outlined in chapter 3, take a different view on lexical

frequency and argue that previous invisible significance of frequency effects emerges

if lexical frequency is measured by Root frequency.

So far, we have discussed results from the three combined East Anglian locations.

Let us now analyse Colchester, Ipswich and Norwich separately to explore whether the

linguistic and social constraints indicate consistency across the urban areas.

4.5 Treating Colchester, Ipswich and Norwich separately

The overall results revealed a contrastive behaviour between sibilant fricatives and

non-sibilant fricatives in the following environment, with non-sibilants disfavouring

deletion mostly due to following retained [h]. If location is not a necessary predictor

for (t,d) since constraints across the English dialects are parallel, with the exception of

following pause, as evinced in the literature (Labov, 1989), we should expect all

linguistic constraints – including the different behaviour of sibilant fricatives and non-

sibilant fricatives as well as the role of following /h/ - to be consistent across the three

localities examined. Colchester, Ipswich and Norwich will be singly analysed, in this

section, and the dataset will be split accordingly. Firstly, I will show how each locality

patterns with the overall East Anglian findings; secondly, results will be theoretically

discussed, and finally I will show the interactions between locality and other predictors.

4.5.1 Results from Colchester

Table 4.5.1 indicated results of a step-up/step-down regression analysis, with speaker

as a random effect, showing that following phonetic segment, voicing agreement, style,

morphological class and preceding phonetic segment have a significant conditioning

of (t,d).

135
Following environment. Analogous to the overall findings, following nasals trigger

deletion the most (.812), followed by sibilant fricatives which favour at .768; liquids

do not behave differently in Colchester, indeed table 4.5.1 suggests that both /l/ and /r/

affect the word-final cluster reduction at .698 and .672, respectively. This variable is

also significantly affected by a following stop (.672) with the same weight as a

following /r/. On the contrary, glides (.389), non-sibilant fricatives (.343), vowels

(.187) and pause (.112) disfavour deletion. Among non-sibilant fricatives, the

remarkable role of /h/ is confirmed since the high /t,d/ retention is due to following

underlyingly consonantal [h]. These findings resemble the overall East Anglian

pattern, illustrated in (3), except for liquids.

(3) East Anglian pattern:

nasals > sibilants > stops > /l/ > glides > /r/ > non-sibilant fricatives > vowels > pause

(4) Colchester:
nasals > sibilants > /l/ > /r/ > stops > glides > non-sibilant fricatives > vowels > pause.

Why do liquids behave differently with respect to the East Anglian pattern? The

deletion rate before a following /l/ (71%) is higher compared to the rates of (t,d)

absence before a following /r/ (41%; though the factor weights are quite close, above).

This suggests that resyllabification does not provide an explanation that holds for the

Colchester data.

This could be due to: (a) the relatively low number of tokens in these categories,

indicated in table 4.5.132 (b) clearness/darkness of onset /l/ (e.g LEAF).

32
When the three locations are combined the dataset shows 81 occurrences before a following /r/ and 44
tokens before a following /l/. In York, however, the number of tokens in the following /r/ and following
/l/ contexts is lower and equals to 32 and 26 occurrences, respectively.

136
Following (b), Baranowski & Turton (2016) suggest that the place of articulation of a

following /l/ may play a notable role. That is, in varieties which exhibit dark initial /l/s,

such as Manchester and American dialects, the tongue tip gesture is missing or delayed

thus /-t,d/ is more likely to be deleted owing to the lack of coronal place.

This explanation might account for the relatively higher deletion before laterals in

Colchester and perhaps Ipswich, compared to Norwich where at least post-vocalic /l/

has a history of being clear after high front vowels (Trudgill 1974). Johnson & Britain

(2007: 299) note for northern East Anglia, “/l/ was clear in all positions… well into the

20th century”.

Voicing agreement. The second significant predictor in the regression analysis is

voicing agreement. The reversed and unexpected finding, as discussed in 4.4.2, holds

for Colchester with heterovoiced tokens favouring the cluster simplification at .763,

and homovoiced tokens disfavouring at .237.

Style. Style is the third predictor to affect this variable, with a notable reduction in

spontaneous speech (.793), whilst in reading passages and word lists the apical stops

are more likely to be retained, indeed they disfavour deletion at .499 and .208,

respectively.

137
Application value = deletion; overall proportion = 0.263
R2 = 0.667; log likelihood = -536.001; N =1615

Constraints Logodds FW % Tokens

Following Env.
p<.001
nasals 1.463 0.812 71 56
sibilant fricatives 1.197 0.768 77 77
/l/ 0.838 0.698 71 17
/r/ 0.715 0.672 41 22
Stops 0.431 0.672 55 136
glides -0.452 0.389 40 94
non-sibilant fricatives -0.649 0.343 33 110
vowels -1.471 0.187 20 470
pause -2.072 0.112 10 633
Voicing agreement
p<.001
heterovoiced 1.172 0.763 60 309
homovoiced -1.172 0.237 18 1306
Style
p<.001
spontaneous speech 1.340 0.793 38 893
reading styles -0.003 0.499 19 375
word lists -1.337 0.208 3 347
Morphological class
p<.01
n't contractions 0.766 0.683 77 66
monomorphemes 0.376 0.593 39 726
semi-weak verbs -0.288 0.429 20 231
regular verbs -0.854 0.299 8 592
Preceding Env.
p<.01
sibilant fricatives 0.594 0.644 28 419
nasals 0.493 0.621 38 684
/l/ 0.205 0.551 20 93
non-sibilant fricatives -0.158 0.461 11 179
Stops -1.135 0.243 5 240

Table 4.5.1 Multivariate analysis of (t,d) deletion in Colchester.

138
Morphological class. The influence of morphological class, in Colchester, is consistent

with the overall East Anglian results and matches the pan-English pattern: negative

contractions favour deletion the most (.683) followed by monomorphemes (.593);

whereas semi-weak verbs (.429) and regular past tense verbs (.30) exhibit a negative

correlation.

Preceding environment. The last predictor which indicates a significant conditioning

of (t,d) is the preceding phonetic segment. Differently from the overall findings,

indicated in (5), in Colchester the greatest effect is contributed by sibilant fricatives

(.644), followed by nasals (.621) and /l/ (.551). Consistently with the findings of

previous studies, both non-sibilant fricatives and stops disfavour deletion.

(5) East Anglian pattern:

nasals > sibilant fricatives > /l/ > stops > non-sibilant fricatives

(6) Colchester:
sibilant fricatives > nasals > /l/ > non-sibilant fricatives > stops

Theoretically, the exchanging of position between nasals and sibilant fricatives on the

one hand, and stops and non-sibilant fricatives on the other, seems not to be problematic

with respect to the East Anglian pattern. In the first case, under the OCP approach, both

sibilant fricatives and /n/ share two features with the target variable. Since the present

study combines all nasals, this result could be justified if preceding /m/ triggered the

most deletion. However, crosstabulations revealed that 59% of (t,d) absence occurs

after a preceding /m/, whilst 81% of deletion was found after a preceding /n/. In the

second case, even though both predictors are marked as disfavouring factors, deletion

rates after non-sibilant fricatives are slightly higher than deletion rates after stops – a

result which goes in a direction opposite to what the OCP predicts. Thus, taking also

into account the high place of /l/, it appears that the OCP does not hold for the

139
Colchester data. With respect to sonority, despite the high position of sibilants in the

hierarchy, it represents an obstacle to the theoretical sonority predictions reported by

Santa Ana (1996).

4.5.2 Results from Ipswich

In Ipswich, the following context remains the strongest predictor for (t,d) deletion,

whereas the interpretation in terms of sonority is not straightforward. Indeed, the high

place of nasals in the hierarchy persists. Table 4.5.2 indicates results of a multivariate

analysis with following phonetic context, voicing agreement, style, preceding phonetic

context and morphological class marked as statistically significant factor groups.

Following environment. The greatest effect is provided by sibilant fricatives and nasals

with no difference in probabilities (.832) following stops and following /l/ also favour

deletion at .758 and .615, respectively, whereas glides (.481), non-sibilant fricatives

(.392), vowels (.258), /r/ (.179) and pause (.153) strongly disfavour. The phonetic

factors mirror the East Anglian pattern despite some slight ranking differences,

indicated in (7). The divergent behaviour of sibilant fricatives and non-sibilant

fricatives is repeated in Ipswich and, once again, following consonantal [h] highly

contributes to the retention of /-t,d/.

(7) East Anglian pattern:

nasals > sibilants> stops > /l/ > glides > /r/ > non-sibilant fricatives > vowels >
pause
(8) Ipswich:
sibilants > nasals > stops > /l/ > glides > non-sibilant fricatives > vowels > /r/ >
pause

In section 4.5.1, we saw that, in Colchester, the magnitude of the liquids effect is nearly

the same, with both /l/ and /r/ favouring deletion. By contrast, in Ipswich, liquids

140
behave as in previous empirical findings of North American dialects, with

resyllabification as a possible explanatory factor.

Voicing Agreement. The reversed order of heterovoiced and homovoiced tokens also

holds for Ipswich. The former favours deletion at .808, whilst the latter disfavours at

.192.

Style. The third significant predictor of the multivariate analysis is style, with /-t,d/

being more likely to be simplified in spontaneous speech (.833) and reading styles

(.625), than in word lists which exhibit a negative correlation (.107).

Preceding environment. In Ipswich, the effect of the preceding segment indicated in

the regression analysis is slightly stronger than the morphological one. Nasals (.695)

show the most deletion, followed by sibilant fricatives (.618) and stops (.601), where

the difference in terms of factor weight is negligible. Preceding consonantal /l/ (.428)

and non-sibilant fricatives (.194) disfavour deletion. These findings are consistent with

the East Anglian pattern, despite the exchange of position between /l/ and stops,

indicated in (10).

(9) East Anglian pattern:

nasals > sibilant fricatives > /l/ > stops > non-sibilant fricatives

(10) Ipswich:
nasals > sibilant fricatives > stops > /l/ > non-sibilant fricatives.

141
Application value = deletion; overall proportion = 0.26
R2 = 0.714; log likelihood = -518.733; N = 1658
Constraints Logodds FW % Tokens
Following Env.
p<.001
sibilant fricatives 1.599 0.832 82 55
nasals 1.598 0.832 68 65
stops 1.141 0.758 58 133
/l/ 0.469 0.615 64 14
glides -0.076 0.481 42 101
non-sibilant fricatives -0.439 0.392 29 123
vowels -1.058 0.258 20 496
/r/ -1.521 0.179 27 26
pause -1.712 0.153 11 645
Voicing Agreement
p<.001
heterovoiced 1.437 0.808 69 286
homovoiced -1.437 0.192 17 1372
Style
p<.001
spontaneous
speech 1.608 0.833 36 973
reading styles 0.509 0.625 21 353
word lists -2.118 0.107 2 332

Preceding Env.
p<.01
nasals 0.825 0.695 42 695
sibilant fricatives 0.480 0.618 21 419
stops 0.410 0.601 10 257
/l/ -0.291 0.428 21 86
non-sibilant fricatives -1.424 0.194 4 201
Morphological
class p<.01
n't contractions 0.435 0.607 82 70
semi-weak verbs 0.326 0.581 23 229
monomorphemes 0.197 0.549 37 750
regular verbs -0.959 0.277 7 609

Table 4.5.2 Multivariate analysis of (t,d) deletion in Ipswich.

142
Theoretically, this reordering of stops and /l/, in the Ipswich pattern, is far from being

problematic as it is in the direction predicted by the OCP: the more features are shared,

the more likely it is that /-t,d/ will be simplified.

Morphological class. Similar to Colchester, n’t negative contractions indicate the

greatest effect on (t,d). However, the final model reports the following ranking:

n’t contractions > semi-weak verbs > monomorphemes > regular verbs, with the latter

being the only predictor to exhibit a negative correlation. Table 4.5.2 indicates that the

factor weight of semi-weak verbs is .581, whereas the weight of monomorphemes

equals .549. Why? Step-down analysis shows that when the preceding segment is

dropped from the model monomorphemes and semi-weak verbs are reordered: n’t > M

> S > R, and they both favour deletion. Cross-tabulations showed that there is an

interaction with the preceding environment. A closer inspection revealed that there is

absence of tokens after preceding sibilant fricatives in the semi-weak category. With

respect to the high place of semi-weak verbs, Tamminga & Fruehwald (2013) also

report a greater deletion rate variance across speakers in semi-weaks than in

monomorphemes (or regular past tense verbs) for the Buckeye corpus (Pitt et al., 2007).

It is worth noting that, whatever the order, both monomorphemes and semi-weaks

favour (t,d) deletion. Let us find out, in the following section, how (t,d) behaves in

Norwich.

4.5.3 Results from Norwich

This section reports findings of a step-up/step-down regression analysis, with following

phonetic segment, style, voicing agreement, preceding environment, morphological

class and sex marked as statistically significant predictors.

143
Following environment. Nasals (.817) are the most favouring factor followed by

sibilant fricatives (.813) which, once again, behave differently from their non-sibilant

counterpart. Stops (.70) trigger the cluster simplification as well as following /r/ (.631).

The latter, however, is inconsistent with the resyllabification theory and does not match

the behaviour of /r/ in Ipswich, where the underlying liquid blocks the deletion of /t,d/

and it is in line with previous US English studies (e.g. Guy, 1980). Following /l/ (.22),

by contrast, disfavours deletion along with non-sibilant fricatives (.497), glides (.389),

pause (.218) and vowels (.204). The reversed behaviour of liquids in Norwich might

be due to the clearness/darkness of onset /l/, that is: the coronal place of articulation of

clear /l/ is suggested to favour retention of (t,d), as for York speakers, owing to the non-

delayed tongue tip gesture (Baranowski & Turton, 2016).

Hence, the same explanation might account for Norwich, where at least post-vocalic /l/

has a history of being clear after high front vowels (Trudgill, 1974). Johnson & Britain

(2007: 299) note for northern East Anglia, “/l/ was clear in all positions… well into the

20th century”. Overall, these results mirror the East Anglian pattern, apart from the

reversed behaviour of liquids and exchange of place between non-sibilant fricatives

and glides.

(11) East Anglian pattern:

nasals > sibilants > stops > /l/ > glides > /r/ > non-sibilant fricatives > vowels >
pause
(12) Norwich:
nasals > sibilants > stops > /r/ > non-sibilant fricatives > glides > /l/ > pause >
vowels.

Style. Parallel to Colchester and Ipswich, /t/ and /d/ are more likely to be deleted in the

spontaneous speech , which favours at .838, than in reading styles (.548) and word lists

(.138).

144
Application value = deletion; overall proportion = 0.248
R2 = 0.688; log likelihood = -497.504; N = 1606
Constraints Logodds FW % Tokens
Following Env.
p<.001
nasals 1.493 0.817 57 76
sibilant fricatives 1.467 0.813 72 60
stops 0.849 0.7 55 155
/r/ 0.535 0.631 24 33
non-sibilant
fricatives -0.012 0.497 35 110
glides -0.450 0.389 38 103
/l/ -1.241 0.224 31 13
pause -1.278 0.218 11 617
vowels -1.364 0.204 16 439
Style
p<.001
infomal 1.641 0.838 37 844
reading passages 0.193 0.548 19 403
word lists -1.835 0.138 2 359
Voicing
Agreement
p<.001
heterovoiced 1.157 0.761 65 294
homovoiced -1.157 0.239 16 1312
Preceding Env.
p<.001
nasals 0.916 0.714 43 624
sibilant fricatives 0.493 0.621 22 388
/l/ -0.316 0.422 13 178
non-sibilant
fricatives -0.455 0.388 8 185
stops -0.638 0.346 5 231
Morphological
class p<.001
n't contractions 1.347 0.794 92 60
monomorphemes -0.146 0.464 35 728
semi weak verbs -0.280 0.431 22 227
regular verbs -0.922 0.285 6 591
Sex
p<.01
males 0.36 0.589 29 803
females -0.36 0.411 21 803

Table 4.5.3 Multivariate analysis of (t,d) deletion in Norwich.

145
Voicing agreement. The previously discussed unconventional finding, which is due to

interaction with the preceding phonetic segment (see section 4.4.2), also holds for

Norwich. Indeed, heterovoiced tokens (.761) exhibit a positive correlation compared to

the homovoiced ones which disfavour at .239.

Preceding environment. Similar to Ipswich, the effect of the preceding phonetic

segment is stronger than the morphological class. The greatest effect is contributed by

nasals (.714) followed by sibilant fricatives (.621), whereas /l/ (.422) along with non-

sibilant fricatives (.388) and stops (.346) show a negative correlation. Similar to the

East Anglian pattern, the findings from Norwich only exhibit a marginal change at the

end of the hierarchy. In terms of the OCP, not only is the high position of /l/ surprising,

but also the higher place of non-sibilant fricatives compared to stops.

(13) East Anglian pattern:

nasals > sibilant fricatives > /l/ > stops > non-sibilant fricatives

(14) Norwich:
nasals > sibilant fricatives > /l/ > non-sibilant fricatives > stops

Morphological class. Not surprisingly, morphological class is marked as a significant

predictor following the ranking n’t > M > S > R. Table 4.5.3 shows that when the

negative contraction category is included in the model it is the only favouring predictor

(.794); however, when n’t tokens are excluded from the analysis monomorphemes

favour deletion whilst semi-weaks and regular verbs disfavour deletion.

Sex. So far, none of the social variables presented any statistical significance except for

sex in Norwich. Results from this location show that men delete significantly more than

females, despite the low rate difference between the two factor groups. In the light of

what has been presented and discussed so far, overall (t,d) seems to behave uniformly

146
across the three East Anglian locations. The next section, indeed, will demonstrate this

graphically.

4.5.4 Interaction between locality and other predictors

This section sets out to discuss significant and non-significant interactions between

locality and three major constraints: preceding phonological environment, following

phonological environment and morphological class. Overall, the three predictors

exhibit minor divergences between them, which are mainly due to slight ranking

differences. Figure 4.5.1 illustrating a significant interaction, shows the trend of (t,d)

in the preceding phonetic segment.

80

60
% deletion

Colchester
40
Ipswich
Norwich
20

0
stops fricatives sibilants nasals /l/

Figure 4.5.1 Interaction between preceding environment and locality, p <.001.

Figure 4.5.2, figure 4.5.3 and 4.5.4 illustrate probabilities of deletion in the following

environment across the three localities and highlights some dissimilarities with the East

Anglian hierarchy (overall results)33. Non-sibilant fricatives and sibilant fricatives

strongly diverge in the overall East Anglian pattern, with the latter favouring the /t,d/

reduction and the former heavily disfavouring deletion. As discussed in section 4.4.1,

this result is due to a following consonantal /h/ which plays a considerable role in

33
The interaction between locality and following environment is not statistically significant (p > .05).

147
disfavouring deletion among non-sibilants. The consistency of this finding in the three

urban areas proposes that (t,d) is highly retained before following initial /h/ - whether

/h/ is realised or dropped. Therefore, this result suggests that the null hypothesis may

prove false. Following pause, which typically varies across speech communities

(Labov 1989; Patrick 1999), consistently disfavours deletion in Colchester, Ipswich

and Norwich.

pause
vowels
non-sibilant fricatives
/r/
glides
/l/
stops
sibilant fricatives
nasals
0 0.2 0.4 0.6 0.8 1

Figure 4.5.2 Probabilities for the following environment in Colchester.

pause
/r/
vowels
non-sibilant fricatives
glides
/l/
stops
nasals
sibilant fricatives
0 0.2 0.4 0.6 0.8 1

Figure 4.5.3 Probabilities for the following environment in Ipswich.

pause
vowels
non-sibilant fricatives
/r/
glides
/l/
stops
sibilant fricatives
nasals
0 0.2 0.4 0.6 0.8 1

Figure 4.5.4 Probabilities for the following environment in Norwich.

148
Figure 4.5.5 illustrates the consistency of the morphological effect across the three

urban areas, proposing the emergence of the expected morphological effect for (t,d) in

East Anglia. Even in this case, the interaction between locality and morphological class

is not statistically significant (p > .05).

100

80

n't
% deletion

60
M

40 S
R
20

0
Colchester Ipswich Norwich

Figure 4.5.5 Morphological class across the three East Anglian localities.

4.6 Summary

This apparent-time survey on (t,d) deletion in East Anglian English has contributed to

the field expanding on the (t,d) literature: a) new phonological insights were revealed

after breaking down the consonant category for the following phonological

environment, revealing that following sibilants and non-sibilant fricatives behave

differently. Moreover, this Chapter reports local dialectal variation after laterals, yet

not before /h/. b) It has shed light on the unsolved problem of morphological class in

British English suggesting the emergence of the expected morphological effect for (t,d)

in East Anglia. Finally, (t,d) deletion does not seem to vary due to morphological

absence in semi-weak verbs. (t,d) is not governed by a sonority hierarchy and does not

report visible OCP effects. (t,d) absence is not the only phonological phenomenon that

can apply word-finally in British English; another typical feature of East Anglian

149
English which is common in most British dialects is (t) glottaling, the subject of the

next chapter.

150
Chapter 5 - (t) GLOTTALING: LITERATURE REVIEW

This chapter extends the discussion on coronal /t/. In Chapters 3 and 4 we have

discussed the reduction of /t,d/ in final consonant clusters but, deletion is not the only

type of variation that /t/ can undergo in British English dialects. In this Chapter, I begin

with a phonetic description of the coronal /t/ and its allophone [Ɂ], with some

terminological remarks which distinguish (t) glottaling from glottal reinforcement.

These remarks also include additional phonetic components identified as a need to

amplify the envelope of variation for (t) (Straw & Patrick, 2007). Before exploring the

place of the glottal variant in phonological theory and in variationist sociolinguistic

theory, I will critically review the diachronic development of the glottal stop in British

English to challenge traditional assumptions which consider London a source of

innovation for (t) glottaling as well as other southern phonological changes. Since the

literature on (t) glottalisation and (t) glottaling is quite extensive, it will be reviewed

selectively here, with the goal to outline previous studies which are relevant to the

present research, as a point of comparison. Finally, a critical evaluation follows which

will lead us to formulate specific research questions.

5.1 Articulatory description of the plosive /t/ and its variant [Ɂ]

As a starting point, I will describe the manner of articulation of plosives which involves

three stages: (1) closure in the vocal tract, (2) compression of lung air behind the

obstruction and (3) airstream release (Hughes et al., 2012). A feature matrix for /t/ is

[-continuant] [-sonorant] [+coronal] [-voiced] (Roca and Johnson, 1999), and it is

articulated by the contact between either the tip or the blade of the tongue and the

alveolar ridge. Its released voiceless segments can be of various kinds: when the

151
release of the underlying plosive is accompanied by a strong burst of air, the phoneme

/t/ is called aspirated and it is articulated as [tʰ]. The latter may be often used before

stressed vowels, whilst unaspirated [t] can be found elsewhere. In final pre-pausal

environment, instead, the distribution of aspirated [tʰ] and unaspirated [t] is in non-

contrastive distribution and varies freely (Wells, 1982).

Cruttenden (1994) observed that the alveolar stop /t/ contact is conditioned by

the place of articulation of a following consonant. Hence, when the following segment

is represented by the approximant /ɹ/, as in try, there will be a post alveolar contact [t̠ ],

whilst when followed by /θ, ð/ as in eighth, the contact will be dental [t̪ ]1. Moreover,

syllable-final /t/ can undergo assimilation processes, generally regressive (i.e. that man

[ðæp mæn]) or coalescent (/t/ + /j/ in What you want [wɒtʃʊ wɒnt), implying variation

in the place of articulation. However, the phonemic assimilation of place is sporadically

finalised; spectrograms, indeed, reveal that some traces on the alveolar ridge

complement the articulation.

In weakly accented contexts, /t/ is liable to affrication as in time [tˢaɪm]. This

happens when the energy of release reaches affrication which is described as a short

“period of audible friction” (Laver, 1994: 363). In many English varieties, /t/ may be

realised as a flap [ɾ], articulated with the tip or the blade of the tongue at the alveolar

ridge (Ladefoged, 2005). In addition to this kind of variation involving plosives,

syllable-final /p, t, k/ can be replaced by the glottal stop.

The articulatory uniqueness of the glottal stop, in the English language, lies in the fact

that it is the only non-oral obstruent-allophone (Pointer, 1996). Its primary articulation

springs from the airstream obstruction and the total closure of vocal folds which

suddenly release. Since there is no air passing during the glottal closure, [ʔ] is

1
Both dental and alveolar variants of /t/ are also referred to as apical or laminal.

152
physiologically a voiceless sound (Gimson 1980; Roca & Johnson 1999). This explains

why the glottal stop is mostly used as an allophone for voiceless obstruents /p, t, k/, as

it shares significant and intelligible features with the phoneme it replaces2. In this case,

[ʔ] shares two features: stop and voiceless, and differs in the place of articulation

(Pointer, 1996).3

It has to be underlined that the articulatory description of the glottal stop given

above is referred to glottalled plosives, and diverges from the articulation of glottalised

variants, which may also involve either a ‘creaky voice’ or ‘harsh voice’.

At this point, some terminological remarks need to be clarified as they appear

sometimes confusing in the literature. Glottaling is adoped to indicate the realisation

of a consonant as a glottal stop, usually in syllable-final or morpheme-final context

when /p, t, k/ are followed by a consonant, such as Scotland (Hughes, et al, 2012)4.

Hence, these variants are often called glottalled plosives, and the total replacement of

/t/ by the glottal stop is usually referred to as T-glottaling (Wells, 1982).

As mentioned above, articulatory features for the glottal stop diverge from the

articulation of glottalised variants. Glottal reinforcement (i.e. pre-glottalisation, post-

glottalisation), in fact, is defined as: “a process whereby the primary supralaryngeal is

accompanied by a secondary stricture of the glottal level” but this articulation fails to

create a glottal constriction since it does not reach the utmost stricture of the full glottal

2
From a taxonomic-phonetic standpoint, the glottal stop violates the biuniqueness requirement by which
“one speech sound must be uniquely assigned to a given phoneme…in a unique way”; see Wells (1982:
54) and Lass (1984) for further details.
3
Other features such as lip rounding and tension are not mentioned here as they are not relevant to the
discussion.
4
The use of the glottal stop as an allophone of /p/ and /k/, however, is more restricted and depends on
the place of articulation of the following consonant: if both following consonant and the segment that is
being replaced share the same place of articulation /p/ and /k/ will be more likely to be glottalled, as in
back garden [baʔ ˈɡɑː(r)d(ə)n].
Since in southern England the accent is non-rhotic, there would be hiatus if non-prevocalic /r/ would
encounter a following vowel, therefore hiatus is resolved by linking /r/, which can also be replaced by
the glottal stop. The latter, in addition, can also stand in place of intrusive /r/. A more detailed description
is provided by Pointer (1996).

153
stop (Laver, 1994: 330). In many British accents glottal reinforcement is linked to the

reinforcement of voiceless and affricate /p, t, k, ʧ/ in syllable final contexts. In such

cases, the glottal stop follows the consonant [tɁ] (Trudgill, 1974) and takes the name

of T-glottalisation when /t/ is the plosive reinforced. However, Straw and Patrick

(2007) observed that the term glottalisation is sometimes used vaguely in the literature

to indicate both glottal substitution and glottal reinforcement; Docherty and Foulkes

(1999: 57) adopted the label ‘glottal(ised)’ claiming that it covers “two distinct

[acoustic] factors”.

In the case of a non-complete closure of the glottis, a distinct phenomenon

usually called creak can be produced (Laver, 1994). Cruttenden (2014) describes creak

as a creaky voice which involves energy to the vocal tract as well as a slow vibration

of the vocal folds. He also distinguishes creaky voice from harsh voice with the latter

referring to the vibration of false vocal folds5.

The ejective stop is another variant of glottal reinforcement, articulated with an

egressive stream of air. In other words, when the oral closure is realised, a total glottal

closure is held. Ejectives resemble glottalised variants due to the glottal constriction

involved during the articulation, yet the relative release timing is different. In the case

of ejective stops, indeed, the oral release is the first (Laver, 1994). As claimed in the

literature, this articulation can also occur with voiceless and affricate /p, k, ʧ/ indicating

that they are more similar to glottalised variants than t-glottaling, however this question

will not be addressed in this thesis as it is beyond the purpose of this study. To amplify

and redefine the envelope of variation for (t), Straw & Patrick (2007: 395), in the

Ipswich study, noted the following variants [th] [t ̚ ] [t’] [d] [ɾ] [ø] [ʔt] [tʔ].

5
Creaky voice refers to the slow rate of vibration of the vocal folds (Laver, 1980).

154
 So far, we have made some terminological remarks and explored the

articulatory features of the phonetic variants, summarised in the tables below:

Phonetic variants of /t/ Glottal variants of /t/

Unaspirated [t] T-glottaling [Ɂ]

Aspirated [tʰ] Creaky voice creak

Affricated [tˢ] Glottal reinforcement [Ɂt] [tɁ]

Tapped [ɾ] Ejective [t’]

Table 5.1.1 Phonetic variants and glottal variants of /t/.

This helps explain the place of t-glottaling within two branches of linguistic theory:

Phonology and Sociolinguistics.

5.2 T-Glottalisation in Phonological Theory

This section attempts to provide a brief overview of how t-glottalisation has been

treated from a theoretical viewpoint and what is meant by lenition/weakness in

phonological theory, however an exhaustive phonological discussion on what counts

as weak including a segment’s contexts or its intrinsic properties is not under scrutiny

here6.

From a phonological standpoint, (t) glottaling can be classified under the label

of lenition processes. Hyman (1975: 165) explains lenition in the following terms: “a

segment X is said to be weaker than a segment Y if Y goes through an X stage on its

way to zero.” This process, also known as debuccalisation, refers to a consonant losing

its place of articulation in the penultimate stage of weakening.

6
For a more detailed account on the current perspectives on lenition see Honeybone (2008).

155
 Zuraw (2009) corroborates Hyman’s (1975) definition of lenition arguing that

weak sounds are those that are vulnerable to deletion.

Harris and Kyne (1990), cited in Docherty et al. (1997: 286), claim that lenition

is divided into two stages: breaking and element-loss: “breaking involves rearranging

the occlusion and coronal elements into a contour structure which is parallel to that

normally assumed for prenasalised stops, light dipthongs or affricates.”

According to the generative phonological theory, lenition is regarded as the spreading

of autosegmental features to neighbouring phonetic segments (Mascaró, 1983).

Carr’s (1999) description of (t) glottaling as a weakening process is explained

as a saving in articulatory time and effort: since the glottal stop lacks oral articulation,

the tongue is free to assume its position for the following phonetic segment. However,

in Carr’s (1999) explanation, it is not clarified whether glottal reinforcement is deemed

as a weakening or strengthening process. The latter occurs when consonants gain a

different feature or their articulation is affected by a neighbouring sound (Pennington,

1996).

Harris (1994: 120) describes t-glottaling as a type of lenition which is a

transitional stage between apical stop and elision

Plosive > ʔ (Glottaling) > Ø (Deletion)

explaining that /t/ can be glottalled on the basis of independent parameters of the

language under investigation, taking also into account universal principles of

phonological structure. Along this line, Kirchner (2004) proposed a constraint ranking

LAZY>>PRES(place), where the place of articulation is preserved and the LAZY

restriction aims at minimising the articulatory effort. Moreover, Kirchner (2004)

suggests the following hierarchy as the most likely for /t/ to debuccalise:

156
 Pre-consonantal syllable > preceding a vocalic nasal > word final preceding a vowel

> preceding a vocalic /l/7 > foot internal.

Many researchers have proposed scales of strength suggesting that segments’

degree of weakness depends on their articulatory or acoustic features (Zuraw, 2009).

Lass & Anderson (1975) believe that one criterion for the classification of consonantal

strength is the resistance to airflow through the vocal tract, that is, the less resistance,

the weaker the segment.

Carr (1991) examined “weakening” and glottalisation in Tyneside English

accounting in particular for environments which trigger weakening more than

glottalisation. By weakening Carr refers to the process by which /t/ is realised as [ɾ] or

[ɹ]8. As Docherty et al. (1997) highlight, Carr’s prediction is that English feet are

trochaic, hence glottalisation is not likely to occur in words such as attack, at Easter,

etc. He claims that, unlike weakening, glottalisation can be found in nouns, adjectives,

prepositions as well as monosyllabic words followed by an unstressed syllable.

However, due to variability within the postulated claim, he proposes that “weakening

applies to feet formed under cliticization post-lexically” (p. 283) which occur before

glottalisation is applied, and tokens such as fitter and fit her (glottalised and weakened

respectively), are apparently regarded to be in complementary distribution.

The Sonority Sequencing Principle (Goldsmith 1990, Laver 1994) has been also

adopted to explain the behaviour of word-final /t/ glottaling in RP (Barrera, 2015).

Barrera’s (2015) results revealed that the more sonorous the following segment, the more

it promotes the glottal stop. However, word-medial /t/ glottaling was found not to follow

completely the sonority scale as vowels were ranked at the bottom of the hierarchy in

7
Kirchner’s (2004) hierarchy does not contain the process of vocalisation, although this phenomenon
needs to be included in the lenition scale.
8
This phenomenon is also known as the T-to-R rule (Wells, 1982).

157
intervocalic context. She claims that the low position of vowels in this phonological context

could be due to the fact that the glottal stop has long been stigmatised intervocalically and

it is mainly related to working-class accents.

A further theoretical prediction related to t-glottalisation has been made by

Halle and Kaye (1990) who, in line with the Complexity Condition, point out that the

replacement of /t/ with [ʔ] is blocked when /t/ is preceded by an obstruent, particularly

if the latter has undergone vocalisation. By contrast, Docherty et al. (1997: 306), from

a variationist perspective, observe that “present phonological accounts do not always

accurately predict patterns of surface variation.” Docherty et al.’s (1997) findings,

indeed, exhibit the use of the glottal stop and [ɹ] in monosyllabic verbs across word

boundaries, even though glottalisation was predicted not to occur in such a context 9.

Moreover, they point out that forms attested during the fieldwork are “frequently forms

that are not predicted by theorists” (p. 288) and disconfirm Carr’s claim that

glottalisation and ‘weakening’ are in complementary distribution, providing evidence

from data systematically collected.

It is clear that a union between phonology and sociolinguistics is yet unachieved and

that the divergent stances taken by these two disciplines on (t) glottalisation are

considered as an issue in terms of “descriptive adequacy” (p. 288). Hence, as Docherty

et al. (1997: 275) note,

“On the one hand, socially situated language samples which have been
systematically collected and analysed constitute a legitimate –indeed often
vital – source of evidence to be utilised by linguists for assessing and
refining theoretical models. On the other hand, variationists cannot operate
in isolation from theoretical concerns, and can benefit from an evaluation
of the competing theoretical frameworks available to them.”

9
Results from this analysis are based on Hartley’s (1992) recordings.

158
Variationist approaches usually diverge from theory-led approaches since initial

formulated theoretical assumptions are not conventional, and theoretical hypotheses

often emerge during systematic analysis of a considerable corpus of data. Eckert

(1991), for instance, collected real data associated with specific social categories before

developing a chain-shift model.

Having provided an overview of how (t) glottalsation is treated in phonological theory

and having briefly explored some dissimilarities between theoretical phonology and

sociolinguistics in respect to the glottal variant, it is now necessary to provide a deeper

insight into sociolinguistic accounts for (t) glottalisation.

5.3 T-Glottalisation in Variationist Sociolinguistic Theory

As mentioned above, sociolinguistic theory, unlike structuralist-generative models, is

constructed on the bases of external evidence (Docherty et al., 1997) and explores the

correlation between specific linguistic features and social factors. The social factors

found to significantly constrain /t/ glottaling are listed as follows: style (e.g. Holmes,

1997), age (e.g. William and Kerswill, 1999), gender (e.g. Milroy et al. 1994) and

social class (e.g. Trudgill , 1974); this explains why glottal variation in /t/ is a variable

rule as it is not independent of social considerations (Wells, 1982). In sociolinguistic

theory, intervocalic /t/ glottaling, as in letter, is traditionally regarded as a marker of

working-class British speech (Trudgill, 1974, 1988; Macauley 1977) as it has diffused

in all social classes and styles, yet still exhibits social and stylistic variation. This

definition springs from a three-fold distinction: (a) indicators, (b) markers and (c)

sociolinguistic stereotype, made on the basis of the interaction between sociolinguistic

variables and the social stratification in a given speech community (Trudgill, 1986).

Indicators refer to stratified linguistic features occurring below the level of social

159
awareness, with no significant difference in the degree of formality, and to which little

or no social import is attached (Labov, 1972). Markers, instead, are variables associated

with a low level of awareness, but show consistent style effects. Linguistic variables

which become popular features of a particular group and which can be overtly

commented on are usually referred to as sociolinguistic stereotypes (Trudgill, 1986).

Formulaically, the status of /t/ can be illustrated as (after Trudgill, 1974: 174):

/t/ → x < [t] ~ [tɁ] ~ [Ɂ] >

x = f (style, class, age)

In other words, /t‘/, /tʰ/ and /t/ are deemed “extrinsic allophones which belong at the

systematic phonetic level”, whilst [tɁ] and [Ɂ] “are not extrinsic allophones, but variant

realisations, depending on sociological variables, of the extrinsic allophones /t‘/and /t/

at the phonetic realisation level”, which are usually found in casual speech of young

urban working-class speakers (Trudgill, 1974: 157). In the literature, /t/ has often been

examined separately in word-final or word-medial position with particular attention to

the phonotactic environment. So far, the bulk of research has predominantly focused

on the environment following /t/, and three contexts are usually compared:

preconsonantal /t/ (PreC), prevocalic /t/ (PreV) and prepausal /t/ (PreP).

In many British dialects results tend to be constant, with PreC deemed the most

influential linguistic factor that favours glottal variants. Straw and Patrick (2007),

indeed, refer to the following traditional ranking PreC > PreV > PreP as the diffusion

pattern10. Hughes et al. (2012) summarise the common findings including syllabic

nasal, syllabic lateral, and intervocalic /t/ in the hierarchy to follow11:

10
This term refers to both its geographical diffusion as well as its spread through the different linguistic
contexts.
11
Note that a following pause is not included in this hierarchy.

160
most frequent word-final pre-consonantal that man

before a syllabic nasal button

word-final pre-vocalic that apple

before syllabic /l/ bottle

least frequent word-internal pre-vocalic better

Table 5.3.1 Most frequent and least frequent environments for t-glottaling to occur.

Recent studies show that [Ɂ] seems to have spread up to the upper class (including

young members of the royal family) and it is adopted by many RP speakers word-

initially before vowels (ant), pre-consonantally, pre-nasally (Hughes et al., 2012) as

well as in pre-vocalic and pre-pausal contexts (Barrera, 2015). These findings suggest

a language change in progress given the results previously reported by Fabricius

(2000), and later confirmed by Altendorf (2003), displaying no pre-pausal and pre-

vocalic t-glottaling in the more formal styles among upper middle-class speakers in

London; whilst, t-glottaling in intervocalic position is still firmly stigmatised in RP.

Barrera (2015: 13) also suggests that, in RP, “lexical frequency seems to be playing a

role in the different progressing stages of the glottal stop word-internally and across

word boundaries.”

It is also important to outline that Fabricius’ (2000: 134) RP speakers exhibit a

regional variation:

“Pre-consonantal glottaling can reasonably be regarded as the ‘first wave’

of glottaling. The ‘second wave’ seems to be pre-pausal category, which
shows a significant difference between the Southeastern category and the
‘rest of England’ category. As we have seen, London and the Home
Counties pattern together on this feature, while the rest of England lags
behind. The ‘newest’ wave of glottaling is evident in the pre-vocalic
category, where the London-raised public school speakers use pre-vocalic

161
 glottaling at a significantly higher rate than speakers from other parts of
England in less formal styles of speech.”

According to the Survey of English Dialects t-glottaling has been traditionally found

in Norfolk, London and the Home Counties (Orton and Tilling, 1969), whilst

glottalised variants have been recognised in East Anglia (Norfolk, Suffolk and Essex)

and East Cambridgeshire. At this point, before exploring in more detail how the glottal

variants behave in southern British dialects and northern British dialects, we should

turn our attention to its diachronic development in order to understand where it has

originated and how it has geographically spread.

5.4 Diachronic development of [Ɂ]

Geolinguistics reveals that the distribution of glottal stops is extremely marked in the

South of England and that in both intervocalic and word-final environment [ʔ] is still

geographically spreading (Straw & Patrick 2007). Yet, where has it started spreading

first? Despite lack of consensus amongst linguists as to where the glottal feature started

and even though no research has shed light on its origin so far, three locations can be

identified as plausible places where it might have developed first: East Anglia, London,

Glasgow.

Trudgill (1974) claims that East Anglia - Norwich in particular - seems to be

“the centre where glottaling has diffused geographically” as its geographical position

might have facilitated the glottal feature to spread. After consulting the Survey of

English Dialects (SED), Trudgill (1974) demonstrated that in the 1950s intervocalic t-

glottaling was only present in one rural southern area of England: northern East Anglia.

If this plausible explanation is accurate, it implies that the glottal stop spread from East

Anglia to London. The Linguistic Atlas of England (LAE), indeed, shows evidence of

162
t-glottaling only in East Anglia and a small area around London, as in figure 5.4.1

(Orton et al., 1978).

London, however, is usually cited as the principal geographical source for the

spread of t-glottaling and it is generally associated with Cockney English - the working

class accent of East End Londoners. Some linguists (Matthews 1972, Wells 1994,

Williams and Kerswill 1999, Altendorf 2004) believe that t-glottaling is actually not

the only phonological innovation that spread in southern England and the rest of the

UK from London, but also other variables such as TH-fronting and L-vocalisation12.

These new variables, according to Altendorf (2004), share some characteristics since

they all stem from non-standard accents, they are all associated with London English

and, in the south-east, it appears that they have been diffusing regionally and socially,

spreading up the social scale and into formal styles. Przedlacka (2001: 48) evinces that

London is a potential and powerful source for linguistic innovations due to its political,

economical and cultural influence “from which innovations normally radiate

outwards”.

12
Johnson and Britain (2007) suggest that the origin of /l/ vocalisation is not rooted in the south-east of
England (or at least in East Anglia) as the phoneme /l/ was reported to be clear in all environments until
the 20th century. Ihalainen (1994) found evidence of /l/-vocalisation in Yorkshire between the 17th and
19th century, Orton (1933) reports its presence in South Durham and Wright (1905) observed it in the
North of England. For further details on the diachronic development of /l/ vocalisation see Johnson and
Britain (2007).

163
Figure 5.4.1 Evidence of glottaling in the Linguistic Atlas of England. Adapted from
Orton et al. (1978).

Wells (1982) suggests that “either this is a new, twentieth-century, phenomenon, or no

phonetician had previously noticed it”. In the survey carried out by Ellis (1989), entitled

On Early English Pronunciation, he provides evidence from several observers from

among Walker (1072- 1807) and Smart (1836), who discussed Cockney, yet none of

them seems to have noticed the presence of glottal stops in earlier centuries. Only Ellis

(1889) reports the loss of final and medial /t/.

The second origin, suggested in the literature, is Scotland. Wright (1905: 287)

found evidence of the glottal stop in “west-mid Scotland, Lothian and Edinburgh”, and

only before schwa [ə] plus liquid consonants /l/ or /r/ in the next syllable (e.g. battle,

better). However, as reported in Wells (1982), in the first edition of Pronunciation of

English by Jones (1909), it is argued that [ʔ] was frequently used to replace /t/ both in

Scotland and London. Andrésen (1968) confirms the use of (t) glottalisation in Scotland

164
claiming that it gradually diffused in the West of Scotland (attested in 1860), in the

East part of Scotland (attested in 1889), in the North of England after twenty years, and

finally it reached the Midlands and London (1909)13. Upon consultation of previous

reference on glottalisation, between 1900 and 1930, Collins and Mees (1996) write that

a general acceptance of [ʔ] started after the Second World War, and it was introduced

even in phonetic transcription such as O’Connor (1952) and in description of RP

(O’Connor 1952, Christophersen 1952). Christophersen (1952) examines the

morphological environments where glottalisation of /p, t, k, ʧ/ can occur, but, as

Holmes (1995) observes, he only discusses glottal reinforcement. Pre-glottalisation,

indeed, is thought to be associated with the north rather than the south of England

(Wells, 1982).

Later, Macafee (1997) adds that the glottal stop has precisely emerged from

Glasgow and rapidly spread to Scotland and throughout the UK. Stuart-Smith’s (1999)

review on Glasgow studies, shows that the glottal feature has been used, since the 19th

century, in both word-medial and word-final contexts, whilst a following vowel was

less likely to foster a glottal insertion. Yet, Romaine’s (1975) findings, based on data

collected in Edinburgh, propose a different ranking resulting in PreV > PreP, where a

following vowel triggers glottalisation more than a following pause14.

However, the lack of earlier descriptive evidence is not a reason to assume that the

glottal stop is a recent sound or emerged in the 19th century. As a result, Abercrombie

(1948) mentioned some studies going back to the 17th century where the glottal stop

was found at the onset of initial vowels. Cruttenden (1994) argues that it is improbable

that t-glottaling is older than 200 years in London. I believe that even if London is a

13
The earliest of evidence of glottal reinforcement for /p/ and /k/ appears not to be mentioned before
1909 (Andrésen, 1968).
14
PreC might have been excluded because it was probably categorical.

165
great potential source for linguistic innovations, there seems not to be strong evidence

to claim that it is the first geographical place where the glottal stop has originated. Its

more recent spread throughout the country, however, is likely due to influence from

London. The LAE, as aforementioned, shows that the glottal variant was present in a

small area of London (Orton et al., 1978), and Ward (1929), cited in O’Connor (1952),

observed that this usage did not occur among London speakers. Moreover, recent

studies propose that “the use of the (t) variable by Ipswich Anglo urban speakers does

not suggest diffusion from London” (Straw & Patrick, 2007: 404), and Schleef (2013)

reveals that t-glottaling is more developed in Scotland (particularly in Edinburgh) than

London. Hence, it seems that the two plausible geographical sources - or “dual

epicentres” as called by Kerswill and Williams (1997) - where the glottal variant might

have originated are East Anglia and Scotland.

5.5 T-Glottalisation in Southern England

Considering its geographical diffusion, it is evident that glottal variation in /t/ is

recurrent in both the north and south of England. Some studies, as discussed above,

have shown that non-standard varieties seem to undergo more recurrent changes,

mostly occurring in the south-east of England. T-glottaling and t-glottalisation, indeed,

have been investigated in London and the Home Counties (e.g. Tollfree 1999; Schleef

2013), Reading (Williams & Kerswill, 1999), Milton Keynes (Williams & Kerswill,

1999), Norwich (Trudgill, 1974, 1988, 1999, 2003), Ipswich (Straw & Patrick, 2007),

and Essex (Altendorf, 2003). Despite the extensive literature, this section places under

scrutiny studies carried out in East Anglian counties (Norfolk, Suffolk and north Essex)

from 1889 until 2007; secondly it focuses on London vernacular.

166
5.5.1 East Anglia (1889 - 2007)

Early studies carried out by Ellis (1889), Kökeritz (1932) and the SED (1962) have

considered phonological and syntactical aspects of East Anglian English, yet their work

lacks a variationist perspective15. Ellis (1889), as mentioned earlier, only commented

on the loss of final and medial /t/, whilst Kökeritz (1932), who provided a descriptive

phonological account of the Suffolk dialect of the 20 th century, noted that despite the

wider discussion on English dialects the Suffolk accent had not received full attention.

Earlier work (Binzel, 1912), for instance, focused on the interpretation of spelling

rather than real speech data. Hence, Kökeritz’s (1932) work seems to be the first study

to investigate the phonology of the Suffolk dialect, although the methodology adopted

is not appropriate by modern standards

Claxton (1960: 5) in the second edition of The Suffolk Dialect of the 20th Century, states

that, in Suffolk, “the consonant ‘t’ is usually articulated and not ‘swallowed’ as in the

Norfolk where ‘water’ is pronounced as ‘wa-er’, ‘butter’ as ‘bu-er’, ‘city’ as ‘ci-y’ ”.

In 1962, evidence of the glottal stop in East Anglia, was provided by The Survey of

English Dialects (SED). East Anglia, as mentioned earlier, was identified as the area

where two types of glottal variation occur: glottalised and glottalled variants. The

former was widely found in Norfolk, Essex and Suffolk (mentioned here in a

decreasing order of diffusion assuming an origin in Norfolk), whilst the latter was

predominately used in Norfolk and partly in Essex, as shown in figure 5.5.1.

Later, Orton et al. (1978) in The Linguistic Atlas of English (LAE) reported the use of

[Ɂ] in Norfolk as well as a small area around London.

So far, it seems that little research relevant to language variation and change had been

conducted until Trudgill’s (1968) study. His work on the co-variation of phonological

15
See section 2.5 for further details related to early field methods.

167
features with sociological parameters in Norwich, follows soon after the earliest

community-based research in sociolinguistics by Labov (1963). For his study, Trudgill

(1974) adopted a quasi-random sample extracted from four ward voter registration lists,

where an equal number of participants was randomly selected16. The sample consisted

of 60 speakers17, stratified according to sex, age and social class. To assign each

speaker to a social group, he employed a six point-scale.

Results from the Norwich study show that glottal variants occur intervocalically and in

syllable-final context (i.e. better, bet), yet they are blocked in stressed syllable initial

position (i.e. tea) or in /n_/ and /l_/contexts (i.e. went, felt). A further restriction was

suggested before and after schwa /ə/ or unstressed /ɪ/; however, Trudgill shows that in

the case of stressed /ɪ/, as in went into, the use of glottal stop is permitted. Glottal

variants were also found proportional to class, sex and social context since they

decrease as formality increases, confirming that women tend to use more of the

standard variant than men and that glottal variants were predominantly found in casual

speech of young urban working-class speakers. Trudgill (1988) also highlights that the

glottal stop is spreading into more formal styles and, unlike youngsters, old people tend

to avoid the stigmatised feature in certain situations, and that it may be an endogenous

change.

16
The four wards were: Eaton, Lakenham, Helledson and Westwick.
17
Ten informants were schoolchildren from two schools.

168
 Figure 5.5.1 Distribution of glottal variants. Adapted from Trudgill (1974).

The Norwich study, however, does not comment on phonological environments that

are significant in terms of diffusion. Further details on this point were subsequently

provided by Straw and Patrick (2007) in the Ipswich study.

Despite the extensive discussion on East Anglian dialects provided in the late 20 th

century by Trudgill (1974, 1986), Fisher (2001) and Britain (2002), Ipswich appears to

be sporadically mentioned and no study appears to have reported the ranking of

phonological environments favouring glottal variation in /t/ in this community until

2007 when Straw & Patrick’s study was carried out. The purpose of their preliminary

research, which examines glottal variation of /t/ in word-final environment, is (a) to

compare the dialect of four Anglo speakers in Ipswich to that of four Barbados-born

speakers in the same community, considering whether Ipswich Anglo urban speakers

indicate diffusion; (b) to explore whether the patterning among Barbadian speakers

169
mirrors dialect acquisition; (c) in response to Docherty and Foulkes’ call (1999), they

investigate whether the ‘received wisdom’ of glottal variants is methodologically

secure.

Data were collected from participants who live in the same working class

neighbourhood, and the participants were stratified according to sex and two age-

groups (elder speakers 68-74 and middle-aged 40-55). A total number of 250 tokens18

was analysed by means of acoustic analysis employing five parameters “to reduce the

perceived glottal variation: (1) presence or absence of glottal occlusion; (2) duration of

the gap in milliseconds; (3) presence or absence of laryngealisation; (4) location of

laryngealisation and (5) presence or absence or voicing irregularity.

Results from auditory analysis were consistent with findings from previous British

dialects19 (resulting in the diffusion pattern, as previously discussed), but results from

acoustic analysis turned out to change the picture: glottal replacement was found not to

occur as much as predicted by auditory analysis and a different ranking triggering

glottal variation in /t/ was exhibited. Hence, this peculiar pattern amongst Ipswich

Anglos, called the Ipswich pattern (after Straw and Patrick, 2007: 393), suggests that

the PreV context favours glottal variants the most: PreV > PreC > PreP. This study

concludes that “the use of the (t) variable by Ipswich Anglo urban speakers does not

suggest diffusion from the London area.” (p. 404). However, as the authors note, some

questions remain open: how similar Ipswich and Norwich are; whether the Ipswich

pattern is retained and whether there are signs of diffusion from Norwich.

Not only was Ipswich only sporadically mentioned in the discussion of East

Anglian dialects, but also Colchester, where little systematic research has been

conducted so far. Together with London and Canterbury, Colchester was chosen by

18
The average per speaker equals 31 tokens.
19
Note the PreC, PreV and PreP environments were not previously employed in East Anglia.

170
Altendorf (2003) to investigate Estuary English20. She explored five vocalic variables

and six consonantal variables (t-glottaling included) in 10 female informants, recruited

in three types of schools: comprehensive school (two working class females), grammar

school (six middle class females) and private school (two upper class females). The

three cities were not equally stratified in terms of social class, indeed, in Colchester,

only the middle class was explored. Moreover, if the six middle class females were

proportionally distributed across the three cities, two informants only would be

representative of Colchester.

Results for (t) glottaling exhibit the following ranking: PreC > PreV > PreP >

pre-lateral /l/ > intervocalic position, where the glottal variant is favoured before

consonants more than in intervocalic position. T-glottaling was also found in all social

classes as well as in more formal styles, even though it displays both social and stylistic

variation. Middle class speakers, for instance, would reduce the use of the glottal

variant in PreV and PreP, whilst it was completely avoided intervocalically where the

stigma seems to persist, even in more formal styles.

The upper class informants’ speech diverges from the middle and working class

especially in PreV and PreP contexts where the use of glottal stop is highly reduced

and it is almost completely absent in the formal reading style. Intervocalically and in

pre-syllabic /l/ environments, however, it is still stigmatised. However, an interesting

result was found within members of the working class where (t) glottaling in pre-

syllabic /l/ was used with the same frequency as word-medial pre-consonantal (e.g.

Gatwick). Altendorf (2003) states that (t) glottaling is increasing in PreV and PreP

environments among middle class speakers, in spontaneous speech, with a rate nearly

five times higher than Hudson and Holloway’s (1977) study, as we will see later in the

20
Estuary English was defined as “a variety of modified regional speech, a mixture of non-regional and
local south-eastern English pronunciation and intonation” (Rosewarne, 1984).

171
chapter. As opposed to the limited research in Colchester (Altendorf, 2003), Ipswich

(Straw and Patrick, 2007) and Norwich (Trudgill 1974, 1988, 1999, 2003), (t)

glottalisation has received full attention in London being stereotypically associated

with Cockney, and due to the position of London in England (as political capital) which

is usually deemed a “linguistic centre of gravity” (Wells, 1982: 301).

5.5.2 London vernacular (1938 - 2013)

As noted during the historical development of /t/, Ward, in 1929, commented on the

use of glottal variants that, at that time, were not present in the speech of Londoners.

Early evidence of [Ɂ], in Cockney, was reported by Matthews (1938: 80) who claimed

that: “the chief consonantal feature of the dialect is the prevalence of the glottal stop.”

Sivertsen (1960: 199) claims that, in Bethnal Green, “the alveolar stop, at least when

it is strongly affricated in the environment [V_V], is looked upon as being too ‘posh’

for a Cockney to use: [‘'beʦə](= better) is ‘posh’, ['betə] is normal, and ['beɁə] is

‘rough.’ She found [Ɂ] intervocalically in tokens such as getting with a syllabic /n/ as

well as word-finally before vowels (e.g. right in). However, intervocalically across

boundaries, EE were more likely to adopt the alveolar approximant (T-to-R rule),

although this feature was not typically described for London English.21 In addition, the

glottal variant was adopted more often by men than women, as confirmed later by other

studies.

Beaken (1971), cited in Wells (1982), investigated the speech of schoolchildren

in the East End of London, reporting the use of glottalised /p t k/ in word-final position.

21
Hickey (2005) reports that the T-to-R rule is common in Northern England and in vernacular Dublin
English.

172
In more detail, the proportion of glottal variants in this context was 73:4 in males’

speech and 46:14 in girls’ speech.

Hudson & Holloway (1977) reported that t-glottaling was more frequent in

PreC than PreV contexts, with a larger percentage of glottaling in PreV for middle-

class males, working-class females and working class males, whilst middle-class

females exhibit 20 percent of prevocalic glottaling.

Tollfree (1999), in her study conducted in different regions of South East

London22, observed that both age groups (15-30 and 54-89) of South East London

Regional Standard speakers (SELRS) produced glottalised variants in PreC context

across a word boundary (e.g. ticket box), as well as in word-internal environment (e.g.

nightmare). In preceding syllabic /n/ (e.g. button) old SELRS did use the glottal feature,

yet it was stigmatised in preceding syllabic /m/, preceding lateral /l/ (e.g. bottle) and

intervocalic contexts. In PreV cross-word boundary as well as PreP position, (t)

glottalisation was sporadically used. In South East London English speakers (SELE)

glottal variants were near-categorical in PreC and PreP and, unlike SELRS, t-

glottalisation was frequently adopted in PreV context and intervocalically. She also

reported that “t-glottalisation […] operates freely between the stem and the

compounds” (p. 171) (e.g. a put-on), yet it is blocked when the segment which precedes

/t/ is a non-resonant consonant (e.g. project, sister, chapter) and in word-internal foot-

initial onset position (e.g. particular). By contrast, in word-internal non-foot initial

onset context (e.g. printer, botany, Saturday) it was often attested. Her results also

reveal that t-glottalisation seems to be “highly sensitive to prominence patterns” (p.

172) and is more likely to operate in tokens such as litter and butler, where the

22
The data were gathered in Peckham, Sydenham and Penge (working class), as well as Dulwich,
Beckenham and Bromley (middle class), were a total number of 90 informants was interviewed. Her
participants were divided into two groups: SELE (South East London English) and SELRS (South East
London Regional Standard).

173
prominence is [1 2] and [1 3 2] respectively, but the process is blocked in items like

pretend [2 1] and contemptible [2 1 3 4].

Schleef (2013) examined t-glottaling in London and Edinburgh among teenagers

stratified by sex and age. His results show that the variable is phonologically

controlled, but most importantly, it intersects with morphological categories and shows

evidence of lexical diffusion. Word-final /t/, in both London and Edinburgh, exhibits

the typical pattern PreC > PreP > PreV, with following nasal, liquid and fricative

favouring (t) glottaling, whereas following plosive, glide, pause and vowel disfavoured

it in both locations. As for the preceding environment, preceding nasals and liquids

disfavoured glottal realisation of /t/, whilst preceding vowels 23 favour it. Glottal

replacement was more frequent in informal style and occurred more often in word-final

position than in word-medial context, yet no gender difference was identified. It seems

that this finding is not very surprising as it suggests that when changes approach an

endpoint, the gender difference tends to disappear (Labov, 2010). From a grammatical

standpoint, function words, progressive verbs as well as past participle were found to

favour t-glottaling; on the contrary, adjectives and nouns disfavoured it. He also found

that the number of syllables plays a role in Edinburgh, as words of four and five

syllables seem to trigger t-glottaling more than words of three syllables. Finally, the

proportion of application value reveals that t-glottaling is more developed in Edinburgh

than London, hence he claims that:

“this makes speculations on the longer history of T-glottalling in

Edinburgh, when compared with London, very plausible. Furthermore,
considering the high degree of similarity of the factors constraining
variation in /t/, a diffusion scenario seems more likely than independent
development.” (p. 212).

23
It has to be noted that both preceding and following vowel, were not coded according to their place
of articulation.

174
In contrast to the long history of t-glottaling in Edinburgh, this phenomenon is recently

being acquired as a prestigious feature in some southern English varieties, such as

Cardiff English.

5.6 T-Glottaling in Cardiff

Mees & Collins’s (1999) study explored the spread of the glottal variant in Cardiff

English. Word-final t-glottaling was analysed in PreC, PreV and PreP environments,

along with social class24 (working class, lower middle class and middle middle class),

gender, time (1976, 1981 and 1990) and style (reading passage and casual style).

Results for t-glottaling in word-final context exhibit a prestigious spreading with young

middle class females leading the change. By contrast, the glottaling rate in the same

linguistic context was found to be very low among working class speakers. In this

respect, Mees & Collins (1999: 202) claim that:

“the upwardly mobile Cardiff females can be seen to acquire RP-style

glottalisations together with a professional career, a suburban house and a
well-qualified partner. Those who lack such aims are also likely to lack
glottalised forms.”

They also suggest that London’s lifestyle along with public figures and celebrities,

who are often heard using the glottal stop and considered trendy, have an influence on

the way people speak. Hence, t-glottaling is perceived as a prestigious and trendy

feature.

After reviewing, selectively, the literature on t-glottalisation in southern British

dialects, starting from East Anglia – the focal point of this study – to London and

Cardiff, let us now move north above the FOOT-STRUT split.

24
The class stratification of speakers was based on father’s occupation.

175
5.7 T-Glottaling in Northern England

The FOOT-STRUT split, according to the Dialects of English (Trudgill, 1999), is one

of the foremost isoglosses (along with the BATH broadening) marking a dialect

boundary between the north and the south of England25. Like southern dialects, glottal

variants were reported for the West Midlands (Mathisen, 1999), Derby (Docherty &

Foulkes, 1999), Hull (Williams & Kerswill, 1999), Newcastle (Milroy et al., 1994;

Watt & Milroy, 1999), Sunderland (Burbano-Elizondo, 2015), Middlesborough

(Llamas 2001, 2006), Manchester (Baranowski & Turton, 2015) and the Fenland

(Britain, 2015).

The Fenland represents the major dialect transition zone where northern

varieties meet southern ones (Britain, 2014). In this site, (t) glottaling is extensively

adopted, even intervocalically. Moreover, evidence of glottal replacement was also

found in word initial context, if /t/ occurs in an unstressed syllable, as in tomorrow and

today (Britain, 2014).

In the North-East, /t/ is usually studied alongside /p/ and /k/ as they all can be

articulated as stops, glottalled and glottalised variants. Glottalisation of word-medial

/p/ and /k/, indeed, have been frequently reported in the literature as a distinctive feature

of Newcastle and Tyneside English (e.g. Milroy et al. 1994; Docherty et al. 1997; Watt

& Allan, 2003). It was also attested in Durham (Kerswill, 1987) and Sunderland

(Burbano-Elizondo, 2008). However, in spite of being examined together with the other

two voiceless plosives /p/ and /k/, this section will be concerned with glottal variation

in /t/ only.

25
The linguistic north delimited by these isoglosses comprises “from the Scottish border as far as a line
from the Mersey to the Humber, including most of the Midlands. It includes, for example, the
Birmingham-Wolverhampton conurbation, Leincester and Peterbourough” (Wells, 1982: 349).

176
 In Newcastle, Milroy et al. (1994) examined five variants of (t) in the informal

speech style of 32 informants, stratified by social class (16 working class, 16 middle

class), gender and two age groups (16-24; 45-65). A total number of 2,838 tokens were

analysed in word-final environment, and results show that young middle class females

were leading the use of glottalled variants, whilst working class speakers as well as old

speakers fall behind. Hence, glottalled features were associated with middle class

members, whereas glottalised variants were more frequent among working class

speakers. Overall, it seems that females are leading the diffusion of t-glottaling in

Newcastle, and that it is a supralocal change.

Similarly, Burbano-Elizondo (2015) found that, in Sunderland, middle-class

females are leading the change for (t) glottaling, whilst glottal reinforcement [Ɂt]

exhibits a higher percentage in males (38%) than females (10%). She reported that

glottalled features are used 59.8% of the time, with a significantly increasing rate over

time (i.e. among the young) – similarly to Middlesbrough (Llamas, 2001). Comparing

results from Sunderland (Burbano-Elizondo, 2015) with findings from Newcastle

(Docherty, 2007) and Middlesbrough (Llamas, 2001), it was observed that the use of

[Ɂt] decreases as the distance from the North Eastern ‘capital’ city increases, reporting

an increase of [Ɂ]. In terms of geographical diffusion patterns, the increase of [Ɂ] for /t/

in Sunderland is less advanced than in Middlesbrough, yet more than in Newcastle,

contrary to any expectations which suggested that [Ɂ] would spread to larger cities first

and subsequently to smaller localities. Docherty & Foulkes (2005) investigated /t/

glottalisation in Tyneside, where the use of the glottal stop is well-established in word-

final context pre-consonantally, yet it is sporadically adopted intervocalically or pre-

pausally.

177
 Jeffries (2011) examined the acceptability of (t) glottaling in the head of a

primary and secondary stressed foot, as in thirteen and retain. The dataset consists of

acceptability scores from West Yorkshire informants who rated (t) glottaling based on

audio clips. Her results show that the transition /t/ → [Ɂ] occurs when /t/ is not in the

head of a foot, either before an unstressed vowel (e.g. butter) or word-finally (e.g.

what). Even though glottaling is prohibited by foot-based accounts and syllable-based

accounts, she suggests that glottal variants can be found before long, high vowel /i:/ or

/u:/.

Broadbent (2008) shows that as (t) glottaling was advancing in Yorkshire t-to-

r, which was the dominant lenition process, became weaker.

Baranowski & Turton (2015) explored three consonantal variables, two of

which are undergoing a change (T-glottaling as well as TH-fronting) and a stable

variable (H-dropping) in Manchester26. T-glottaling was explored particularly in word-

final postvocalic /t/ (e.g. cat) and intervocalically (e.g. butter). Auditory analysis of 86

speakers, stratified by sex, age and socio-economic class27 (working class and middle

class) was conducted, and a total number of 3,727 and 2,043 tokens, for word-final

postvocalic /t/ and intervocalic /t/ respectively, was analysed.

Results for word-final /t/ find age and the following segment to be the only significant

predictors, with glottal replacement less likely to occur among older speakers 28. Since

gender and social class were not marked as significant predictors, this pattern suggests

that word-final t-glottaling should be considered as an advanced change nearing

26
It has to be noted that, unlike the many North-East cities, Manchester displays few pre-glottalised
variants.
27
Age was used as a continuous variable, whilst socio-economic class was based on occupation since,
according to Baranowski and Turton (2015), other measures for social status were employed, such as
education, yet occupation has usually produced the best model.
28
This finding seems to be in contrast with Davies and Braber’s (2011) study, which reports a frequent
use in the older generation in the East Midlands.

178
completion. Even in Manchester, similarly to other studies reviewed above, young

middle class females displayed the highest rate of glottal variants, yet, according to

Baranowski & Turton (2015), this may be due to age-grading. In intervocalic context,

middle-aged working class males seem to have a remarkable tendency towards the

glottal articulation, with middle-aged middle class females falling behind. This finding

suggest that the change in the intervocalic phonological environment began in the WC

social group. Both glottalisation and glottaling were attested in the so-called -ee/oo

words (e.g. canteen, tattoo), although glottalled features in this context were rare

among Mancunian speakers. Finally, as expected, T-glottaling is sensitive to style-

shifting even in Manchester.

5.8 T-Glottaling in Scotland

Moving further north, we find a second plausible epicentre of glottal(ised) variants that

has received sociolinguistic commentary: Glasgow. Stuart-Smith et al. (2007), for

instance, explored t-glottaling, among eight other consonantal variables, in 32 speakers,

stratified by sex, age (adolescents and adults) and social class (working class and

middle class). The data were collected by means of sociolinguistic interview as well as

a word list with a number of tokens equal to 3,597.

Results, if compared to Macaulay’s (1977) study, indicate a real-time increase of

glottals in Glasgow, with a higher rate for the working class (92.47%) than for the

middle class (56.56%) in informal speech. In the more formal style there is a clear

difference between the two classes: the former displayed a 76.32 per cent of t-

glottaling, whereas in the latter the glottal variant seems to be inhibited given the very

low rate of production (4.65%). Even in Glasgow a lack of gender difference was

179
reported. This study does not comment on the different phonetic contexts constraining

glottal replacement.

No gender effect was also found by Stuart-Smith (1999) who reports mandatory

use of the glottal variant in PreP and a recurrent usage in PreV as well as intervocalic

positions in working class speech.

Recent research on glottal replacement has been carried out in Buckie by Smith

and Holmes-Elliot (2017), with a sample of 24 speakers and a total number of 4,898

tokens coded for linguistic and social factors. The dependent variable was linguistically

constrained by following phonetic segment (PreP and PreV) 29, ambi-syllabic segment

(e.g. bottle, bitten) and onset position (e.g. sometimes), whist the independent variables

explored were age and gender. Results display the following ranking:

Ambi#Syllabic consonant (e.g. bottle) > Coda#Vowel (e.g. that is) > Ambi#Vowel30

(e.g. pretty) Coda#Pause (e.g. I like that.) > onset (e.g. sometimes).

What is surprising from this data is that the usual highly disfavouring ambisyllabic

context, in Buckie, is the most favouring factor. This study displays a dramatic change

in apparent time, from a minor percentage of [Ɂ] in the older generation to a full 90 per

cent among youngsters. A remarkable gender difference is only found among the older

generation where males adopt the non-standard feature more frequently than women,

yet this polarity appears to be levelled in the young generations. Finally, although

glottal variants were usually blocked in prominent syllables, Smith and Holmes-Elliott

(2017) highlight some exceptions, such as the teen numerals (e.g. eighteen) and the

lexical item sometimes.

29
The PreC contexts were excluded as /t/ might assimilate to a following non-sonorant consonant.
30
V_V environment.

180
5.9 T-Glottaling beyond the British Isles.

This phonological feature has also been investigated in other English varieties, such as

New Zealand English and American English.

Holmes (1999) examined word-medial and word-final /t/ glottaling in the speech of

young (18-30) and middle-aged (40-55) speakers from New Zealand, reporting an

increase of (t) glottaling lead by young informants. Glottal variants were found to be

more frequent among working class speakers than their middle class counterparts, with

women being in the lead in both classes. While glottal variants occurred at a higher rate

word-finally, evidence of glottaling was also found in intervocalic position (both word

medial and word final) even though not very frequently. Moreover, the pre-pausal

environment turned out to be the most favoured context for /t/ glottaling to occur.

Roberts (2006) conducted the first study on /t/ glottaling in Vermont analysing

47 speakers, aged between 8 and 80. The change is led by males, particularly by 9 year-

olds and teenagers. In terms of constraints, besides the following environment, the

preceding context was also included in the analysis revealing that preceding vowels

trigger glottaling, whereas preceding consonants inhibit it. As regards the following

environment, pre-pause, pre-nasal and pre-glide favoured the glottal stop, whilst

obstruents, vowels and liquids were marked as disfavouring factors.

To examine the variability of (t), Eddington and Taylor’s (2009) study mainly

focused on formal style, excluding informal speech and word-medial tokens. Their

work included 20 collocations of word-final pre-vocalic /t/ constrained by frequency,

preceding and following stress, preceding and following vowels, gender, age and,

location. The 58 participants gathered were divided on the basis of geographical

location: 22 speakers were from Western states, 20 were from Utah, and 16 from non-

western states. Results show that only following vowel, age and region surfaced as

181
significant factors. With respect to following vowels, following high vowels had a

significant effect, whereas following back vowels inhibited the use of glottal variants.

As for age, young and middle-aged females (in their 20s, 30s and 40s) adopted

glottaling more than males aged 20-30. As regards location, high frequency of glottal

stops were found among Western speakers, rather than among Non-Western ones.

Eddington & Brown (2020) examined the production and perception of word-

final and word-medial /t/ glottalization in five US states: Indiana, Mississippi, New

Mexico, Utah, Vermont. The production study shows that age is the only significant

social factor when /t/ occurs word-medially. /t/ in pre-nasal word-medial (e.g. button)

is more likely to be produced as oral releases by young speakers than by their older

counterparts. The preceding vowels of word-medial /t/ followed by nasals was found

to condition glottalization, with /ɪ/ and /ə/ favouring oral releases and /æ/ and /o/

disfavouring it. As for word-final /t/, age and gender are marked as significant

predictors, with young speakers and women being more likely to realise /t/ as a glottal

stop in the PreV environment. The perception study revealed that people who use

glottal variants are considered less educated and less friendly.

5.10 Summary

At this point, the above literature review leads us to make some considerations. Firstly,

some studies treat t-glottaling and t-glottalisation together (e.g. Tollfree, 1999), whilst

others only focus on glottal replacement (e.g. Schleef 2013; Baranowski & Turton

2015; Smith & Holmes-Elliott 2017, etc.).

Secondly, as Schleef (2013: 205) claimed: “in some locations, T-glottaling is a new

feature; while in other locations it is not.” It is evident, in fact, that t-glottaling is in the

process of ongoing change in some parts of the UK, as revealed by generational

182
differences whereas, in other sites, it does not seem to be a new feature especially in

specific phonetic contexts (e.g. PreC environment). In the PreC environment, indeed,

the change seems to approach an endpoint due to high rate of occurrence and lack of

gender differences. On the contrary, a more recent change can be observed in the most

constraining and stigmatised environments, namely intervocalic position and in

prominent syllables, where the glottal variant has started spreading and, as expected, it

is rarely used by speakers belonging to the old generation. Moreover, since [Ɂ] is the

non-standard and stigmatised variant, it is not surprising to find males using it more

(Labov, 2001). However, Milroy et al.’s (1994) findings from Newcastle show a higher

rate of glottal variants for middle-class females than for middle-class males, and

suggest that this may be associated with a supralocal change, as also observed in Cardiff

(Mees & Collins, 1999).

Schleef (2013) proposes that the variability of /t/ was found not to be merely

conditioned by phonological factors, but also by grammar which plays a notable role.

The usage of glottal(ised) variants has also increased dramatically in all social classes

and styles. However, as Altendorf & Watt (2008: 209) claim: “social differentiation is

retained by differences in frequency and distribution of the glottal variant in different

phonetic contexts”. In the light of the above considerations, this study addresses the

following research questions, as already outlined in Chapter 1:

Descriptive aims:

1) How does (t) glottaling behave in East Anglia?

2) Is the variability of word-final /t/ glottaling conditioned by phonological factors

only?

183
 - Since East Anglia is considered one of the places where the glottal stop has

originated, the hypothesis is that the change, in this area, might be nearing

completion.

3) Is word-medial /t/ behind word-final /t/ glottaling in the change?

- The hypothesis is that word-medial /t/ glottaling might be behind word-final

/t/ in the change, as the former includes intervocalic /t/ – a typically

stigmatised phonological environment.

4) Do external (i.e. social) factors play a notable role?

- The hypothesis is that social factors play a remarkable role if the change is

not nearing the end.

5) Do Colchester, Ipswich and Norwich exhibit similar patterns?

- Given the geographic proximity, the use of this feature is not expected to

considerably diverge between the three localities.

Theoretical aims:

6) Is (t) glottaling governed by a universal sonority hierarchy in East Anglia?

184
Chapter 6 - (t) GLOTTALING: RESULTS AND
DISCUSSION

This Chapter presents and discusses the results of the second linguistic variable

investigated in this survey: (t) glottaling. Firstly, I will recap some coding procedures

related to the linguistic constraints; secondly, I will briefly go through the analytical

procedures adopted during the mixed-effects Rbrul regression analysis. I will then

present the overall findings (the three locations together) and afterwards the results

from each locality to observe whether the role of internal and external factors conform

to the overall East Anglian pattern. Since the (t) database is split into two parts: word-

final /t/ (WF) and word-medial /t/ (WM), the last two procedures will be repeated for

both analyses.

6.1 Analytical procedure – linguistic constraints

As outlined, the phonetic environment which has received the greatest attention in the

literature is the following phonetic segment – commonly divided into three main

contexts: PreC (e.g. that man), PreV (e.g. that apple) and PreP (e.g. what?).

This study follows Roberts (2006), Eddington and Taylor (2009)1, Schleef (2013), and

Barrera (2015) in coding for both preceding and following environments. Roberts

(2006) divided the consonant category into: obstruents, liquids, glides and nasals, and

has also coded for vowels and following pause. Schleef (2013) maintained the same

1
Both Roberts (2006) and Eddington and Taylor (2009) researched (t) glottaling in American English.

185
coding scheme for the following context but, in line with Barrera (2015), limited the

preceding environment to vowels2, nasals and liquids3.

The current study, for both preceding and following phonetic segment, codes vowels

as front, central and back. Trudgill (1974) shows that /t/-glottaling is blocked when

followed by a schwa /ə/ or unstressed /ɪ/, hence, to better control the behaviour of

preceding and following schwa /ə/ - provided the restriction of the rule application -

central /ʌ/ was coded together with back vowels. It is unlikely that the reclassification

of these few tokens would have a different statistical effect on these factor groups given

the number of occurrences. Following consonants are distinguished between nasals,

liquids, stops, fricatives and glides. The coding for the preceding environment is not

only limited to the factor groups examined in previous studies, but along with vowels,

nasals and /l/4, I also included preceding stops, sibilant fricatives and non-sibilant

fricatives. In the literature, preceding non-resonant consonants (e.g. project, sister,

chapter) were found to block /t/- glottaling (Tollfree, 1999), hence they are usually

excluded from analysis (Schleef, 2013). However, my hypothesis is that if there is a

“rapid increase in the use of [Ɂ] across all dialects studied to date in the UK” (Smith &

Holmes-Elliott, 2017) and, for this reason, (t) glottaling is advancing so fast across

geographical, social and linguistic constraints, it is likely that [Ɂ] might have just started

diffusing in those linguistic contexts where /t/ was categorically retained. After all,

there are records in which the glottal stop occurs after a preceding stop, i.e. post-

glottalization, which is generally considered a characteristic of broad accents, and

occasionally referred to as basilectal (Cruttenden, 1994). Examples of post-

glottalisation after /p, t, k/ are: keeper [kipɁə], speaker [spikɁə], water [watɁə], which

2
Barrera (2015) distinguished between front, central and back vowels.
3
Nasals and liquids were collapsed in Schleef’s analysis.
4
L- vocalised tokens are coded as vowels (e.g. bolt [bəʊt]).

186
typically occur in Tyneside (Wells, 1982). Even though the glottal stop does not occur

in word-final consonant clusters in the above examples, they show that the sequences

[pɁ], [kɁ] and [tɁ]5 can apply. Hence, /t/ after a preceding stop /p, k/ could potentially

have a glottal gesture. Docherty et al. (1997: 290) report the use of glottal(ised) variants

[Ɂt] in words like chapter, doctor, where the rhymal consonant is a stop, but claim that

“it is at present not clear whether the process also affects words like after and custard

where the preceding rhymal consonant is a fricative.” However, occasional glottal

replacement [Ɂ] of the labio-dental /f/ was found in Cockney (Cruttenden, 1994), and

occasional glottalization after fricatives (e.g. fifteen) was reported by Hartley (1992) in

the Tyneside adult conversational speech.

Consonant clusters excluded from analysis are str- (e.g. stroke) and /tr/ (e.g.

mattress). In the latter, /r/ can syllabify with the preceding /t/, hence the alveolar stop

is less likely to undergo glottaling6. Indeed, the few tokens containing /t/ in pre-liquid

/r/ position were all near-categorically non-glottal(ised). Since there was no variation,

they were excluded post-coding from the Rbrul analysis (Johnson, 2009). /t/ at the onset

of a stressed syllable (e.g. attack) are typically excluded as the stress pattern blocks (t)-

glottaling (Schleef, 2013). In this research, however, the –ee/-oo words such as tattoo

and seventeen were taken into account, in line with the Manchester (Baranowski &

Turton, 2015) and the Buckie studies (Smith & Holmes-Elliott, 2017) which show that

the glottaling process is advancing in stressed positions7.

5
This type of glottalized pronunciation [tɁ] is also commonly found in Suffolk and Norwich
(Trudgill, 1974).
6
In this environment, pre-glottalisation (e.g. mattress [mæʔtrəs]), seems to be more common among
English speakers (Wells, 1982).
7
/t/ → [ʔ] lenition after preceding high and long front (-ee /i:/) or back (-oo /u:/) vowels have also been
observed by Harris & Kaye (1990). As they highlight, these type of words (e.g.-teen items) are subject
to the English Rhythm Rule, (Prince, 1983), according to which the secondary-primary stress pattern of
a token (e.g. fourteen) turns into a primary-secondary stress pattern, if followed by a stressed word (e.g.
fóurtèen mén) as a result of primary stress clash. These findings were also reported for West Yorkshire
(Jeffries, 2011).

187
Word-medial /t/ was initially coded by syllable contexts, yet due to low number of

tokens in some categories they were grouped together (see section 6.5. for further

details). Tokens followed by the non-sonorant consonant /s/, such as it’s, were excluded

as [t] tends to assimilate to following segments. Similarly, occurrences with a following

/t/ and /d/ were not coded. Finally, a maximum of 10 occurrences per speaker were

coded for high-frequency words such as it, at, but, out, that, not.

6.2 Analytical procedure – Rbrul

Across the 36 informants interviewed, 3,051 and 1,872 tokens were respectively coded

for /t/ in WF and WM positions. The number of occurrences analysed for the WF

dataset averages 84.75 tokens per speaker, while 52 tokens per participant were

extracted for the WM contexts.8

Rbrul is the statistical tool employed to carry out the mixed-effects regression analysis,

with speaker and word as random effect. To spot the most relevant predictors and to

detect any potential interaction, all factors were included in the first run. Once the

salient predictors were identified, I carried out numerous cross-tabulations between the

dependent variable and each linguistic factor to control for knock-outs. In the next

step, I ran an interaction factor in Rbrul which revealed the interaction between style

and the following phonetic segment, as discussed later in the chapter. Finally, a step-

up/step-down regression analysis was carried out. The type of response is binary: /t/-

glottaling vs. plain /t/, with /t/-glottaling as application value. Table 6.2.1 indicates the

constraints and their related factor groups of the final model for both WF and WM /t/.

8
In both cases, the number of occurrences exceeds the common statistical threshold (Guy, 1980).

188
 Constraints Factors
Preceding manner of articulation nasals (e.g. prevalent; printer)
sibilant fricative (e.g. machinist; sixteen)
non-sibilant fricatives (e.g. craft; fifteen)
/l/ (e.g. Walt; multiple)
stops (e.g. craft; factory)
high vowels (e.g. quit; eighty)
mid vowels (e.g. it; nightmare)
low vowels (e.g. habitat; apartment)
Following manner of articulation nasal (e.g. eat meet; nightmare)
fricatives (e.g. favourite food; itself )
stops (e.g. rent because; netball)
liquid (e.g. newest little; completely)
glides (e.g. brought you; Gatwick)
high vowels (e.g. updated)
mid vowels (e.g. upset about; potato)
low vowels (e.g. went on; regret and)
pause (e.g. what?)
Syllable stress Primary stress – (t) occurs in primary
stressed syllables (e.g. sit; eighteen);
Non-primary (e.g. operate; community)
Style Spontaneous speech, reading styles, word
lists
Word frequency low frequency (1-3); high frequency (4-7)9
Social class working class, middle class
Age young (18-28), middle (35-50), older (60+)
Sex female, male

Table 6.2.1 Constraints on the (t) variable for the WF and WM dataset.

6.3 Overall Results and Individual Variation for word-final /t/

This section outlines the overall results of word-final /t/ and presents the step-up/step-

down regression analysis of the above model reporting the findings from the three

localities together. The figure below reveals that /t/ glottaling in word final context is

well-distributed across class, sex and age. Working class males lead the young group

9
Word frequencies presented as Zipf- values can be divided into low-frequency words (values 1-3)
and high-frequency words (values 4-7) (Van Heuven et al., 2014).

189
with 83% of glottal realisations; women of both classes are slightly ahead among

middle-aged speakers; whereas the least glottal realisations were found among middle

class females in the old group.

100

80
% glottaling

WC females
60
WC males
40 MC females
20 MC males
0
young middle old

Figure 6.3.1 Overall results for word-final /t/ glottaling.

Having delineated, in brief, the profile of word-final /t/ in East Anglia, let us turn the

attention to the mixed-effects regression analysis for an in-depth discussion of the

findings. Model comparison was carried out through a log-likelihood ratio test, and the

best model achieved in the multivariate analysis shows that preceding environment,

following environment and style are marked as significant predictors. Class, sex and

age do not exhibit statistically significant responses. The lack of social effect is

consistent with recent research carried out in both London and Edinburgh (Schleef,

2013), yet they do not match those studies in which (t) glottaling has a social effect

(e.g. Smith & Holmes-Elliott, 2017)10 - this issue will be addressed later in the chapter.

10
Despite the social effect found in Buckie, these social influences were weakening over time.

190
 Application value = glottaling; overall proportion = 0.703
R2 = 0.599; log likelihood = -1125.923; N = 3051
Constraints Logodds FW % Tokens
Preceding environment
p<.001
central vowel 1.985 0.88 95 37
front vowels 1.099 0.75 80 762
back vowels 0.887 0.71 79 1411
nasals 0.846 0.70 74 350
/l/ 0.446 0.61 57 94
stops -2.062 0.113 23 126
fricatives -3.201 0.039 18 271

Style
p<.001
spontaneous
speech 1.656 0.84 82 2091
reading styles 0.083 0.521 53 615
word lists -1.739 0.149 30 345
Following environment
p<.001
nasals 2.285 0.908 91 87
liquids 0.688 0.665 94 107
glides 0.553 0.635 87 271
stops 0.481 0.618 91 233
fricatives -0.066 0.484 80 421
central vowel -0.597 0.355 79 227
pause -0.951 0.279 50 1069
front vowels -0.962 0.276 75 533
back vowels -1.432 0.193 62 103

Table 6.3.1 Multivariate analysis of (t) glottaling in the whole dataset.

Table 6.3.1 reports no significant effect of lexical frequency for word-final /t/. The

following sections draw the attention to those predictors that surfaced in the mixed-

effects regression analysis and discusses factor groups in order of significance.

6.3.1 Preceding environment - word-final /t/

The most powerful constraint for word-final /t/, in East Anglia, is the preceding

environment. As detailed in Chapter 5, most of the studies on (t) glottaling have limited
191
the preceding context to vowels. Despite the restriction on the use of [Ɂ] before and

after a schwa /ə/ (Trudgill, 1974), Table 6.3.1 shows that, amongst vowels, a preceding

schwa /ə/ promotes the most glottaling (.88). Crosstabulations revealed that 82% of

tokens with a preceding /ə/ occurred in unstressed syllables. But does this also hold for

Norwich? (this question will be answered in section 6.4.3). Front vowels are the second

most favouring segment with a probability of .75, followed by back vowels which

favour at .71 and have the highest token number. It is interesting to note the strong

influence of preceding vowels since they are not divided by any other factor group in

the hierarchy. Vowels also exhibit the highest effect size amongst Edinburgh (.67),

London’s teenagers (.61) (Schleef, 2013), and in Vermont English (.57) (Roberts,

2006), while in RP the use of [Ɂ] is only favoured by back vowels (.57) - front and

central vowels disfavour it (Barrera, 2015). In phonological theory, front vowels and

coronal consonants are deemed to be members of the natural class of coronal sounds

(Clements and Hume, 1995). Hence, /t/ is more likely to be realised as a coronal feature

rather than with a glottal stop which differs in the place of articulation. However, even

though front and central vowels were marked as disfavouring factors in RP (Barrera,

2015), Barrera reports a visible variability. This means that not only has (t) glottaling

spread from one linguistic environment to another (e.g. after a preceding schwa /ə/),

but has also spread, in the same linguistic contexts, to more formal or statusful varieties.

This would explain the high use of [Ɂ] after certain central and front vowels in East

Anglian English, and it would suggest that the diffusion of glottal variants after a

preceding schwa has completed its linguistic change in the vernacular, at least in East

Anglia.

192
 1

0.8

0.6

0.4

0.2

0
central vowel front vowels back vowels
EA RP

Figure 6.3.2 Probability of WF /t/ glottaling between EA and RP.

The fourth most favouring effect on (t)-glottaling concerns nasals11. The difference in

probabilities between nasals (.70) and low vowels (.71) is negligible, suggesting that

the distinction between these factor groups might not be relevant. Laterals are the last

favouring preceding predictor to trigger glottal(ised) variants. By contrast, Schleef

(2013) found nasals and liquids to inhibit (t)-glottaling, confirming Roberts’ (2006)

findings where preceding consonants disfavoured [Ɂ] at .34. Trudgill (1974) reported

analogous findings in Norwich English with no [Ɂ] in the /n_/ and /l_/ contexts, as in

went and felt. Preceding stops, as expected, favour /t/ retention and disfavour /t/

glottaling at .113. Fricatives have the lowest probability resulting in .039. In this study

fricatives are not treated separately (sibilants vs. non-sibilants) as a likelihood ratio test

of the model with combined fricatives against the model with separate fricatives did

not reveal a significant difference between models ( x2 (1) = 0.16, p > .05).

No comparison can be made with previous studies with respect to preceding stops and

fricatives – they were usually excluded as obstructing (t)-glottaling12. This research

reports a slightly visible variability of glottal variants when /t/ is preceded by stops /p,

11
The nasal category mainly refers to /n/ (n = 350) which hugely outnumber preceding /m/ (n = 3).
12
A further run not shown here, reveals that when stops and fricatives are excluded from the analysis,
nasals and preceding /l/ disfavour glottaling, along the line of Schleef’s (2013) study.

193
k/ and fricatives /f, s/, yet they all remain the most inhibitory factors for (t) glottaling

to occur. Since [Ɂ] is dramatically spreading, and since diffusion can also occur

linguistically (from one environment to the other) (Straw & Patrick, 2007), the use of

glottal stops in these contexts might be accounted for as an early stage diffusion into

another linguistic environment.

From a theoretical standpoint, the sonority hierarchy seems to be a partial

explanatory factor as more preceding sonorous segments favour glottal(ised) variants,

while less sonorous segments disfavour it. The sonority scale, which refers to the

ranking of speech phones by amplitude, is proposed by Goldsmith (1990) and Laver

(1994) as follows: (most sonorous) vowel > glide > liquid > nasal > fricative > affricate

> stop/plosive (least sonorous)13.

This ranking is nearly parallel to the East Anglian findings with preceding vowels

favouring (t) glottaling, while a reverse order occurs with respect to nasals and /l/, as

well as stops and fricatives. The exchange of place between nasals and liquids, in terms

of sonority, should not be problematic given the small probability difference between

them. If we explore in more detail the sonority scale within the vowel system, we also

find a reversed order to what is theoretically predicted: (most sonorous) low open

vowels > mid vowels > high close vowels (least sonorous) (Laver, 1994). However,

the present study shows mid vowels at the top of the ranking followed by front and

back vowels.

13
This phonotactic principle, whose purpose is to describe the structure of a syllable in terms of sonority,
is known as Sonority Sequencing Principle (SSP). For an in-depth discussion on the syllable structure
(onset, nucleus and coda) see Goldsmith (1990) and Laver (1994).

194
 100

80
% glottaling

60

40

20

0
central front back nasals /l/ stops fricatives
vowel vowels vowels

Figure 6.3.3 Rates of /t/ glottaling for the preceding environment.

6.3.2 Style-shifting in word-final /t/

The style-shifting analysis seeks to investigate the distribution of the (t) variable in

conversational and controlled speech. It is argued that when speech is unselfconscious,

the style is closer to the vernacular, while when speech is more self-conscious it will

be more closely to the standard variety (Labov, 1966). Trudgill (1974) demonstrates

that the glottal(ised) variants, in Norwich English, are inversely proportional to social

class and social context. High classes in formal style exhibit low levels of glottalization,

whilst lower classes glottal(ise) more frequently in spontaneous speech. Interestingly,

Trudgill’s (1988) real time study revealed that the variable (t), intervocalically and in

word-final /t/, slightly increased in casual style. Conversely, in more formal styles there

was a dramatic spread suggesting how “a change having gone almost to completion in

casual speech, continues to spread from style to style” (Trudgill, ibid: 44). Holmes-

Elliott’s (2019) real time study in Hastings, which explores the development of (t)-

195
glottaling (among other linguistic variables) 14 from childhood to adolescence, found

higher rates of glottaling with individuals showing convergence over time and moving

in the same direction. The significant effect of style, in Hastings, shows lower rates in

more formal contexts15. Figure 6.3.4 displays the distribution of (t) across style-

shifting, which is the second most significant constraint in the present analysis.

Figure 6.3.4 Rates of (t) glottaling by three speech styles.

This outcome, which shows that the (t) variable is highly sensitive to style-shifting,

corroborates Trudgill’s (1988) findings and it is in line with previous studies (Milroy

et al. 1994; Tollfree 1999; Williams & Kerswill 1999; Mees & Collins 1999; Stuart-

Smith et al. 2007). Indeed, the graph illustrates a very high rate of glottal(ised) variants

in spontaneous speech, followed by reading passages - where the use of [Ɂ] is almost

in a neutral position (.52) - and the word lists, which are marked as a disfavouring

factor. The question whether /t/-glottaling is losing its stigma remains open, since it is

“increasingly tolerated in more careful register” (Kerswill & Williams, 2000) and it is

14
This real time study also includes: GOOSE fronting, TH-fronting and /s/-realisation (Holmes-Elliott,
2019).
15
Kerswill & Williams (2000) argue that “children slowly gain sociolinguistic maturity in a manner
that involves a gradual increase in the number of styles that are perceived and treated in an adult way”.

196
diffusing to more formal styles in younger speakers (Stuart-Smith 1999; Marshall

2002).

Conversational speech style and preceding environment. It has been proven that

linguistic changes begin in the vernacular spontaneous speech before continuing to

spread from style to style (Trudgill, 1988). Along this line, figure 6.3.5 illustrates the

distribution of word-final /t/ by preceding phonological environment in casual speech.

. Figure 6.3.5 WF (t) glottaling by preceding environment in casual speech.

If glottal(ised) variants were previously found not to occur before and after a schwa

(Trudgill, 1974), in the present sample [Ɂ] is strongly promoted by all vowels,

regardless of their highness or backness, and a preceding schwa leads the hierarchy.

The position of nasals and /l/ in the hierarchy is noticeable as are their high glottaling

rates. Stops, non-sibilant and sibilant fricatives occupy the last places in the ranking,

but they still show evidence of glottaling whose rate does not go below 26%. Let us

now explore the (t) variable by following phonetic segment.

197
6.3.3 Following environment in word-final /t/

As previously discussed, the bulk of research on the (t) variable has mostly focused on

the following phonetic segment, which is considered the most fruitful constraint. This

ruling predictor, called the diffusion pattern after Straw & Patrick (2007), refers to the

ordering of diffusion in different linguistic contexts: PreC > PreP > PreV. This pattern

is repeated in many southern communities, such as London (Hudson & Holloway 1977;

Tollfree 1999; Schleef 2013), Milton Keynes (William & Kerswill 1999) and in many

northern places such as Derby (Foulkes &. Docherty 1999a), Hull (Williams &

Kerswill 1999), and Edinburgh (Schleef, 2013). In the Manchester data, PreC leads the

ranking, yet no significant difference was found between PreP and PreV (Baranowski

& Turton, 2015). Previous research (e.g. Trudgill, 1974) shows that the use of glottal

stops and glottal(ised) variants is very common in Essex, Suffolk and Norfolk

(Trudgill, 1974), however, as Straw & Patrick (2007: 392) note, “the [linguistic]

environments have not previously been applied to glottal variation in East Anglia”.

In 2007, these environments were adopted to constraint the (t) variable in Ipswich,

where the PreV context was found to favour glottal variants the most, as the following

ranking shows: PreV > PreC > PreP. Since this hierarchy differs from the diffusion

pattern, it is referred to as the Ipswich pattern (after Straw & Patrick 2007).

The extension of the following phonetic segment in some British dialects shows the

following rankings for word-final /t/ glottaling:

in London (Schleef, 2013):

nasal (.74) > liquid (.64) > fricative and affricate (.53) > plosive (.45) > glide (.41) >
pause (.37) > vowel (.34)
in Edinburgh (Schleef, 2013):

nasal (.78) > fricative and affricate (.68) > liquid (.54) > plosive (.48) > glide (.46) >
pause (.43) > vowel (.16).

198
The overall results from the present East Anglian study confirm the diffusion pattern,

with more glottaling in PreC than PreP and PreV. However, when the vowel category

is subdivided a following schwa precedes pause in the hierarchy. Figure 6.3.6 shows

that, in the following environment, nasal (.908), liquid (.665), glide (.635) and stop

environments (.618) favour the glottal(ised) variants, while fricative (.484), central

vowel (.355), pause (.279), high vowel (.276) and low vowel environments (.193)

disfavour it. The social distribution in this context does not reveal prominent results,

besides all social factors, statistically, do not play a role. In Tyneside, on the contrary,

word-final /t/ pre-vocalically was found to be more common among young middle-

class men.

Following pause is usually referred to as an arbitrary facor for (t,d) (Patrick, 1991) as

deletion of apical stops, after a pause, varies across localities. Similarly, pause appears

to be localised even for the distribution of (t) glottaling as glottal(ised) variants do not

occur when followed by a pause in Tyneside (Docherty et al., 1997), in London

(Schleef, 2013), in Edinburgh (Schleef, 2013), yet this predictor holds true for Milton

Keynes (Kerswill & Williams, 1992). Similarly, working-class adults in Glasgow

retain their categorical use of glottals when /t/ occurs before a following pause (Stuart-

Smith, 1999).

100

80
% glottaling

60

40

20

Figure 6.3.6 Rates of (t) glottaling by following environment.

199
The high rates of glottal variants before a following nasal or a liquid is comparable to

the London and the Edinburgh studies. By way of contrast, in East Anglia, the place of

glides and fricatives is reversed: the former favours (t) glottaling, whereas the latter

disfavours it.

The high position of glides should not be surprising as the probability of

glottaling in this context is already well-established in RP, where glides are the second

most favouring predictor (.63) (Barrera, 2015). However, the low place of fricatives in

East Anglia remains a mystery, for now, since (t) glottaling before this phonetic

environment has recently been attested even in RP (Barrera, 2015).

Due to the low position of vowels in the hierarchy, the Sonority Sequencing Principle

could not be adopted to explain the ranking of the following environment.

An additional theory on the sonority hierarchy, known as the Dispersion Principle,

suggests that “a language will preferentially maximize sonority difference in the

syllable onset, but minimize it in the coda” (Clements, 1990: 177). Even in this case, a

following vowel – at the onset – maximises sonority and does not apply to the data

discussed here. Perhaps, sonority simply does not govern word-final /t/-glottaling.

Conversational speech style and following environment. Let us now explore the

distribution of word-final /t/ for following phonetic segment in casual speech.

Figure 6.3.7 displays a rate which goes above 85% before nasals, liquids, stops and

fricatives; while the graph line decreases when glottal(ised) variants are followed by a

schwa, pause, front and back vowels. This, as expected, suggests that the use of

glottal(ised) variants in PreC has nearly completed its linguistic change. Indeed, it has

been proven that word-final /t/-glottaling, in the PreC environment, is also well-

established in RP (Fabricius 2000; Kerswill 2007; Barrera 2015). Fabricius (2000)

200
comments on the loss of stigma of word-final pre-consonantal /t/-glottaling having

reached the young upper-middle class generation.

Figure 6.3.7 Rates of (t) glottaling by following environment in casual speech.

A closer examination of the distribution of glottal(ised) variants in casual speech

reveals that fricatives exhibit a high rate of glottaling behaving like other following

consonants. This reconfirms that (t) is sensitive to style-shifting and shows how

phonetic environments vary according to speech style.

6.3.4 Non-significant factors – overall findings for word-final /t/

The correlation between speech style and social variables has long been a key for (t)-

glottaling. However, this study does not report a significant effect for any of the social

predictors investigated. Why? The lack of statistical significance of social factors has

recently been reported in Manchester (Baranowski & Turton, 2015)16, in London,

Edinburgh (Schleef, 2013) and Hastings (Holmes-Elliott, 2019).

16
Age is a significant factor in Manchester. See Baranowski & Turton (2015) for further details.

201
It is argued that the absence of social class and gender significance of a variable

suggests that the advanced change is nearing the completion (Baranowski & Turton,

2015). Appendix II displays statistical details from Rbrul one-level analysis. Despite

being non-statistically significant, sex, age and social class are in the right direction

that we would expect. The youngsters are slightly ahead of middle-aged, while old

speakers fall behind. Indeed, Trudgill (2004a) notes that this supralocal consonant

feature is on the increase among young East Anglian speakers. This trend is illustrated

in figure 6.3.8. showing the distribution of /t/-glottaling across individuals (with the

young at the left), whose proportion of the variance equals 0.487.

100

80
% glottaling

60

40

20

+18 87
Figure 6.3.8 Rates of WF /t/ glottaling by individuals' age.

Males glottal(ise) /t/ more than females, whereas middle-class speakers fall behind

working-class ones. However, in some speech communities (e.g. Cardiff) young

middle-class speakers were found to promote glottaling the most (Mees & Collins,

1999). The change from above for Cardiff, where glottals are borrowed, contrasts with

the old change for East Anglia where glottals are not a borrowing, hence not a change

from above.

202
The correlation between sex and class did not exhibit noteworthy results, yet the

relationship between age and class, as in figure 6.3.9, shows that: (1) young working-

class speakers glottal(ise) the most; (2) the degree of glottaling between middle and

working-class middle-aged speakers is levelled, while (3) old speakers use the glottal

stop [Ɂ] to a smaller extent. Altendorf (2003: 91) argues that: “T glottalling has

increased in all social classes17, styles and phonetic contexts in London and the south-

east … where it is now very widespread indeed.”

100

80
% glottaling

60
working class
40 middle class
20

0
young middle-aged old

Figure 6.3.9 Rates of word-final /t/ glottaling by class and age.

Previous research demonstrated that (t) glottaling is a well-established local variable

with a long history of use in certain sites, despite the evaluation of some more recent

change. East Anglia, indeed, is considered one of the geographical areas where the

glottal stop started spreading first – from Norwich to London (Trudgill, 1999). Origin

aside, given the long presence of the glottal feature in this geographical and linguistic

area, it is not surprising that the three localities investigated behave similarly. Results

from one-level analysis display a high use of glottals in Ipswich > Colchester >

Norwich, with Norwich showing the smallest amount of [Ɂ], but not significantly. The

17
Note that Altendorf (2003) defined social class merely by type of school: working-class speakers
attended comprehensive schools, while middle-class speakers were those who attended grammar
schools.

203
fact that Norwich exhibits lower rates of glottaling might be due to the relatively lower

number of tokens extracted from this city. The last predictor which did not reach

statistical significance is lexical frequency. In this analysis, lexical frequency is binary:

high frequency words vs. low frequency words. The former refers to the SUBTLEXUK

Zipf-values 4-7, whilst the values attached to the latter are 1-3.

Finally, in the attempt to answer the question posed at the beginning of this section -

why don’t social factors play a salient role? – I believe that the lack of social effect in

East Anglia is not really surprising if compared to contemporary studies, and it adds to

the general ‘loss of stigma’ argument.

So far, we have presented and discussed the results from the three locations together to

gain an overall understanding of the behaviour of (t) in the linguistic East Anglia area.

Let us now examine each location separately to monitor the consistency of linguistic

and social constraints across geographical space.

6.4 Treating the three localities separately

The overall findings revealed that linguistic constraints and style trigger glottaling the

most in East Anglia. The preceding environment proved the most fruitful constraint in

terms of effect size and linguistic diffusion, showing that [Ɂ] has spread into other

phonetic contexts (e.g. after a preceding schwa /ə/). The following phonetic segment

reported analogous findings with respect to the internal diffusion of glottals, which is

also conditioned by style-shifting. Since the three localities are sited in what Trudgill

(2001a) defines as ‘linguistic East Anglia’, we would expect Colchester, Ipswich and

Norwich to conform, overall, to the East Anglian pattern. However, internal difference

might surface in the behaviour of word-final /t/ when the three localities are

investigated separately.

204
6.4.1 Results from Colchester – word-final /t/

Results from the multivariate multiple regression analysis with speaker as random

effect show that the most powerful predictors in order of significance are style,

preceding environment, following environment, as illustrated in table 6.4.1.

Style. In Colchester, /t/-glottaling is distinctly conditioned by style-shifting, indeed it

is the most favouring constraint. As expected, results from style-shifting match the

overall pattern with /t/ being more likely to glottal(ise) in spontaneous speech than in

more formal style, whilst in words lists /t/ is more likely to be realised as an alveolar

stop.

Preceding environment. The second most favouring predictor is the preceding phonetic

segment with a marginal change at the top of the ranking, compared to overall findings.

(1) East Anglian pattern:

central vowels > front vowels > back vowels > nasals > /l/ > stops > fricatives

(2) Colchester:
front vowels > central and back vowels > nasals > /l/ > stops > fricatives

In Colchester, high vowels favour (t)- glottaling slightly more than central and back

vowels. It has to be noticed that in the Colchester dataset, central and back vowels were

grouped together in the preceding environment owing to the low number of tokens in

the mid category and due to the similarities in terms of probability weight between the

two factor groups. The rest of the hierarchy matches the East Anglian pattern, with

stops and fricatives as disfavouring factors.

205
 Application value = glottaling; overall proportion = 0.726
R2 = 0.644; log likelihood = -349.502; N = 1097
Constraints Logodds FW % Tokens
Style
*** p<.001
spontaneous
speech 2.096 0.890 87 769
reading styles 0.028 0.507 49 206
word lists -2.121 0.107 23 122
Preceding Env.
*** p<.001
front vowels 1.035 0.738 81 544
central & back
vowels 1.030 0.737 78 277
/l/ 0.928 0.717 65 20
nasals 0.684 0.665 69 118
Stops -1.441 0.191 27 41
fricatives -2.236 0.097 32 97
Following Env.
*** p<.001
nasals 2.840 0.945 93 29
liquids 1.063 0.743 98 42
glides 0.693 0.667 91 92
stops 0.182 0.545 89 73
fricatives -0.188 0.453 83 169
central vowels -0.833 0.303 83 77
pause -1.002 0.268 52 378
front vowels -1.054 0.258 78 200
back vowels -1.700 0.155 65 37

Table 6.4.1 Multivariate analysis of WF /t/ glottaling in Colchester.

Following environment. The following phonetic segment is marked as the third

significant predictor to condition (t). Factor groups within this constraint influence (t)

glottaling in the same probability order as the overall East Anglian pattern, matching

the diffusion pattern18:

PreC > PreP > PreV.

18
Barring the inclusion of glides among consonants.

206
As discussed in section 6.3.4, none of the external factors, style aside, reached

statistical significance when the three areas were combined; this result still holds for

Colchester when it is separately analysed. Let us now turn the attention to the second

site investigated: Ipswich.

6.4.2 Results from Ipswich – word-final /t/

This section reports the results from a step-up/step-down regression analysis, with

preceding environment, following environment and style, as illustrated in table 6.4.2.

Preceding environment. In Ipswich, the greatest effect is contributed by the preceding

phonetic segment, with central & back vowels (.832), nasals (.80) and front vowels

(.735) favouring glottal(ised) variants, whilst /l/ (.493), stops (.126) and fricatives

(.071) exhibit a negative correlation. These findings go in the direction of the overall

East Anglian pattern, despite slight ranking differences.

(3) East Anglian pattern:

central vowels > front vowels > back vowels > nasals > /l/ > stops > fricatives

(4) Ipswich
back vowels > nasals > front vowels > /l/ > stops > fricatives.

Vowels, in Ipswich, are separated by nasals, and back vowels trigger slightly more

glottaling than front vowels. It has to be noticed that central vowels have been excluded

from the regression analysis as the use of the glottal stop in this context turned out to

be categorical. This provides additional empirical support to the linguistic diffusion of

[Ɂ] (e.g. after a preceding schwa). The exchange of place in hierarchy between sibilant

and non-sibilant fricatives is not statistically relevant.

207
Following environment. The second most powerful constraint is the following context,

with nasals (.957), glides (.668), stop (.586) and liquids (.565) promoting the use of the

glottal stop, while fricatives (.49), vowels (.37, .252, .161) and pause (.247) do not

encourage (t)-glottaling.

(5) East Anglian pattern:

nasals > liquids > glides > stops > fricatives > central vowels > pause > front vowels
> back vowels

(6) Ipswich:
nasals > glides > stops > liquids > fricatives > central vowels > pause > front vowels
> back vowels

Previous studies show that nasals and liquids are the most preferable environments for

[Ɂ] to occur, whereas glides and plosives were found to disfavour glottal(ised) variants

in both London and Edinburgh (Schleef, 2013). However, the high place of glides and

stops in the above hierarchy confirms the strong effect of these predictors in East

Anglia - in Ipswich in particular - where glides and plosives separate nasals and liquids,

as in (6). Straw & Patrick (2007) demonstrate that, in Ipswich, word-final /t/ does not

follow the usual diffusion pattern (PreC > PreP > PreV), yet the apical stop is more

likely to be glottal(ised) before following vowels than before following consonanta and

pause yielding the following ranking: PreV > PreC, PreP – which is referred to as the

Ipswich pattern (after Straw & Patrick, 2007)19. By contrast, the findings from the

present study are not analogous to the Ipswich pattern, as all vowels strongly disfavour

/t/-glottaling.

19
This pattern was found amongst Ipswich Anglo urban speakers, whereas the use glottal variants for
Barbadians was higher in PreP position.

208
 Application value = glottaling; overall proportion = 0.729
R2 = 0.469; log likelihood = -469.458; N = 1114
Constraints Logodds FW % Tokens
Preceding Env.
p<.001
central & back
vowels 1.601 0.832 83 317
nasal 1.389 0.80 79 155
front vowels 1.021 0.735 78 488
/l/ -0.030 0.493 47 19
stops -1.414 0.196 29 52
fricatives -2.567 0.071 19 83
Following Env.
p<.001
nasals 3.110 0.957 97 40
glides 0.700 0.668 90 95
stops 0.348 0.586 95 81
liquids 0.263 0.565 95 37
fricatives -0.039 0.49 84 146
central vowels -0.530 0.37 83 78
front vowels -1.087 0.252 77 198
pause -1.115 0.247 51 405
back vowels -1.650 0.161 64 34
Style
p<.001
spontaneous
speech 1.394 0.801 83 760
reading styles 0.038 0.506 54 244
word lists -1.392 0.195 43 110

Table 6.4.2 Multivariate analysis of word-final /t/ glottaling in Ipswich.

Style. Another significant constraint which helps explaining the variability of (t)

glottaling, in Ipswich, is style.

6.4.3 Results from Norwich – word-final /t/

Similar findings, with respect to style, were found in Norwich. Table 6.4.3 reports the

results from a multivariate analysis, with preceding environment, following

environment and style marked as significant predictors in the conditioning of (t).

209
 Application value = glottaling; overall proportion = 0.716
R2 = 0.559; log likelihood = -296.004; N = 749
Constraints Logodds FW % Tokens
Preceding Env.
p<.001
central and back
vowels 1.239 0.775 79 205
front vowels 1.052 0.741 75 379
nasal 0.900 0.711 69 77
/l/ 0.363 0.59 58 55
stops -3.555 0.028 9 33
Following Env.
p<.05
nasals 1.705 0.846 81 16
stops 0.927 0.717 93 74
liquids 0.830 0.696 96 26
glides 0.427 0.605 85 76
fricatives -0.147 0.463 75 98
central vowels -0.568 0.362 81 63
pause -0.993 0.27 53 249
front vowels -1.072 0.255 76 117
back vowels -1.110 0.248 60 30
Style
p<.001
spontaneous
speech 1.538 0.823 84 491
reading styles 0.182 0.545 63 145
word lists -1.721 0.152 25 113

Table 6.4.3 Multivariate analysis of word-final /t/ glottaling in Norwich.

Before comparing the above results to the East Anglian pattern, we should draw

attention to the number of tokens analysed in this research site. The original number of

tokens for word-final /t/ in Norwich equalled 840. However, in the pre-fricative context

(n = 91) /t/ was categorically retained, hence these occurrences were excluded from the

regression analysis, giving a total of 749 with an average of 62 tokens per speaker.

Central vowels were collapsed with back vowels as, in this environment, /t/ was nearly

210
categorically glottal(ised). It is unlikely that this would have a significant statistical

influence, given the very low tokens in the central vowel category (n = 10).

Preceding environment. Results from Norwich are near identical to the East Anglian

pattern: central and back, as well as front vowels (.775, .741) strongly favour the glottal

stop [Ɂ] along with nasals (.711) and /l/ (.59), while stops (.028) strongly disfavour it.

Following environment. The second greatest effect is contributed by the following

phonetic context with nasals (.846) at the top of the ranking followed by stops (.717),

liquids (.696), and glides (.605) which exhibit a positive correlation; similar to Ipswich,

stops precede liquids in the ranking, whereas fricatives (.463), vowels (.362, .255.,

.248) and pause (.27) consistently disfavour /t/-glottaling even in Norwich.

Interestingly, pause, which seems to be a localised constraint (Docherty et al., 1997),

consistently disfavours (t) glottaling in Colchester, Ipswich and Norwich.

It is not surprising that certain following vowels disfavour [Ɂ] as previous studies

carried out in Norwich reported some restrictions on the application of the diasystemic

rule: /t/ x <[t] ~ [tɁ] ~ [Ɂ]> (Trudgill, 1974). As discussed early in the chapter,

one of the limitations refers to the use of the glottal stop after a preceding schwa.

However, the present survey has demonstrated that (t) glottaling in this linguistic

environment has completed the change in East Anglia, as this finding was consistent

in all three localities. Moreover, /t/-glottaling after a preceding schwa has been recently

attested even in RP (Barrera, 2015). The second restriction concerns the following

environment: /t/-glottaling is blocked when followed by a schwa /ə/ or unstressed /ɪ/,

as in put it [phot‘əɁ], yet in the case of stressed /ɪ/ the glottal stop can apply as in went

into. Schematically, Trudgill (1974: 175) summarises the rule as follows:

211
 Figure 6.4.1 Restrictions on the application of (t) glottaling. Adapted from Trudgill
(1974).

This second restriction is consistent with the findings of the present study. So far, we

have presented and discussed the findings from word-final /t/ in East Anglia, as a

linguistic area first, and secondly, we have investigated the three urban areas separately

to verify the consistence of constraint ranking. Let us now explore /t/ glottaling in word-

medial position.

6.5 Overall results and individual variation for word-medial /t/

This section draws attention to word-medial /t/ from the whole East Anglian dataset.

The findings suggest that the change in status of /t/ glottaling in word-medial position

is behind that of /t/ glottaling in word-final environments due to the lower rates of rule

application and to the noticeable variation across social variables. Despite the

differences between socio-economic classes, this predictor does not play a significant

statistical influence on this variable. Males of both classes are in the lead among the

young group where the least glottal realisation occurs among middle class females

(22%); middle-aged speakers seem to use a relatively similar amount glottal variants,

except for middle class females (16%); in the old group, working class males

glottal(ise) the most reaching a rate of 51%. Furthermore, the use of word-medial /t/

glottaling appears to be aged-graded among working class males.

212
 100

80

% glottaling WC females
60
WC males
40 MC females
MC males
20

0
young middle old

Figure 6.5.1 Overall results for word-medial /t/ glottaling.

In the literature, word-final /t/ and word-medial /t/ are typically investigated separately

as two different phonological contexts where the rule application is not identical. The

preceding phonetic segment of word-medial or word-internal /t/ is typically restricted

to vowels (e.g. intervocalically). To account for word-medial /t/, I initially

distinguished four syllabic contexts aiming at a more detailed descriptive analysis:

(a) /t/ in word-medial syllable final (e.g. nightmare, before a morpheme boundary)20,
(b) intervocalic position (e.g. better)
(c) syllabic /l/ contexts (e.g. bottle), and
(d) /t/ at the onset position (e.g. seventeen, sometimes21).

Tokens with syllabic /n/ (e.g. button) were initially included in the multivariate

analysis, yet /t/ was categorically retained. While the intervocalic position includes all

the unstressed contexts, the onset position includes the stressed ones (e.g. the -ee/-oo

tokens, such as tattoo). The latter, as previously outlined, was commonly found to be

20
This study follows Tollfree (1999) in treating /t/ in word-medial syllable final contexts in tokens like
nightmare.
21
Word-medial foot initial tokens (e.g. canteen) were all preceded by a consonant, while word-medial
syllable final positions (e.g. Gatwick) were mostly followed by a consonant, except for 6 occurrences
out of 163, which are followed by a vowel. The latter result from the regular past -ed affixation (e.g.
adopted, reported).

213
the most constraining environment, yet several studies have demonstrated that, in some

British dialects22, /t/-glottaling has started spreading through the prosodic hierarchy

(Harris & Kaye 1990; Baranowski & Turton 2015; Smith & Holmes-Elliott 2017).

However, due to the low number of tokens in certain factor groups23, the four syllabic

contexts were collapsed into one more general category: word-medial /t/ - in line with

previous research (e.g. Schleef 2013; Hall-Lew et al. 2019). However, when the data

are divided by syllable context the hierarchical distribution of /t/ glottaling is the

following:

morpheme boundary > syllabic /l/ > intervocalic > onset.

80

60
% glottaling

40

20

0
morpheme boundary syllabic /l/ intervocalic onset position

Figure 6.5.2 Rates of (t) glottaling by syllable type.

Where comparison is possible, the above hierarchy seems to resemble Buckie for the

high use of non-standard variants in pre-syllabic lateral /l/ position24, yet this context

is strongly disfavouring in other varieties (see Stuart-Smith 1999) and it is considered

as the second least favouring environment in Hughes et. al’s (2012) ranking, illustrated

22
So far, there is evidence of /t/-glottaling at the onset position in Manchester and Buckie.
23
In Buckie, syllabic context and following environment were combined “within one elaborated
category” as the following environments result in: pause, vowel and syllabic consonant (Smith &
Holmes-Elliott, 2017: 12).
24
In Buckie, /t/ glottaling rates in pre-syllabic lateral position go beyond 75 per cent.

214
in Chapter 5. The intervocalic position follows with a modest difference, employing

lower rates of application as previous descriptions of other dialects have shown.

However, the argument that the ongoing spread is most noticeable intervocalically,

seems to also hold for East Anglia. In Derby, for instance, only younger speakers

produced 22% of glottals, while older ones produced only four medial glottal tokens

(Docherty & Foulkes, 1999).

On the other hand, the onset environment exhibits a marginal use of [Ɂ] in this

context (for a robust use of [ʔ] in onset positions see Smith & Holmes-Elliott 2017).

This finding suggests that, despite the low probability of application, in East Anglia, /t/

is not categorically retained at the onset, and that glottaling might have started the next

stage of this process, that is diffusion in stressed environments. In a similar vein,

Bermúdez-Otero (2010) argues that a phonological process, over time, may spread

through the prosodic hierarchy, affecting inclusive environments. With reference to

glottalization in word-initial syllable onset positions, Docherty et al. (1997: 290) argue

that:

“It is in general unwise to make a negative claim to the effect that glottalisation
does NOT occur in this position, even if the occurrences are rare… In the
environments reviewed above which are said to block glottalisation, we can state
more accurately that the probability of glottalisation occurring is lower than in,
say, presyllabic-lateral positions.”25

From a statistical standpoint, the best model achieved in the multivariate analysis

shows that word-medial /t/ is significantly constrained by the preceding phonetic

segment, lexical-frequency, following phonetic segment, sex, age and stress26.

25
This argument also holds for the application of glottaling / glottalisation in word-final /t/ after
preceding fricatives. As noted earlier, the phonological process is marked as a highly disfavouring factor,
but in this context glottal(ised) variants are not categorically blocked.
26
For the word-medial sample, style was not included as a predictor in the regression analysis as /t/ was
categorically glottal(ised) in informal speech, whilst in the more formal style (i.e. reading passages) it
was categorically realised as a plain /t/. Hence, this constraint had to be excluded from the multivariate
analysis since the only remaining factor group was the word lists, with a rate of 10% glottaling.

215
Preceding environment. In the word-medial sample, the greatest effect is contributed

by the preceding phonetic segment, similarly to word-final /t/. The most favouring

factors are vowels, with front ones (e.g. natter) leading the ranking (.80). Central and

back vowels (e.g. senator; cutter) favour at .555, slightly ahead of the .50 threshold and

not very far from being a neutral context. Conversely, preceding nasals (e.g. canteen),

stops and fricatives (e.g. laptop; westfield) disfavour at .441 and .203, respectively.

Comparable findings were found in London and Edinburgh where among teenagers the

glottal stop was more likely to be produced after a preceding vowel, than after a

preceding nasal or liquids. In the East Anglian database, preceding laterals are excluded

from analysis owing to the low number of tokens in this category (n = 4).

216
 Application value = glottaling; overall proportion = 0.361
R2 = 0.530; log likelihood = -911.106; N = 1872
Constraints Logodds FW % Tokens
Preceding Env.
p<.001
front vowels 1.386 0.80 50 919
central & back
vowels 0.222 0.555 32 541
nasals -0.238 0.441 12 263
stops & fricatives -1.37 0.203 7 149
Word-frequency
p<.001
high-frequency 0.689 0.666 46 1289
low-frequency -0.689 0.334 14 583
Following Env.
p<.001
stops & fricatives 1.502 0.818 75 77
nasals 0.274 0.568 68 41
liquids -0.256 0.436 54 167
central & back
vowels -0.656 0.342 35 478
front vowels -0.864 0.297 30 1109
Sex
p<.01
males 0.428 0.605 44 916
females -0.428 0.395 29 956
Age
p<.01
young 0.691 0.666 43 574
middle-aged -0.107 0.473 32 661
old -0.584 0.358 33 637
Syllable stress
p<.01
non-primary 0.174 0.543 37 810
primary -0.174 0.457 35 1062

Table 6.5.1 Multivariate analysis of (t) glottaling in the whole WM dataset.

From a theoretical viewpoint, the sonority hierarchy appears to be an explanatory factor

in this preceding word-medial context, where highly sonorous preceding segments

induce /t/ to glottal(ise): front vowels > central and back vowels > nasals > stops and

fricatives. A closer inspection revealed that the trend of males and females goes in the

217
same direction, yet males are ahead of females especially when /t/ follows a front

vowel, but not significantly.

70
60
50
% glottaling

40
females
30
males
20
10
0
nasals stops and central and front vowels
fricatives back vowels

Figure 6.5.3 Rates of WM /t/ glottaling by preceding environment and sex, p > .05.

With respect to age, figure 6.5.4 shows that old speakers produce slightly more

glottaling in pre-nasal, pre-stop and pre-fricative environments. By way of contrast,

middle-aged speakers near-categorically do not adopt the glottal stop when /t/ follows

stops and fricatives; additionally, they are also behind the two age groups after

preceding front vowels. Whereas young speakers are ahead in the central and back

vowel environment, the rates of glottaling after a preceding front vowel are slightly

higher among old speakers. In the same phonetic context, young and middle-aged show

a rate of application higher than 30% (for middle-aged) and lower than 45% (for young

speakers).

218
 60

% glottaling 45
young
30 middle
old
15

0
nasals stops and central and front vowels
fricatives back vowels

Figure 6.5.4 Rates of WM /t/ glottaling by preceding environment and age, p < .01.

Lexical frequency. The strong effect of lexical token frequency has long been argued

to be an influential factor in previous studies of lenition, such as the (t,d) variable - in

some English varieties (Bybee 2000b). By way of contrast, there seems to be a limited

number of studies which took this predictor into account to explore specifically /t/

glottaling. Schleef (2013) found lexical frequency to be significant word-medially in

both London and Edinburgh, whereas its effect on word-final /t/ emerged only in

London. Barrera (2015) found a word-frequency effect in her RP sample in both

phonological contexts.

From a qualitative viewpoint, Baranowski et al. (2016) argue that high-frequency

words exhibit high rates of (t) glottaling in Manchester, yet the rate of change over time

is the same for high and low frequency words. Hence, there is no evidence that high

frequency words change faster. Their focus is rooted on how /t/ glottaling exhibits

lexical frequency effects, and whether or not the magnitude of this effect increases as

the change progresses. Their findings support modular theories of phonological

representation – different from episodic models, which argue in favour of high

frequency words changing faster.

219
Quantitatively, results from East Anglia confirm that the probability of /t/-glottalling,

in apparent time, is higher in high-frequency words, whose magnitude is twice the size

of low-frequency tokens.

Following environment. The third most relevant predictor to account for the variance

of (t) is the following phonetic segment, with stops and fricatives (e.g. netball; outside),

and nasals (e.g. nightmare) favouring glottal(ised) variants at .818 and .568,

respectively. Liquids (e.g. settle) and following vowel environments (e.g. fainted27;

obscenity) are marked as disfavouring factors.28 Findings from London (Schleef, 2013)

reveal that, word-medially, obstruents (.89), nasals and liquids (.61) favour the glottal

replacement, while following vowels strongly disfavour it (.07). The difference

between East Anglia and London in the behaviour of following liquids remains

unclear29. These results are at odds with Schleef’s (2013) findings in Edinburgh, where

the following environment does not affect word-medial /t/-glottaling in a statistically

significant way (Schleef, 2013).

In Glasgow, categorical absence of glottal stop occurs word-medially in Class

1 adults. (Stuart-Smith, 1999). The glottaling process is attested in word-internal non-

foot initial onset context (e.g. printer, botany, Saturday) (Tollfree, 1999) even in East

Anglia, yet there seems to be no relevant influence of the following vowel type. The

fact that the following environment of word-final and word-medial /t/ vary in the

27
In Norwich English, final -ed can be realised as /-ət/, as in hundred [hʌndɹəɁ], where word-final glottal
replacement can occur (Trudgill, 1974). Similarly, /t/ in fainted is followed by schwa, and it is above
reported as an example of pre-central vowel.
28
Note that glide tokens were excluded from the analysis owing to the low number of tokens (n = 9). In
7 out of 9 tokens, /t/ was preceded by the labio-velar glide /w/ as in Gatwick, network, between etc.;
while /t/, only in two cases, was followed by the palatal glide /j/ where /t/ was retained, as in fortune. In
many varieties of English such cases, in the absence of yod-dropping, are typically palatalised [tʃ]. For
a more detailed account on /j/, in East Anglia, see Trudgill (1974) who reports a retained use of /j/ in
use, ewe, value (but not curlew).
29
This might be due to the fact that liquids are collapsed with nasals in Schleef’s study, yet it is inexplicit
whether these two factors were combined because they were strongly marked as disfavouring factors or
due to the low number of tokens in one of those categories.

220
ranking to a small degree suggests that, at least in East Anglia, /t/ glottaling is under

the control of similar phonetic constraints which move in the same direction. Indeed,

the rankings below show that /t/-glottaling is favoured pre-consonantally, but not pre-

vocalically.

(7) word-final /t/

nasals > liquids > stops > fricatives > central vowels > pause > front vowels > back
vowels
(8) word-medial /t/
stops + fricatives > nasals > /l/ > central and back vowels > front vowels

Following consonants, which exhibit a different order ranking, diverge only in the

behaviour of /l/ word-medially. Following liquids favour /t/ glottaling in word-final

position, yet pre-syllabic lateral /l/ disfavours it in word-medial position.

Let us turn attention to the interaction between the social and the linguistic dimensions.

Crosstabulations between the following environment and sex revealed that women tend

to glottalise more than men when /t/ is followed by nasals, stops and fricatives; while

before liquids and vowels men use the non-standard variants more frequently than

women. This finding is not surprising as it is in line with the literature: word-medial

pre-consonantal has been regarded as a non-stigmatised context (Wells 1997; Tollfree

1999; Altendorf 2003; Hughes et al. 2005), hence glottals are preferred by women;

following syllabic /l/ as well as vowels have been viewed as “stigmatised”

environments, and thus males are ahead in the use of glottal(ised) variants.

221
 100

80

% glottaling
60

40 females
males
20

0
stops & nasals liquids central & front
fricatives back vowels
vowels

Figure 6.5.5 Rates of /t/-glottaling by following environment and sex, p <.01.

Crosstabs by age and following phonetic segment (see figure 6.5.6 revealed that young

speakers are ahead of the middle and old ones, particularly before nasals and front

vowels. The latter is not a surprising finding as glottal(ised) variants word-medially -

precisely in intervocalic context – were found to be preferred by the youngsters30 (see

Milroy et al., 1994). Older speakers show a strong linear effect, with a decreasing of

the non-standard feature as sonority increases. Interestingly, middle-aged speakers

once again fall behind, but this time in different phonetic contexts: before nasals,

central and back vowels, and slightly before liquids.

100

80
% glottaling

60
young
40 middle

20 old

0
stops and nasals liquids central and front
fricatives back vowels
vowels

Figure 6.5.6 Rates of WM /t/-glottaling by following environment and age, p <.001.

30
The tendency of using the glottal stop was particularly associated with young middle-class females
(Milroy et al., 1994).

222
Age and sex. The most marked difference between word-final and word-medial /t/ is

linked to the social dimension which surfaces only word-medially. Sex exhibits a

higher proportion of variation compared to the age of speakers. The fact that class is

not listed among significant predictors in the logistic regression analysis should not be

surprising. Indeed, it has been argued that in terms of influence on linguistic variability,

gender should be considered prior to social class, as the role played by gender is greater

than social class influences in many speech communities (e.g. Milroy et al., 1994).

In East Anglia, males produce more glottal stops than females, in line with the

assumption that “this variant is a male working-class norm” (Docherty et. al, 1997:

305). Previous studies in Norwich (Trudgill, 1974), in Edinburgh (Romaine, 1975), in

London (Hudson & Holloway, 1977), Glasgow (Stuart-Smith, 1999) and many others

show that males were leading the change31 (see Collins and Mees (1999) for opposite

results). Figure 6.5.7 displays a young male preference for the glottals, followed by

older males and finally by the middle-aged counterpart. Females, overall, produce

lower rates of these variants with middle-aged being slightly ahead of younger

speakers, while older speakers fall behind. Statistical details report the influence of

young speakers, whose probability weight is .66, far exceeds that of middle and older

speakers who disfavour /t/-glottaling at .47 and .36, respectively. This means that the

clear gender-related preference in the youngsters diminishes in older speakers and

neutralises in the middle age group.

31
Docherty et al. (1997) found a female preference in the use of glottals in Tyneside, and argue that the
traditional and more common finding (males > females) may be due to the fact that many studies do not
distinguish glottal stops from reinforced stops.

223
 75

60

% glottaling
45 young
middle
30
old
15

0
males females

Figure 6.5.7 Rates of WM /t/- glottaling by sex and age.

Stress. The last predictor which surfaced in the mixed-effects analysis is stress. As

outlined, in chapter 2, stress was coded as a binary factor: primary vs. non-primary –

depending whether /t/ occurred in a stressed (e.g. haters) or non-stressed syllable (e.g.

favourite).

Variation in realisation of /t/ has been previously accounted for with reference

to stress (e.g. Holmes 1995; Tollfree 1999). Holmes (1995) reports that main stress

triggers the use of glottals more than reduced stress, whereas Tollfree (1999) claims

that “T-glottalisation is optional where the stress on the syllable following /t/ is less

than that borne by the preceding syllable, i.e. in non-foot-initial onset position” as in

guilty (Tollfree 1999: 172).

The stress pattern found in East Anglia, for word-medial /t/, suggests that

unstressed syllables containing /t/ favour glottal(ised) variants, whereas syllables with

primary stress disfavour glottals. This predictor, however, is the least statistically

powerful and the probability difference between primary and non-primary stress is

quite small: the former exhibits a 37 per cent of glottals, whereas the latter 35 per cent.

So far, we have discussed the results of the logistic regression analysis exploring closer

those predictors which proved the null-hypothesis false. Let us now inspect the glottal

224
variation at an individual level to see whether each speaker is participating in the

change. The individual variation was controlled for by including speaker as a random

effect in the mixed-effects model. Figure 6.3.8 displayed a good clustering of young

individuals with respect to word-final /t/; we would expect the same result word-

medially given the leading position of young speakers. However, figure 6.5.8 exhibits

more dispersion in the young and middle age groups than in the old counterpart, where

individuals tend to be clustered. A closer examination reveals that the low rates of

glottals among young speakers are due to a relevant number of tokens where /t/ occurs

at the onset position, such as nineteen. When these words are excluded from the dataset

glottaling rates increase in the young generation.

80
70
60
50
40
30
20
10
0
young middle old

Figure 6.5.8 Percentages of WM /t/ by individual speaker.

Although the factors that appear to constrain word-medial and word-final /t/ are

markedly divergent, the effect size of the preceding and following phonetic segment is

remarkably similar. The statistical effect of social factors, instead, emerged only word-

medially with age and sex conditioning (t).

225
6.5.1 Non-significant factors – word-medial /t/

Social class, the third social factor included in the model, is marked as a non-significant

predictor along with location. Working-class speakers adopt glottals more than their

middle-class counterpart, but not significantly. Similarly, no striking difference was

detected for the three locations. One-level analysis shows that Colchester and Ipswich

favour /t/-glottaling, while Norwich disfavours it, but not significantly. Indeed, the

tables below illustrate the similarity in the glottaling rates among the three localities.

Although location is not a statistically relevant predictor, it is worth exploring in detail

how /t/ behaves word-medially in Colchester, Ipswich and Norwich to inspect the

internal diffusion of this phenomenon across geographical areas.

6.6 Treating Colchester, Ipswich and Norwich separately.

Overall, word-medial /t/ in East Anglia appears to be phonologically and socially

conditioned. By exploring closer the three localities, we will see whether the same

factors surface uniformly.

6.6.1 Results from Colchester – word-medial /t/

Table 6.6.1 below shows results of word-medial /t/ according to the statistically

prominent constraints. The most favouring factors, in order of significance, are the

preceding phonetic segment, the following environment, lexical frequency and sex.

Preceding environment. The influence of the preceding phonetic segment does not

seem to resemble the overall East Anglian pattern. In Colchester, front vowels favour

at .766, stops and fricatives are not very far from being a neutral context favouring /t/-

glottaling at .507, whereas central and back vowels along with nasals disfavour at .452

and .265, respectively.

226
(1) East Anglian pattern
front vowels > central and back vowels > nasals > stops and fricatives

(2) Colchester pattern

front vowels > stops and fricatives > central and back vowels > nasals.

Clearly, the Colchester pattern is not in line with the sonority hierarchy of the overall

East Anglian pattern given the high place of stops and fricatives in the ranking, and

given the low position of nasals, which is the least favouring segment for /t/-glottaling

to occur.

Application value = glottaling; overall proportion = 0.388

R2 = 0.559; log likelihood = -296.004; N = 749
Constraints logodds FW % Tokens
Preceding environment
p<.001
front vowels 1.184 0.766 53 315
stops & fricatives 0.028 0.507 21 34
central & back
vowels -0.194 0.452 28 34
nasals -1.018 0.265 8 70
Following environment
p<.001
stops & fricatives 2.275 0.907 84 25
nasals & liquids -0.28 0.43 60 70
front vowels -0.94 0.281 34 331
central & back
vowels -1.055 0.258 32 152
Lexical frequency
p<.001
high frequency 0.755 0.68 48 424
low frequency -0.755 0.32 14 154
Sex
p<.05
males 0.602 0.646 50 283
females -0.602 0.354 28 295

Table 6.6.1 Multivariate analysis of WM /t/ in Colchester.

Following environment. Following nasals typically highly promote glottals in words

such as witness yet, in Colchester, they favour production of the coronal stop. This
227
might be a result of the re-coding of nasals, grouped together with liquids here, owing

to insufficient number of tokens in the nasal category to carry out an accurate statistical

analysis. Following vowels, in agreement with the overall pattern, disfavour /t/-

glottaling. The exchange of place between front vowels with central and back ones is

not relevant since they all inhibit glottals. The only favouring predictors for the

following environment are stops and fricatives, showing a probability of .907. Typical

findings, indeed, show that following obstruents as in Whitby, Watford enhance the use

of glottal variants.

(3) East Anglian pattern

stops and fricatives > nasals > liquids > central and back vowels > front vowels

(4) Colchester pattern

stops and fricatives > nasals and liquids > front vowels > central and back vowels.

Lexical frequency. The third most favouring predictor is lexical frequency, with high

frequency tokens being subject to glottals more than low frequency ones.

Sex. The least favouring constraint is sex - the only social factor which reached

statistical significance in Colchester, with males being ahead of females in producing

glottal(ised) variants, exhibiting a probability of .646.

6.6.2 Results from Ipswich – word-medial /t/

In Ipswich, the preceding environment, lexical frequency, following context, stress and

age have a significant influence on the response. Table 6.6.2 shows that factor groups

are nearly identical to the East Anglian pattern.

Preceding environment. The preceding phonetic segment reaffirms its robust

conditioning in Ipswich, with preceding vowels promoting the use of glottals, while

preceding consonants favour /t/ retention particularly stops and fricatives, which

228
strongly disfavour /t/ glottaling. This finding counteracts the Colchester pattern where

obstruents are the second most preferred environment for glottals to occur.

(5) East Anglian pattern

front vowels > central and back vowels > nasals > stops and fricatives

(6) Ipswich pattern

front vowels > central and back vowels > nasals > stops and fricatives.

Application value = glottaling; overall proportion = 0.39

R2 = 0.748; log likelihood = -283.850; N = 694
Constraints logodds FW % Tokens
Preceding environment
p<.001
Front 2.053 0.886 56 314
central & back vowels 0.493 0.621 39 196
nasals 0.346 0.586 15 114
stops & fricatives -2.893 0.053 2 70
Lexical frequency
p<.001
high-frequency 0.864 0.696 54 428
low-frequency -0.864 0.304 15 266
Following environment
p<.001
stops & fricatives 1.956 0.876 92 24
nasals & liquids -0.401 0.401 63 63
central & back vowels -0.407 0.4 44 173
front vowels -1.149 0.241 30 434
Age
p<.001
young 1.851 0.864 56 195
middle -0.885 0.292 27 274
old -0.966 0.276 39 225
Stress
p<.001
non-primary 0.48 0.618 44 297
primary -0.48 0.382 35 397

Table 6.6.2 Multivariate analysis of WM /t/ in Ipswich.

Lexical frequency. The second most favouring predictor is word frequency, with high

frequency tokens triggering the use of glottals more than low frequency ones.

229
Following environment. Results from the following phonetic segment are uniform with

the overall pattern despite a marginal difference, as shown below:

(7) East Anglian pattern

stops and fricatives > nasals > /l/ > central and back vowels > front vowels

(8) Ipswich pattern

Stops and fricatives > nasals and liquids32 > central and back vowels > front vowels.

Age. The penultimate significant predictor in this analysis is age - the only social factor

that influences word-medial /t/ in Ipswich - with younger speakers favouring [Ɂ], while

middle and older speakers disfavour it.

Stress. Unlike the findings from Colchester, stress surfaced in the Ipswich regression

analysis yielding the ranking: non-primary > primary.

6.6.3 Results from Norwich – word-medial /t/

By contrast, syllable stress, in Norwich is not marked as a significant factor; this

suggests that glottals occur word-medially regardless of whether /t/ belongs to a

stressed or unstressed syllable. The significant factors that emerged in this multivariate

analysis include the preceding phonetic segment, lexical frequency and the following

phonetic segment, as illustrated in table 6.6.3.

Preceding environment. When stops and fricatives are included in the analysis, they

are the only predictors to be marked as strongly disfavouring factors, as shown in the

table below. However, when obstruents are removed from the analysis front vowels

favour at .67, whilst central and back vowels together with nasals disfavour at .48 and

32
Similar to Colchester, nasals and liquids had to be collapsed even in Ipswich.

230
.35, respectively. In this case, the restriction of glottals to occur before a preceding

schwa (Trudgill, 1974) would still hold for Norwich word-medially.

(9) East Anglian pattern

front vowels > central and back vowels > nasals33 > stops and fricatives

(10) Norwich pattern

front vowels > central and back vowels > nasals > stops and fricatives.

Application value = glottaling; overall pattern = 0.302

R2 = 0.413; log likelihood = -300.502; N = 600
Constraints logodds FW % Tokens
Preceding Env.
p<.001
front vowels 1.351 0.794 40 290
central & back vowels 0.573 0.64 28 186
nasals 0.079 0.52 12 79
stops & fricatives -2.003 0.119 2 45
Lexical frequency
p<.001
high frequency 0.529 0.629 33 437
low frequency -0.529 0.371 11 163
Following Env.
p<.01
stops & fricatives 0.808 0.692 53 28
nasals & liquids 0.247 0.562 49 75
central and back -0.504 0.377 27 123
front vowels -0.552 0.365 26 344

Table 6.6.3 Multivariate analysis of WM /t/ in Norwich.

Lexical frequency. In line with the literature and with the trend which has been

discussed in this chapter, high frequency tokens show a probability of .629, which is

nearly double the size of that of low frequency words (.371).

33
In the overall pattern, nasals were marked as a disfavouring factor.

231
Following environment. Similar to word-final /t/, this phonological context rejects the

null hypothesis at p <.01, whereas in both Colchester and Ipswich levels of significant

are smaller (p<.001). This would suggest that (a) (t) glottaling is completing the change

in this environment in Norwich; (b) Norwich is ahead of Colchester and Ipswich in the

rule application word-medially as also suggested by the lack of social effect.

When zooming into this phonological constraint both nasals and the lateral /l/ promote

glottals in Norwich, despite being grouped together, resembling word-medial findings

from London (Schleef, 2013).34 Following vowels favour the retention of the coronal

stop.

(11) East Anglian pattern

stops and fricatives > nasals > /l/ > central and back vowels > front vowels

(12) Ipswich pattern

stops and fricatives > nasals and /l/ > central and back vowels > front vowels

6.7 Summary

This Chapter has examined word-final and word-medial /t/ in East Anglia.

The results suggest that word-final /t/ has completed its social change and is spreading

in phonological space (e.g. after a preceding schwa where the use of glottals was found

to be previously blocked, at least in Norwich). It appears that glottal variants in this

phonetic context have nearly completed the change in East Anglia and have also spread

to more formal varieties, such as RP. Word-medial /t/ is both phonetically and socially

conditioned with young males being ahead of females. However, in Norwich no social

34
When we looked at the overall results (the three localities together) nasals and /l/ were treated
separately, as the number of tokens in the nasal category was above the statistical threshold of 30.
However, when the dataset was divided according to the three urban areas, nasals and /l/ had to be
grouped together.

232
significance was reported, and the restriction found by Trudgill (1974) still holds before

a following schwa. The latter also holds for Colchester and Ipswich.

Chapter 4 and chapter 6 have investigated two phonological variables: (t,d) in

word-final consonant clusters, and (t) in word-final and word-medial position. (t,d) and

word-final (t) intersect in word-final consonant clusters, yet their intersection appears

to have been ignored so far. Wells (1999a, in answer to question number 12 in the list

of ‘Frequently Asked Questions’), remarks that “no-one can know what will happen in

the future: if the glottal stop is indeed a stage on the route to disappearance [elision]

[…].” In the attempt to provide an answer to the above claim – twenty years later – the

phonological co-variation of (t) deletion and (t) glottaling in word-final consonant

clusters will be examined in the chapter to follow.

233
Chapter 7 - INTERSECTION BETWEEN (t) DELETION
AND (t) GLOTTALING IN WORD-FINAL CONSONANT
CLUSTERS: RESULTS AND DISCUSSION

So far, we have discussed the co-variation of phonological and sociological variables -

one of the main goals of sociolinguistics research. In this Chapter, the focal point will

be the co-variation of the two linguistic variables previously examined: (t) deletion and

(t) glottaling, which intersect in word-final consonant clusters. To start with, I will

provide some terminological remarks, followed by a categorisation of the two

variables, and a brief review of the concept of co-variation between linguistic variables;

I will then turn to the subject of lenition, and phonological rule ordering following with

an outline of analytical procedure; finally, I will present and discuss the overall results

(three localities together) of this intersection analysis.

7.1 Terminological Remarks

Intersection between (t) deletion and (t) glottaling means that in words like can’t, kept,

the non-standard variants of the /t/ may be elided (e.g. [kɑːn]) or may have a glottal

gesture (e.g. [kɑːnɁ])1. To avoid any kind of terminological confusion between (t)

glottaling in word-final context (e.g. what, that) and (t) glottaling in word-final

consonant cluster (e.g. kept, event), the former will be referred to as word-final /t/

glottaling2, whilst the latter will be called (t) glottaling in the C(C)t environment.

This question does not concern the (t,d) variable, in which both /t/ and /d/ occur in

C(C)t and C(C)d positions. This analysis, instead, set out to examine only patterns of

1
The terms intersection and covariation will be used interchangeably in this chapter.
2
See Chapter 5 for further details on the use of this term throughout this survey.

234
variation between (t) deletion and (t) glottaling in the C(C)t environment, even though

the voiced alveolar /d/ can be subject to glottal realisations.

Temple (2014) provides evidence of a small number of tokens where the /d/

undergoes glottalisation in the York (t,d) dataset, as in second-hand shops

[sɛʔ͡n̩ˈhant͡ʔʃɒps], where the first voiced alveolar is deleted, whilst the second is

devoiced and glottalised. Glottal realisation of /d/ is common in other English dialects,

such as AAVE (e.g Fasold, 1972) and Norwich English. With respect to Norwich

English, Trudgill (1974) demonstrates that final -ed can be realised as /-ət/; thus, the

voiceless alveolar is likely to be replaced by the glottal stop as in hundred [hʌndɹəɁ].

There seems not to be reason to consider final (-d) a sociolinguistic variable in

Norwich.

7.2 Status of the two phonological variables in East Anglia

Let us briefly return to the sociolinguistic categorisation of these two phonological

features. We have seen that the status of (t,d), in East Anglia, is that of a stable variable;

whereas the profile of (t) is explained as follows: /t/ glottaling in word-final position is

a change in progress which has reached social completion in many parts of the UK

(Baranowski & Turton 2015), including East Anglia where glottal variants are now

spreading in phonological space; word-medial /t/ glottaling, by contrast, is both

phonetically and socially conditioned.

These two linguistic variables are usually analysed as parallel models: non-

standard /t,d/ deletion vs. /t/ retention, or non-standard /t/ glottaling vs. coronal forms

of /t/. As indicated earlier, it is common practice to code glottal variants along with

apical stops, when examining (t,d) deletion, as the glottal gesture is treated as a

presence, as opposed to deletion – the complete loss of the segment. Even though these

235
two variables intersect in British English, no systematic investigation has been carried

out on their intersection, as Amos et al. (2018) recently observed.

7.2.1 Research questions for intersection

Given that British English allows for more than two alternations in word-final

consonant cluster, as in /kept/ which can be realised as [kɛpt], [kɛɁ], [kɛpɁ], [kɛp]; and

since the results from East Anglia display a slight visible variability of /t/ glottaling in

environments previously stigmatized (e.g. after preceding stops and fricatives3), the

two non-standard features - (t) deletion and (t) glottaling – will be explored in a more

restricted linguistic environment, that is in word-final consonant clusters (e.g. silent,

fault, kept), as the elision of apical stops (i.e. (t,d) absence) typically occurs in this

phonological context. This analysis will be examined by means of a mixed-effects

binary regression analysis to observe patterns of variation (see section 7.7). The

research questions which arise in this regard are summarised as follows:

• Does intersection change the frequency of deletion?

o of glottal forms?

• How should they be properly counted?

7.3 Covariation between linguistic variables

Most of sociolinguistic research deals with covariation between linguistic variables and

social factors. Trudgill (1974: 64) states that “a phonological variable can be defined

as a phonological unit which is involved in co-variation with sociological parameters

or with other linguistic variables.” The traditional variationist research which focused

3
See Tollfree (1999).

236
on linguistic/social covariation has been later expanded including covariation of

explanatory linguistic factors as prime goal (Wolfram 1993; Patrick 1999). In this

Chapter, we will mainly focus on exploring the covariation of (t) deletion and (t)

glottaling in the C(C)t linguistic context.

Covariation can be also explored in terms of spatiality, with isoglosses

mirroring dialect boundaries (Labov et al. 2006); in terms of style shifting of multiple

variables (Rickford & McNair-Knox, 1994), in terms of interspeaker covariation

(Tamminga, 2019), in which speakers are compared across a range of variable features.

Comparison between speakers across phonological variables has been also carried out

in British creole (Patrick, 2004). Some linguists have looked at the intersection between

variables in terms of feeding and bleeding, such as Anttila (2002b).

In a volume devoted to the theme of covariation between linguistic variables,

Guy and Hinskens (2016: 5) state that “the question of whether and how in a given

speech community (or in a coherent sector of a speech community) multiple variable

phenomena are interrelated has received little attention until recently.” Several studies

have addressed this matter. Horvath & Sankoff (1987) investigated 20 vocalic variants

in Sydney English, showing how the distribution of speakers is similar or different in

linguistic space, rather than examining the distribution of linguistic variables in the

social dimension.

Patrick (1999) observed the covariation between phonological and

morphological features by examining (t,d) deletion at the intersection with past

marking in Jamaican Creole. His results from the creole continuum show that the

regular affixation of /-t,-d/ as a past tense marker is not compulsory, that is some

sentences “are systematically ambiguous as to time reference” (Patrick, 1999: 169).

237
This explains the high /t,d/ deletion rate in regular past tense verbs which is due to

morphological absence (see Chapter 3 for further details).

Guy (2013) explored four binary variables among working class speakers of

Popular Brazilian Portuguese. Some of the variables in his work were found to be in

covariation, even though some speakers showed an idiosyncratic behaviour. Guy

(2013) also suggests distinguishing between socially motivated covariation from

interactions driven by structural linguistic correlations among variables, as even if

covariation may be facilitated by structural relationships, sociolinguistic coherence

surfaces separately.

In characterising the expectation of covariation, Guy and Hinskens (2016: 2)

argue that “the orderly variables that define the community should collectively behave

in parallel (i.e. cohere), that is, variants (or rates of use of variants) that index a given

style, status, or a social characteristic should co-occur.” Coherence, in this context,

regards to what degree multiple co-existing linguistic variables show an analogous

distribution. This parallel behaviour springs from the concept of speech communities

being sociolinguistically coherent; that is, speakers who belong to higher classes would

adopt all linguistic features associated with their class status.

However, this issue is a prime theoretical debate in contemporary

sociolinguistics, as Guy and Hinskens’ (2016) argument appears to be at odds with the

claim that speakers actively, idiosyncratically adopt the social signalling of variants in

their communities of practice in order to construct identities, stances, and styles – a

view known as bricolage after Eckert (2008). The issue of at what point such individual

initiatives become community patterns, however, will not be addressed here as it is

beyond the purpose of this research.

238
 One of the questions raised by Guy and Hinskens (2016: 4) is: “Which features

correlate and which do not? To what extent, and in what ways, do the characteristic

variables associated with a dialect or speech community co-vary? Which co-varying

linguistic features / domains are involved in change in progress and which tend to be

constant?” With respect to the social dimension the question asks: “Are there socially

identifiable leaders of change who tend to use all the innovative variants together, or

are different innovations subject to differentiated social interpretations and

individuated patterns of usage?”

As regards Philadelphia vowel changes4, Tamminga (2019) suggests that

interspeaker co-variation springs from a shared social motivation.

The above considerations appear not to be equally applicable to the present

analysis, as studies which have addressed the issue of covariation so far have

investigated multiple variables to observe a potential coherence within the speech

community. In the present Chapter, conversely, we will be dealing with covariation

between two non-standard features. The relationship between (t) deletion and (t)

glottaling in the C(C)t environment, here, will be mainly explored through the lens of

lenition, and through the rule ordering of feeding and bleeding, rather than that of

coherence.5

7.4 A brief account of lenition

The terms ‘lenition’ and ‘weakening’, as outlined in Chapter 5, are interchangeably

used in phonology (e.g. Carr 1993, Hock and Joseph, 1996) and imply a notion of

4
The vowels investigated include the following lexical sets: FACE, PRICE, TOOTH, DOWN, GOAT,
THOUGHT.
5
Guy & and Hinskens (2016) also address the matter of covariation and coherence among different
linguistic systems, raising questions related to language contact.

239
consonantal strength6 (Ashby & Maidment, 2005). According to Lass and Anderson

(1975: 151), “[…] strength is equated with resistance to airflow through the vocal tract,

and weakness with lack of such resistance.” Two current definitions of this process,

reviewed by Honeybone (2008), show that lenition and weakening are used as

synonyms:

“Outside the domain of assimilation in place of articulation, the most

common segmental interaction between consonants and vowels (or,
sometimes, other sonorants) is lenition or weakening. Typical examples of
lenition involve either the voicing of voiceless stops, or the voicing and
spirantisation of stops...” (Odden, 2005: 239)7.

“lenition (also called weakening): consonants can be arranged on scales of

strength.... The scales can be summed-up by saying that a consonant is
stronger the more it differs from vowels; a consonant becomes weaker the
more it comes to resemble a vowel.” (Ashby & Maidment, 2005: 141).

Even though the two terms are indistinguishably used, originally, lenition derives from

the Latin lenire (to soften) and it is not associated with strength or weakness

(Thurneysen, 1898). By lenition Thurneysen (1898: 43) simply means “a decrease in

the intensity of articulation”.

The notion of lenition scale has been extensively debated in theoretical

phonology, yet no proposal seems to cover all stages which lead to the total segment

loss (Honeybone, 2012). Some hierarchies, for instance, do not include the stage of

debuccalisation (e.g. Ewen & van der Hulst, 2001). A compelling proposal, with

respect to coronal segments, has been provided by Harris (1994) who considers

glottaling as a transitional stage to the segment loss:

6
It is argued that consonantal strength can be due to the stress or prosodic prominence of syllables (Ladd
1996); the place of articulation of consonants (Foley, 1977); the manner of articulation of consonants
and voicing (Honeybone, 2008). Six additional types of consonantal strength which have been identified
include “inherent strength, positionally-endowed strength, static comparative strength, strength shown
through dynamic spontaneous change, simple non-inhibitory relative strength, and strength to inhibit
process-innovation.” See Honeybone (2008) for further details.
7
See Honeybone (2008) for issues which arise from this definition, such as linking lenition to inter-
sonorant context.

240
 Plosive > ʔ (Glottaling) > Ø (Deletion).

Kirchner (2004: 3) argues that lenition refers to both diachronic alternations and

synchronic sound change “whereby a sound becomes weaker or where a weaker sound

bears an allophonic relation to a stronger sound”. Despite being a debated issue in the

literature, he deems the characterization of weakening, in relation to consonants, a

“reduction in constriction degree or duration” (Kirchner, 2001: 3). But what type of

changes are usually regarded as lenition? This process includes: degemination, such as

the reduction of a long to a short consonant; flapping, that is the reduction of a stop to

a flap; spirantisation - the reduction from a stop to a fricative or approximant; reduction

of other consonants; debuccalisation – the reduction of an oral to a laryngeal (e.g. t →

Ɂ; s → h); and at its most extreme, complete elision (e.g. t → Ø).

Wide attention has been devoted to the lenition of coronal stops in Present-Day

English, including Irish English, some other British dialects (e.g. Liverpool English),

and American English particularly with respect to ‘tapping’ or ‘flapping’8 whose

application is limited to the intervocalic context v__v, or intersonorant environment

(Honeybone 2012). In Liverpool English, all the underlying stops are affected by

lenition, especially /t/, /k/, and /d/ (Watson, 2007), where the process of affrication and

spirantization are mostly involved. The relationship between t-to-r and t-lenition has

been explored in Liverpool English by Honeybone & Watson (2013), who claim that

the two phonological processes have different characteristics. Intermediate stages

between /t/ and zero were also identified in Dublin English by Hickey (2009b: 400),

who suggests that “lenition can be seen as a scale with the full plosive /t/ at one end

and zero at the other, with identifiable stages in between.” These stages in between are

illustrated as follows:

8
See Carr and Honeybone (2007) for terminological remarks.

241
(1) Vernacular Dublin English (Hickey, 2009b):
t - t̪ - Ɂ - h/r - Ø
button but water water what

In more standard varieties, however, this process is only attested for the first stage.

Indeed, Hickey (2009b) claims that the transitional scale to lenition was not continued

in Supraregional southern Irish English as the latter developed among middle class

speakers who wanted to distance themselves from vernacular Dublin English, by

avoiding glottalisation as an advanced stage of lenition. Therefore, this internal

phonetic development of the lenition process is blocked by social motivations, as

shown in (2):

(2) Supraregional southern Irish English (Hickey, 2009b):

t - t̪
button but

Among those illustrated above, few examples occur in coda position, and none of them

occurs in word-final consonant cluster. This is not entirely surprising since C(C)t

environments are typically excluded from (t) glottaling analysis9.

Honeybone (2012) summarises some types of change regarded as lenition and

the linguistic contexts in which they occur. The first, conditioned changes, can be

influenced by neighbouring segments or other phonological properties; the second,

strongly unconditioned changes, where the phonological environment does not play a

prominent role; the third weakly unconditioned changes, which are not context-free,

but not brought about by the phonological properties of the neighbouring segments. He

9
See Chapter 5 for a detailed discussion as to why tokens in this linguistic context were
typically excluded in previous research.

242
suggests that “lenition involves only those types of change that can show this weakly

unconditioned patterning.” (p. 785).

Despite the existence of intermediate stages in the lenition scale, tapping will not

be an object of examination in the present analysis for a twofold reason: firstly, taps

were not included in the (t) glottaling analysis; secondly, tapping is largely attested in

intervocalic position, or medially in tokens such as winter (Wells, 1982). Hence, in this

Chapter, I will mainly refer to Harris’ (1994) lenition scale where the only intermediate

segment, prior to elision, is the glottal stop. The focus on lenition, here, largely

disregards ‘fortition’10 and ‘strengthening’ as these notions go beyond the purpose of

this survey. Under the light of lenition and the theories of feeding and bleeding, I will

attempt to shed light on the appropriate sequences of rule applications.

7.5 A brief account on rule ordering – feeding and bleeding

Determining the sequences of rule applications was a highly debated issue during the

nineteen seventies. In the early days of Generative Phonology, the assumption was that

rules need to apply in a specific order as part of the grammar of the language.

The major claim, within this aspect of Generative Grammar, is that one rule can

influence the operation of a following one. To explain this process, Nathan (2008)

provides as an example the transition from coronal stops /t,d,n/ to flaps 11 – a process

which is affected by stress. For the next stage - deletion - to occur, there must be a

condition which generates the flap formation since flaps are allophones of other sounds.

This idea of sequential rule ordering was subsequently called into question as some

rules were not structurally connected (i.e. did not have any influence on one another),

10
It is a hot issue in the literature (cf. Buizza 2011b) as to whether affricates account for lenition or
fortition.
11
Flaps undergo deletion in relatively informal speech style.

243
hence it would be onerous to match rules which had a different output. A notable

contribution to this discussion was provided by Kiparsky (1968) who explored the

diachronic rule orderings of languages. His discussion of rule interactions starts by

exploring those cases where the application of a rule lays the basis for a later rule to

apply. This process is referred to as feeding after Kiparsky (1968) as the previous rule

feeds the subsequent one, that is “if Rule A increases the numbers of forms to which

Rule B can apply, the order A – B is a feeding order” (Gussenhoven & Jacobs, 2011:

114). Besides the flapping and flap deletion phenomena above mentioned, another

example of feeding ordering is the rule of fortis plosive insertion which feeds pre-

glottalization in British English (Gussenhoven & Jacobs, 2011), is illustrated in the

examples below:

Underlying Fortis stop Pre-glottalization

prins prints prinɁts prince
leŋθ leŋkθ eŋɁkθ length

Conversely, if a previous rule creates a context which prevents the following

rule from applying, the rules are claimed to be in bleeding order, as the first rule bled

the second one. In other words, “if Rule A decreases the number of the forms to which

Rule B can apply, the order A – B is in bleeding order” (Gussenhoven & Jacobs, 2011:

115). An example of bleeding order, as reviewed by Nathan (2008) in his commentary

on Kiparsky’s (1968) discussion, is ‘l-darkening’; the allophone of /l/ for numerous

speakers is [ɬ] even foot-internally or syllable finally in words like velar. However, if

the suffix -ity is attached to velar, the stress shifts to the next syllable resulting in a

syllable-initial clear [l]. Therefore, in this case, the stress shift blocked the application

of l-darkening. Gussenhoven & Jacobs (2011) illustrates the bleeding ordering through

the ɪ-insertion rule between the voiceless alveolar fricative /s/ and its voiced counterpart

– the plural marker /z/. The English rule related to the devoicing of segments when

244
these are followed by voiceless consonants, as in books [bʊks], does not apply when /ɪ/

is inserted, as in buses [bʌsɪz].

From a maximal rule transparency viewpoint, both feeding and bleeding are

treated as natural orders due to their application transparency on the surface. Having

reviewed the main points of rule ordering, let us briefly outline the analytical

procedures employed before turning the attention to the logistic regression analysis.

7.6 Analytical procedure

The total number of tokens which intersect (i.e. occur for both variables) equals 1,275

- a lower amount than the totals of 4,879 (t, d) and 3,051 for word-final (t), due to the

exclusion of C(C)d clusters. In detail, an average of 35.4 occurrences per speaker was

analysed – a number which (a) still conforms to the general statistical law, and (b)

reaches the ideal of 30 tokens per environment (Erickson and Nosanchuk 1992). In this

regard, the present analysis will be only carried out to account for East Anglia as a

whole (i.e. by grouping the three localities together) in order not to lower the suggested

threshold per environment. Together with glottal replacement with [ʔ], the dataset

includes the few cases of glottal reinforcement of [t] with [tʔ] (n = 2) or [ʔt] (n = 12),

and the few cases where a period of creaky voice occurred (n = 4).

Commonly, the tool of choice to analyse binary outcomes is logistic regression.

However, when a sociolinguistic variable has more than two alterations, multinomial

logistic regression seems the appropriate tool. In case the variants are related in an

ordinal way, ordinal logistic regression is employed. The drawback of running

multinomial models, however, is that we lose the precision we would obtain from the

full mixed-effects model. In this respect, Gorman & Johnson (2013: 226) state that “we

are unaware of any software that fully supports mixed-effects multinomial models.”

245
Since speakers in the sample produced many tokens, it would be ideal to run a mixed-

effects model with a per-subject intercept. Hence, since the previous chapters examined

/t,d/ vs. deletion, and /t/ vs. glottaling, in this last chapter I will look at /t/ deletion vs.

/t/ glottaling, as a result of parallel models.

Deletion has been selected as application value in the mixed-effects regression

analysis following the order of the lenition scale proposed by Harris (1994) - plosive

> ʔ (glottaling) > Ø (deletion) - according to which /t/ glottaling is closer than /t/

deletion to what is considered the ‘standard’12. Treating word-final /t/ glottaling (e.g.

it, get) as a realisation close to the ‘standard’ is not surprising since this feature (before

a consonant) was found to be well-established even in RP (Kerswill 2007; Barrera

2015). Moreover, Fabricius (2000: 147) suggests:

“As a recommendation for foreign language teaching then, it seems

reasonable to describe t-glottalling as an emerging standard pronunciation
in word-final environments.”

The computational formula employed for this intersection analysis resembles what is

usually referred to as “Labov Deletion” (Rickford et al., 1991: 106), and commonly

used in the study of the AAVE copula be:

D D
C + D.

In the above formula, D stands for deletion (e.g. "He Ø talkin), whilst C stands for

contraction (e.g I'm here). Full forms (e.g. she will be here tomorrow; she was here

yesterday) are not included in the formula (cf. computational formula “Straight

Deletion” (Rickford et al., 1991: 106)).

Following the above, the computational formula adapted to explore the intersection of

(t) deletion and (t) glottaling in the C(C)t environment is:

12
In this analysis, treating /t/ glottaling as ‘more close to the standard’ does not mean that it is more
standard that /t/ deletion which never had the same level of stigma attached.

246
 D
G+D

where D represents deletion and G represents glottal(ised) forms. Before attempting to

answer the above research questions, let us briefly review what is meant by covariation

between multiple linguistic phenomena, and how linguists are currently dealing with

this issue.

Due to the limited amount of occurrences, in comparison with the main analyses

of (t,d) deletion and (t) glottaling, several factor groups had to be collapsed to avoid

knock-outs. This prevents us from providing an in-depth account of the linguistic

constraints examined, and it is likely that some differences will be concealed.

Predictors whose factor weight were the same were collapsed, such as following

unstressed syllables and pause, following stops and pause, as well as following liquids

and glides. The constraints and their related factor groups included in the statistical best

fit model are illustrated in table 7.6.1.13

13
Note that morphological class, with its related factor groups: monomorpheme, semiweak, and regular
past tense, is not included in the intersection analysis as this constraint has not been explored in relation
to (t) glottaling, neither in the (t) analysis of the present survey nor in previous research on (t) glottaling.
Hence, comparing a constraint between two non-standard variables without knowing how it behaves in
the (t) vs. /t/ analysis, might lead to misleading results.

247
 Constraints Factors

Preceding manner of articulation nasals (e.g. different)

fricatives (e.g. left)
/l/ (e.g. built)
stops (e.g. asked)

Following manner of articulation nasal (e.g. last month)

fricatives (e.g. can’t help)
stops (e.g. must be) + pause (e.g. she
stopped?)
liquids (e.g. don’t like) + glides (e.g.
Walt was)
vowels (e.g. went on)

Syllable stress (on the cluster) Primary stress – /t/ occurs in primary
stressed syllables (e.g. cost)
Non-primary (e.g. different)
Stress on the following syllable Unstressed (e.g. past eleven) + pause
(e.g. agreement)
Stressed (e.g. best way)
Voicing Agreement Homovoiced (e.g. contact)
Heterovoiced (e.g. parent)
Style Spontaneous speech, reading styles,
word lists
Word frequency low frequency (1-3); high frequency (4-
7)
Social class working class, middle class

Age young (18-28), middle (35-50), older

(60+)
Sex female, male

Table 7.6.1 Constraints on the intersection between (t) deletion and (t) glottaling.

To my knowledge, the intersection between (t) deletion and (t) glottaling has

not been explored prior to this study, therefore this will prevent us from making

comparisons with previous research. This discussion, therefore, will be limited to

comparison, where possible, with the findings discussed earlier.

248
7.7 Overall Results and Individual Variation

This section presents the step-up/step-down regression analysis of the above model and

reports the findings for the three locations together. The best model achieved in the

multivariate analysis shows that preceding environment, following environment, stress

on the following syllable, style, sex, and syllable stress (on the cluster) are statistically

significant. In the main two analyses of (t,d) deletion and (t) glottaling (e.g. street), as

discussed in Chapters 4 and 6, none of the social factors surfaced as significant.

However, when (t) deletion and (t) glottaling in the C(C)t environment are examined

at their intersection, sex emerges as a significant predictor. Further details on the effect

of sex will be provided in section 7.7.5. Even though the multivariate analysis does

not include the standard [t], figure 7.7.1 illustrates the distribution of /t/ in word-final

consonant clusters across the whole lenition scale, in the East Anglian dataset.

Figure 7.7.1 Distribution of /t/ in word-final consonant clusters.

The sections to follow will be devoted to the predictors which surfaced in the logistic

regression analysis (deletion vs. glottaling) in order of statistical significance.

249
 Application value = deletion; overall proportion = 0.725
R2 = 0.532; log likelihood = -580.158; N = 1275

Constraints logodds FW % Tokens

Preceding
environment
p<.001
stops 1.957 0.876 96 47
fricatives 0.937 0.719 89 346
nasals -0.382 0.406 68 814
/l/ -2.512 0.075 29 68
Following
environment
p<.001
nasals 0.589 0.643 92 90
fricatives 0.500 0.622 82 244
pause + stops 0.250 0.562 71 489
vowels -0.604 0.353 68 264
liquids and glides -0.735 0.324 65 188
Stress on
following syllable
p<.001
unstressed +
pause 0.542 0.632 77 949
stressed -0.542 0.368 59 326
Style
p<.001
reading styles 0.688 0.666 75 247
spontaneous
speech 0.443 0.609 73 983
word lists -1.131 0.244 49 45
Sex
p<.05
males 0.597 0.645 75 664
females -0.597 0.355 70 611
Syllable stress
p<.05
unstressed 0.195 0.548 70 326
stressed -0.195 0.452 74 949

Table 7.7.1 Multivariate analysis of /t/ deletion vs. /t/ glottaling.

250
7.7.1 Preceding phonetic segment

The most robust predictor in the intersection between (t) deletion and (t) glottaling in

the C(C)t linguistic context is the preceding phonetic environment, with preceding

stops (.88) and fricatives (.72) favouring deletion, while preceding nasal and preceding

/l/ disfavour at .41 and .07, respectively14. This means that, in word-final consonant

clusters, both nasals and /l/ favour (t) glottaling.

0.9
0.8
probability of deletion

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
stops fricatives nasals /l/

Figure 7.7.2 Probability of deletion according to the preceding phonetic segment.

Interestingly, the preceding environment is the first most weighty constraint bearing a

resemblance to the main analysis of glottaling in word-final position (e.g. hate), as

discussed in Chapter 6. Typically, in binomial models for both (t,d) deletion and word-

final /t/ glottaling – where non-standard versus standard features are compared – the

preceding linguistic context is not a very powerful predictor: for (t,d), it is considered

a “tertiary constraint” (Guy, 1980: 20); for word-final /t/ glottaling, the preceding

environment is left unexplored in many British dialects. This finding for East Anglia is

thus unprecedented.

14
It should be highlighted that when preceding /l/, a strongly disfavouring predictor, is excluded from
the analysis, stops are the only factor which favours deletion; whereas fricatives and nasals favour /t/
glottaling.

251
 Rice (1992) suggested to also take into account the phenomenon of sonority

when discussing phonetic weakening, or lenition. Indeed, sonority seems to be an

explanatory factor when moving forward to the lenition scale. The results reveal that

less sonorous preceding segments favour deletion,15 whilst more sonorous segments

favour glottal variants. These findings resemble the trend of word-final /t/ glottaling

(e.g. habit) even in the behaviour of factor groups, with preceding nasals and /l/

triggering the use of the glottal variant, while fricatives and stops disfavoured it.

When examining the distribution of /t/ between the preceding phonetic segment

and sex of participants, the trend of men and women goes in the same direction, as

illustrated in figure 7.7.3. However, men delete more than women when /t/ follows the

lateral /l/, nasals and stops; while women exhibit a greater deletion rate when /t/ follows

fricatives.

100

80
% deletion

60

40

20

0
stops fricatives nasals /l/

males females

Figure 7.7.3 Deletion by preceding phonetic segment and sex, p <.001.

Before attempting to provide an explanation, let us make some considerations referring

back to Harris’ (1994) lenition scale (plosive > ʔ > Ø). Deletion represents an advanced

weakening stage and glottaling is located in an intermediate position being closer to

15
This finding is in line with previous North American (t,d) studies (e.g. Santa Ana, 1996), however
note that Santa Ana (1996) examined /t,d/ deletion over standard /t/, thus the comparison with the
above results is not entirely the same.

252
the ‘standard’ /t/ 16. Word-final /t/ glottaling (before a consonant) is now attested even

in RP (Kerswill 2007; Barrera 2015), and the lack of social effect found in recent word-

final /t/ research (e.g. Holmes-Elliott 2019; Chapter 6 of the present study) appears to

add to the ‘loss of stigma’ argument. This leads us to suggest that, if the glottal stop is

losing its stigma in word-final /t/ (e.g. opposite), it should not be surprising to find men

favouring deletion in the C(C)t environment – the stage of lenition.

7.7.2 Following phonetic segment

The second most significant predictor is the following phonetic environment, with

nasals (.64), fricatives (.62), stops and pauses (.56) favouring deletion, while vowels

(.35), liquids and glides (.32) disfavour it. Since the probability values for pause and

stops were the same, the two factor groups were collapsed into one category. Similarly,

liquids and glides were grouped together due to probability similitude.

0.9
0.8
0.7
Probability of deletion

0.6
0.5
0.4
0.3
0.2
0.1
0

stops+pause fricatives nasals approximants

Figure 7.7.4 Probability of deletion according to the following phonetic segment.

16
This lenition hierarchy, however, does not seem to be universal as there are dialects of English which
show T-deletion but not T-glottaling. In those cases, the glottal stop cannot be considered an intermediate
stage between /t/ and zero.

253
Figure 7.7.4 shows that the probability of deletion slightly changes when running a

binomial model with two non-standard variables. Indeed, while the behaviour of nasals,

stops, vowels, and glides resembles (t,d) results found for East Anglia, the behaviour

of following pause differs. In Chapter 4, we saw that following pause is marked as a

disfavouring factor in the three localities, so /t,d/ is more likely to be retained 17. When

/d/ is left out of the dataset, and when deletion is compared over glottaling, pause seems

to favour /t/ reduction. The high rate of deletion before fricatives is not surprising since

they were strongly marked as a disfavouring factor in the (t) analysis (see Chapter 6).

Sonority does not play a key role in this linguistic environment, as nasals, the most

favouring predictor, are followed by less sonorous segments such as fricatives and

stops whose probabilities of deletion are slightly lower.

Crosstabulation between the following phonetic environment and sex, as in

7.7.5, shows that women are very slightly ahead of males in the deletion of /t/ before

following nasals; while in every other environment males are leading. If compared to

figure 7.7.2 we can observe that deletion is led by males in the same factor groups

(stops and liquids) in both preceding and following environments.

100
80
% deletion

60
40
males
20 females
0

Figure 7.7.5 Deletion by following phonetic segment and sex, p < .001.

17
Following pause is also a disfavouring factor for /t/ glottaling, see Chapter 6.

254
7.7.3 Stress on the following syllable

The third most significant predictor which surfaced in the mixed-effects logistic

regression is stress on the following syllable. This predictor has been included in the

present analysis following Rice’s (1992) suggestion to consider sonority when taking

lenition into account18. By way of theoretical background, sonority is argued to be

influenced by prosodic factors (de Lacy, 2007), hence, the inclusion of this predictor is

used as a means of explaining the role of sonority in this weakening process. The

sonority scale I will refer to is the following, after Clements (1990):

VOWELS > GLIDES > LIQUIDS > NASALS > OBSTRUENTS

The view of treating the sonority scale as universal is at odds with recent

alternatives advocated by Prince (2001) and de Lacy (2006) who suggest to “avoid

positing universally fixed ranking.” A more gradient approach proposes that the

sonority of sounds can slightly oscillate depending on the syllable position they are in;

or it can differ based on the physiological properties of the speakers, such as intensity,

duration, etc. (Parker, 2002). Since, however, sonority is not crucial in this work, I will

not discuss this point further. In Chapter 3 we saw that sonority governs (t,d) in

numerous North American studies (e.g. Santa Ana, 1996), yet in East Anglia and other

British dialects sonority is not an explanatory factor for word-final /t,d/ reduction.

18
What is usually referred to as ‘strength’, as opposed to a weak sound, is also simply called sonority.

255
 unstressed + pause
stressed

0 0.2 0.4 0.6 0.8

Figure 7.7.6 Probability of deletion by stress on the following syllable.

Turning back to the intersection, figure 7.7.6 reveals that following unstressed syllables

and pause are more likely to favour (t) glottaling, whereas following stressed syllables

trigger (t) deletion. Among the unstressed syllables a high number of tokens (N = 22)

is represented by following schwa (e.g. passed away) which, being a vowel, is the first

most sonorous feature of the sonority scale. Hence, the more sonorous the following

segment, the less likely is deletion to occur. This finding is not surprising, as following

vowels are typically marked as a disfavouring factor for (t,d) due to resyllabification

processes, as discussed in Chapter 4. In this case, instead of referring to the

resyllabification of /t/ onto the following vowel since the voiced plosive is left out of

the analysis, the use of the glottal stop might suggest an approach to increase the

difference in terms of sonority between syllable coda (less sonorous than vowels) and

the syllable onset19.

Following syllables which hold a prominent stress are mainly fricative-initial

(N = 177/614)20 followed by stops (N= 151/614). This finding, linked to the sonority

19
A similar explanation was provided by Fuchs (2015), who explored word-initial glottal stop insertion
in V#ɁV and C#ɁV positions, to explain the use of glottal stop insertion after sonorants.
20
This shows the highest number of tokens among following stressed syllables where /t/ deletion occurs.

256
hierarchy, reveals that less sonorous following segments trigger /t/ deletion. Overall, it

seems that the second stage of lenition – (t) glottaling – occurs if the following segment

is not stressed; whereas the last weakening stage – deletion – takes place when followed

by stressed syllables.

7.7.4 Style-shifting

The fourth most favoring predictor which surfaced in the mixed-effects regression

analysis is style. Reading styles and spontaneous speech favor deletion at 0.67 and 0.60,

respectively; whereas word lists disfavour at 0.24. This implies that, when deletion is

examined against glottaling, a higher use of glottal stops occurs when words are

realized in isolation. Considering the increase of glottals across speech styles in the

binary analysis of [Ɂ] vs. /t/ (e.g. forget), as discussed in Chapter 6, it is not unexpected

to find a relatively high rate of deletion in spontaneous speech and reading styles in

word-final consonant cluster, where Ø vs. [Ɂ] are examined at their intersection. This

suggests that the speakers might be advancing towards the lenition scale, moving from

[Ɂ] to zero Ø in the C(C)t context.

100
80
% deletion

60
40
20
0
informal reading word lists
passages

Figure 7.7.7 Probability of deletion in the C(C)t linguistic context, across style.
.

257
Along this line, it should not be surprising to find word lists disfavouring deletion (and

thus favouring glottaling). Indeed, in the (t) analysis discussed in chapter 6, /t/ is

typically realised as an apical stop in words in isolation; therefore, when moving

towards the lenition scale, the underlying /t/ is expected to be realised with glottal

variants first – the second stage of lenition – before being eventually deleted.

The influence of syllable stress on the word-final cluster is discussed below in

7.7.6, as it is the least significant predictor.

7.7.5 Sex

The fourth statistically significant predictor, and probably the most salient as a

sociolinguistic explanatory factor, is sex. Figure 7.7.8 illustrates that females adopt

glottal variants more than males, whilst males delete /t/ more than females.

males females

Figure 7.7.8 Probability of deletion by sex of participants.

If deletion is regarded as the most advanced stage of lenition, if the segment loss is

treated as a stage weaker than glottaling, and if glottaling is well-established even in

RP in word-final position, it is not surprising to find males performing more lenition.

Another way of saying this, however, is that for the variable which is stable males

delete more, whereas for the variable which is a change in progress women are leading

by glottalling more. This finding confirms results from chapter 4, with males being in

258
the right direction and leading in the deletion of /t,d/21, whilst results from chapter 6

showed that males glottal more than women when /t/ occurs word-finally (i.e. /t/ after

a preceding vowel (e.g. that) and /t/ after a preceding consonant (e.g. silent)). This

suggests that women are leaders in the use of glottal variants in a more specific

environment, that is when /t/ occurs in word-final consonant cluster (e.g. silent). A

closer inspection shows that this trend also holds across different speech styles, as in

figure 7.7.9.

100

80
% deletion

60
males
40 females

20

0
informal speech reading passages word lists

Figure 7.7.9 Rates of deletion by style and sex.

7.7.6 Syllable stress

The least statistically significant predictor is syllable stress on the word-final cluster.

This constraint was included in the model as English is a stress-based language and

“East Anglian dialects of English show greater stress-effects than most others”

(Trudgill, 2018). Viewed through the lens of lenition, coronal /t/ is more likely to

undergo deletion in unstressed clusters, whereas in stressed ones /t/ is subject to

glottaling.

21
However, sex was not marked as a significant predictor for (t,d).

259
 stressed
unstressed

0 0.1 0.2 0.3 0.4 0.5 0.6

Figure 7.7.10 Probability of deletion by syllable stress.

In line with phonological theory, Schiering (2006, cited in Trudgill 2018) claims that

stress-based phonologies exhibit a “strong erosive force in reducing and deleting

unstressed syllables.”

7.8 Summary

This Chapter has attempted to examine the intersection between (t) deletion and (t)

glottaling in the C(C)t environment. Overall, results show that the frequency of deletion

remains reasonably stable if results are compared to findings for the complete set of

(t,d) forms (see Chapter 4). The influence on the variation of underlying /t/ through the

lenition scale is linguistically driven, suggesting that (t) glottaling and (t) deletion in

word-final consonant cluster are in feeding order, i.e. glottalling can lead to deletion22.

Sex is the only statistically significant social predictor: when looking at the lenition

scale women seem to prefer the use of the glottal stop, whereas in men’s speech /t/

undergoes more deletion. The age of participants, social class, voicing agreement and

word-frequency do not play a relevant role. Numerous questions, however, remain

open to discussion (e.g. how should frequencies of deletion and glottaling in the C(C)t

context be properly counted?).

22
This does not apply to all English dialects as, for some dialects that have both (t) deletion and (t)
glottaling, the diachronic ordering seems to be T-deletion > T-glottaling.

260
Chapter 8 - CONCLUDING REMARKS AND FUTURE
WORK
This research has focused on two well-studied phonological features: (t,d) deletion and

(t) glottaling by preserving the structuralist roots of the variationist paradigm (see

Wolfram 1993; Patrick 1999). Along this line, this survey has treated covariation of

explanatory linguistic factors as a prime goal and has shed light on the intersection

between two linguistic variables: (t) deletion and (t) glottaling in word-final consonant

clusters. This chapter brings together the findings of the target phonological variables,

summarises their sociolinguistic status and proposes directions for future work.

Overall results suggest that the profile of (t,d), in East Anglia, is that of a stable

variable as none of the social factors examined (social class, age and sex) reached

statistical significance. However, when the three communities were investigated

separately, the only social predictor which surfaced in the multiple logistic regression

analysis is sex in Norwich. The fact that age is not significant in any of the three

communities suggests that (t,d) has probably been stable over years, at least in apparent

time.

Conversely, in some northern English varieties social factors play a remarkable

role such as in Tyneside English (social class, age and sex). In this variety (t,d) is

advancing through the life cycle of phonological processes resulting in the incoming

stem-level progression. This suggests that BrE dialects are behind US dialects in terms

of stem-level – a result which could explain the lack of morphological effect in York.

In East Anglia, rates of deletion are lower than Manchester and even lower than York,

261
except for monomorphemes, yet there is no evidence of change in progress. Its status

of a stable variable, in East Anglia, aligns with previous US English studies.

Unlike previous research carried out on (t,d), the present study removes all

actually-glottalled/glottalized tokens from the (t,d) database. This survey has also shed

light on the unsolved problem of morphological effect in British English dialects (I

refer specifically to those studies which treated (t,d) as one variable, such as York,

Manchester and Tyneside English) reporting the emergence of the expected

morphological effect for (t,d) in East Anglia, with n’t negative contractions and

monomorphemes favouring deletion, whereas semi-weak verbs and regular past tense

verbs disfavour it. The strong morphological effect is proven by the consistency of the

above findings in all three speech communities.

With respect to phonological constraints, I have proposed a closer inspection

into the following phonetic segment whose robust effect in British English mirrors that

of previous studies. The present survey has shown that the classical coding (obstruents

> glides > /r/> /l/ > vowels > pause) is not sufficient to account for this universal

constraint, and has revealed salient phonological differences which surfaced after

breaking down the obstruent category. It is likely that these differences, as summarised

below, might have been obscured in previous research: (a) sibilants and non-sibilant

fricatives in the following environment were found to behave very differently - the

former strongly favours deletion, whereas the latter strongly disfavours it; (b) following

initial /h/ plays a prominent role as (t,d) is highly retained whether following initial /h/

is dropped or not. This finding helps to explain why non-sibilant fricatives strongly

disfavour deletion, and points the way to future investigation on (t,d) in English dialects

where following /h/ is either frequently retained or frequently dropped.

262
 Some predictors still deserve a closer examination, such as /t,d/ after a

preceding /l/. In this study, /l/ was coded as a lateral when it was phonetically

consonantal, yet when /l/ was vocalised it was coded in the vowel category. In future

research, it would be interesting to compare the rates of (t,d) deletion after a preceding

consonantal /l/ and rates of (t,d) absence after a preceding underlying /l/ which

undergoes vocalisation to observe any potential difference.

From a theoretical viewpoint, Lexical Phonology does not provide an

explanation which holds for these data (the probability of retention for

monomorphemes is comparatively lower than that of regular verbs), and the Competing

Grammars approach revealed that (t,d) does not appear to vary due to morphological

absence in semi-weak verbs. The OCP, according to which the more features of

phonetic context are shared with /t, d/, the more likely deletion is to occur (Guy &

Boberg 1997), does not hold for the East Anglian data either. Sonority, whose

hypothesis is that less sonorous preceding segments favour deletion, appears not to be

an explanatory factor in East Anglia as nasals occur at the top of the ranking in the

preceding phonetic environment. Word-frequency did not surface as a significant

predictor.

The second target linguistic variable was (t) glottaling which has been

examined in both word-final and word-medial positions. Nearly 20 years ago, Milroy

et al. (1994) argued that linguists do not seem to have an accurate idea of the main

constraints that govern this variable. 20 years later, Schleef (2013: 202) claimed that

what was previously described by Milroy et al. (1994) “has barely changed.” Since this

variable has been greatly explored in relation to the following phonological

environment – commonly divided into three main phonological contexts: PreC (e.g.

that man), PreV (e.g. that apple) and PreP (e.g. what?) (e.g. Williams & Kerswill 1999;

263
Straw & Patrick 2007), the present survey has contributed to a closer examination of

this constraint by providing a finer distinction of the above classical coding. Moreover,

this study has examined the role of the preceding phonological environment, which has

received little attention in the (t) literature so far.

Overall results showed that word-final /t/ glottaling has completed its social

change in East Anglia, indeed none of the social factors was found to be statistically

significant. This finding resembles recent research carried out in the UK (e.g. Holmes-

Elliott, 2019), and appears to add to the general ‘loss of stigma’ argument. The largely

ignored preceding phonological environment turned out to be the most powerful

constraint. Glottaling in post-sonorant position (e.g. bolt, ant) did not occur

categorically, indeed overall results reported a rate of 74% and 57% for /t/ post-nasals

and post-laterals, respectively – a finding which is in contrast with the high rates of

glottaling found in Manchester in the same contexts (see Baranowski & Turton, 2020).

Findings from East Anglia suggest that word-final /t/ glottaling is diffusing in

phonological space even in environments where it used to be blocked. The role of

frequency, even in this case, was not statistically significant.

A direction for future research relates to word-frequency. The latter was not

marked as a significant predictor in the present study, yet recent research (Purse &

Tamminga, 2019) suggests that different measurements of frequency (e.g. the

frequency of the root) has a significant statistical effect for some linguistic variables.

Besides testing the effect of root-frequency, it would be also interesting to explore

differences and correlations between the low/high frequency of root and the low/high

frequency of suffix in East Anglian English.

Whole-word frequency, however, was found to be an explanatory factor for word-

medial /t/ glottaling – a change in progress which has not reached social completion

264
yet in East Anglia. Indeed, sex and age of participants were marked as significant

predictors. Given the above results linked to word-frequency, could there be a potential

relationship between whole-word frequency and variables which are still progressing

in the change? Word-medial /t/, in East Anglia is less advanced and it is spreading both

socially and linguistically. Overall results showed that young males are ahead of

females and that word-medial /t/ is also spreading after a preceding schwa – an

environment previously blocked, at least in Norwich. Syllable stress was marked as the

least predictor to condition /t/ glottaling word-medially, with non-primary syllable

stress (e.g. community) favouring the non-standard variant and primary syllable stress

disfavouring it (e.g. Westminster). The sonority hierarchy is an explanatory factor as

more sonorous preceding segments were found to trigger the glottal stop. In terms of

spatiality, Norwich appears to be ahead of Colchester and Ipswich in the word-medial

rule application, as also indicated by the lack of social effect.

The last part of this work has focused on the covariation between (t) deletion

and (t) glottaling in the C(C)t linguistic context to explore the interplay between

multiple variable phenomena in the three speech communities. Overall results showed

that when moving towards the lenition scale (see Harris, 1994) linguistic factors play a

remarkable role, with less sonorous preceding segments favouring deletion, whereas

more sonorous ones favour (t) glottaling.

With respect to style-shifting, it appears that the speakers are advancing towards the

lenition scale moving from [Ɂ] to zero Ø, showing that (t) glottaling and (t) deletion are

in feeding order. Sex is the only social factor which reached statistical significance with

males being ahead of females in favouring deletion – the last stage of the lenition scale.

Future research on the intersection of these two non-standard features, in the C(C)t

context, could consider applying alternative methods to compute the incidence of

265
deletion. Since the computational formula employed in the present survey has been

adapted from what is commonly referred to as “Labov Deletion” (Rickford et al., 1991:

106), it would be interesting to observe whether the application of the “Straight

Deletion” formula (Rickford et al., 1991: 106) could affect the results. The “Straight

Deletion” formula, adopted to examine the variation of copula be in AAVE, includes

the full form of the copula be variable (e.g. was); whereas in the present survey the

label “full form” could be intended as the first stage of the lenition scale (i.e. the

realisation of the standard plosive [t]). Rickford et al. (1991) show that the application

of different computational formula can affect the overall outcome and, along the line

of their results, it would be no surprise if the overall deletion rate was lower when

employing the “Straight Deletion” method.

Romaine (1984: 228) suggests that “by looking at the way in which variants distribute

themselves synchronically in the social structure of a speech community and

understanding the social meaning that are attached to them, we get some idea of relative

chronology and directionality.” When taking into account the directionality of a

variable in time, social and linguistic space, I would suggest considering whether it

intersects with other linguistic variables. The examination of both linguistic and social

patterns involved in the intersection will provide a thorough understanding of the

concepts of stability vs. change in progress and could be also valuable for

sociolinguistic reconstruction tasks. The present survey, unusual in highlighting

intersection, suggests that stable variables like (t,d) are only stable when they do not

intersect with ongoing changes - including late-stage changes like word-final /t/

glottaling. Intersection analyses also add to our understanding of (a) reactivation of

change and (b) completion of change. It is a core principle in variationist

sociolinguistics to distinguish language change from variation. A prerequisite of

266
language change is variation, but some variation involves no change and can be stable

over time. Social distribution can mark the endpoint of a change, that is, absence of

variation by age, gender, social class, neighbourhood or ethnicity implies the

completion of a change (Labov, 2010). In the case of (t) glottaling and (t,d) deletion

in the UK, Milroy et al. (1994) argue that, in Newcastle, the glottal variant is a supra-

local change and women are instrumental in the diffusion of these non-standard

variants (for similar results in Sunderland see Burbano-Elizondo, 2015); Woolford

(2018) shows that, in Tyneside English, (t,d) is advancing through the life cycle of

phonological processes, as opposed to East Anglia where (t,d) is a stable variable and

word-final /t/ glottaling is a late-stage change. This suggests that when glottal forms

began to spread in some areas of the UK, (t,d) became ‘changing’ again (e.g.

reactivation of change); as glottaling goes to completion, (t,d) becomes stable (e.g.

completion of change)

267
 REFERENCES

Abercrombie, D. (1948) Forgotten Phoneticians. In D. Abercrombie, Studies in Phonetics and

Linguistics. Oxford: Oxford University Press.

Akers, G. (1977). Phonological Variation in the Jamaican Continuum. PhD dissertation,

Harvard University.

Akers, G. (1981). Phonological Variation in the Jamaican Continuum. Ann Arbor MI:

Karoma.

Altendorf, U. (2003). Estuary English. Levelling at the interface of RP and south-eastern

British English. Tubingen: Gunter Narr Verlag.

Amos, J., Johnson W. & Kasstan J. (forthcoming). Reconsidering the variable context: A

phonological argument for (t) and (d) deletion.

Andrésen, B. S. (1968). Preglottalisation in English standard pronunciation. Norwegian

Studies in English No 13. Oslo: Norwegian Universities Press.

Anttila, A. (2002). “Variation and phonological theory.” In Chambers, J.K., Trudgill, P. and

Schilling-Estes, N. (eds.) The Handbook of Language Variation and Change, (pp. 206-

243). Blackwell, Malden, Massachusetts, and Oxford, U.K.

Ash, S. (2002). “Social Class.” In Chambers, J.K., Trudgill, P. and Schilling-Estes, N. (eds.)

The Handbook of Language Variation and Change. (pp. 402-422). Oxford: Blackwell.

Ashby, M. and Maidment, J. (2005). Introducing Phonetic Science. Cambridge: Cambridge

University Press.

Baggs, A. P., Board, B., Crummy, P., Dove, C., Durgan, S., Goose, N. R., Pugh, R. B. Studd,

P. and Thornton, C. C. “Modern Colchester: Town development”, in A History of the

268
 County of Essex: Volume 9, the Borough of Colchester, Janet Cooper and C R Elrington

(London, 1994), pp. 199-208. British History Online http://www.british-

history.ac.uk/vch/essex/vol9/pp199-208 [accessed 20 December 2019].

Baranowski, M. & Turton, D. (2015). “Manchester English”. In Hickey, R. (eds) Researching

Northern English. (pp. 293-316). Philadelphia: John Benjamins.

Baranowski, M. & Turton, D. (2016). The morphological effect in British English t/d-deletion.

Paper presented at LAGB 6th -9th September.

Baranowski, M., Bermúdez-Otero, R., Bailey, G. & Turton, D. (2016). Ahead but not faster:

the effect of high token frequency on sound change. Paper presented at NWAV-45, 4th

November.

Baranowski, M. & Turton, D. (2020). “TD-deletion in British English: New evidence for the

long-lost morphological effect.” Language Variation and Change 31, 1-23.

Barrera, B. B. (2015). A Sociolinguistic Study of T-glottalling in Young RP: Accent, Class and

Education. Unpublished PhD thesis, University of Essex.

Batterham, M. (2000). “The apparent merger of the front centering diphthongs – EAR and AIR

– in New Zealand English”. In Bell, A. and Kuiper, K. (eds.) New Zealand English.

Wellington: Victoria University Press, pp. 111-145.

Baayen, R. H. (2008). Analyzing Linguistic Data. A Practical Introduction to Statistics Using

R. Cambridge University Press.

Bailey, C. J. (1991). Variation theory and second language learning; linguistics and social

constraints on interlanguage tense marking. Doctoral dissertation. Stanford University.

Bayley, R. (1994). “Consonant cluster reduction in Tejano English”. Language Variation and

Change 6, 303-326.

269
Bayley, R. (2013). “The Quantitative Paradigm.” In Chambers, J.K. and Schilling, N. (eds.)

The Handbook of Language Variation and Change. (pp. 87-107). Oxford: John Wiley

& Sons, Inc.

Bell, R. T. (1976). Sociolinguistics: Goals, Approaches and Problems. London: Batsford.

Bermúdez-Otero, R. (2007). Diachronic phonology. In The Cambridge handbook of

phonology, ed. Paul de Lacy. (pp. 497–517). Cambridge: Cambridge University Press,.

Bermúdez-Otero, R. (2015). Amphichronic explanation and the life cycle of phonological

processes. In The Oxford handbook of historical phonology, eds. P. Honeybone & J. C.

Salmons. (pp- 374-399). Oxford: Oxford University Press.

Britain, D. (2001). “Welcome to East Anglia!: Two major dialect ‘boundaries’ in the Fens”, in

Fisiak, J. and Trudgill, P. (eds.) East Anglian English. (pp. 217-242). Cambridge: D.S.

Brewer.

Britain, D. (2002). “Space and spatial diffusion.” In Chambers, J.K, Trudgill, P., Schilling-

Estes, N. (eds), The Handbook of Language Variation and Change. (pp. 603-637).

Blackwell, Oxford.

Britain, D. (2013). “Space, Diffusion and Mobility”, in Chambers, J.K. and Schilling, N. (eds.)

The Handbook of Language Variation and Change. 2nd edn. (pp. 471-500). Chichester:

John Wiley and Sons Ltd.

Britain, D. (2014). “Where North meets South?: Contact, divergence, and the routinisation of

the Fenland dialect boundary.” In D. Watt and C. Llamas (eds.), Languages, borders

and identity. (pp. 27-43). Edinburgh: Edinburgh University Press.

Britain, D. (2015). “Between North and South: The Fenland.” In Hickey, R. (ed.). Researching

Northern English. (pp. 417-435). Amsterdam: Benjamins.

270
Broadbent, J. M. (2008). “t-to-r in West Yorkshire English.” English Language and Linguistics

12, 141-168.

Brysbaert, M., and New, B. (2009). “Moving beyond Kûcera and Francis: A critical evaluation

of current word frequency norms and the introduction of a new and improved word

frequency measure for American English”. Behavior research methods 41, 977–990.

Buizza, E. (2011b). “/t/ lenition in RP English spontaneous speech.” Paper presented at the 8th

UK Conference on Language Variation and Change, Edge Hill, September 14th, 2011.

Burbano-Elizondo, L. (2008). Language variation and identity in Sunderland. PhD thesis,

University of Sheffield.

Burbano-Elizondo, L. (2015). “Sunderland.” In Hickey, R. (eds) Researching Northern

English. (pp. 183-204). Philadelphia: John Benjamins.

Butcher, K. (2019) “The moan/mown long-mid vowel merger in East Anglia: Exploring

correlations of GOAT and GOOSE variations”. Paper presented at UK Language

Variation and Change (UKLVC) 12, 3-5 September 2019.

Bybee, J. (2002) “Word frequency and context of use in the lexical diffusion of phonetically

conditioned sound change.” Language Variation and Change 14, 261-290.

Carr, P. (1991). “Lexical properties of post-lexical rules: Postlexical derived environment and

the Elsewhere Condition”. Lingua 85, 255-268.

Carr, P. (1999). English Phonetics and Phonology: An Introduction. Oxford: Blackwell.

Carr, P., and Honeybone, P. (2007). “English phonology and linguistic theory: An introduction

to issues, and to Issues in English Phonology”, Language Sciences 29, 117-153.

Cedergren, H. and Sankoff, D. (1974). “Variable rules: Performance as a statistical reflection

of competence.” Language 50, 233-55.

271
Chambers, J.K. and Trudgill, P. (1998). Dialectology. 2nd edn. Cambridge: Cambridge

University Press.

Chambers, J.K., Trudgill, P. and Schilling-Estes, N. (2002). “Studying Language Variation:

An Informal Epistemology.” The Handbook of Language Variation and Change. (pp.

3- 14). Oxford: Blackwell.

Chambers J. K., and Schilling-Estes, N. (2013). The Handbook of Language Variation and

Change. Chichester: Wiley-Blackwell.

Chomsky, N. and Halle, M. (1986). The Sound Pattern of English. New York: Harper and

Row.

Ciancia, C. & Patrick, P. L. (2019). (t,d) deletion and the unsolved problem of morphological

effect in British English. Paper presented at the International Conference on Language

Variation in Europe (ICLaVE) 10, Fryske Akademy, 26-28 June.

Ciancia, C. & Patrick, P. (2019). Revealing new phonological insights into (t,d) deletion: An

ol’ variable unexplored in the South East of England. Paper presented at New Ways of

Analyzing Variation (NWAV) 48, University of Oregon, 10-12 October.

Clark, L. and Trousdale, G. (2013). “Using participant observation and social network

analysis”. In Manfred Krug and Julia Schlüter (eds.), Research methods in language

variation and change. (pp. 36-52). Cambridge: Cambridge University Press.

Claxton, A.O.D. (1968). The Suffolk Dialect of the Twentieth Century. 3rd edn. Ipswich: The

Boydell Press Ltd.

Christophersen, P. (1952). “The glottal stop in English”. English Studies 33, 156-63.

Clements, G. N. (1990). “The role of the sonority cycle in core syllabification”. In J. Kingston

& M. Beckman (eds.) Papers in Laboratory Phonology I. (pp. p. 282–333)

Cambridge: CUP.

272
Clements, G. and Hume, E. (1995). “The Internal Organisation of Speech Sounds.” In J.

Goldsmith (ed.) The Handbook of Phonological Theory. Oxford: Blackwell. 245-307.

Cloke, P., Crang, P., and Goodwin, M. (2014). ‘Urban and Rural Geographies: Introduction’,

in Cloke, P., Crang, P., and Goodwin, M. (eds.) Introducing Human Geographies. 3rd

edn. Abingdon: Routledge, pp. 688-689.

Collins, B. and Mees, I. (1996). “Spreading everywhere? How recent a phenomenon is

glottalisation in Received Pronunciation?” English World-Wide 17, 175-187.

Docherty (eds.) Urban Voices: accent studies in the British Isles. London: Arnold. 185-202.

Coulmas, F. (1997). “Introduction”. In Coulmas, F. (eds.) The Handbook of Sociolinguistics.

Oxford. Blackwell, pp, ix-1.

Cruttenden, A. (2001). Gimson’s Pronunciation of English: 6th Edition, Arnold Publishing.

Cruttenden, A. (2014). Gimson’s Pronunciation of English: 8th Edition. New York: Routledge.

Davis, D. and Braber, N. (2011). An investigation into dialect through oral history: The East

Midlands. http://www.le.ac.uk/emoha/community/dialect/home.html. (Last Accessed:

December 2019).

de Lacy, P. (2006). Markedness: Reduction and Preservation in Phonology. Cambridge

Studies in Linguistics 112. Cambridge University Press.

de Lacy, P. (2007). The interaction of tone, sonority, and prosodic structure. In P. de Lacy

(Ed.), The Cambridge Handbook of Phonology (Cambridge Handbooks in Language

and Linguistics. (pp. 281-308). Cambridge: Cambridge University Press.

Docherty, G. J., Foulkes, P., Milroy, J., Milroy, L., & Walshaw, D. (1997). “Descriptive

adequacy in phonology: a variationist perspective”. J. Linguistics 33, 275-310.

273
Docherty, G. and Foulkes, P. (1999). “Derby and Newcastle: Instrumental phonetics and

variationist studies.” In Foulkes, P., Docherty, G. (eds), Urban Voices: Accent Studies

in the British Isles. (pp. 47-71). Arnold, London.

Docherty, G. J. (2007). “Prosodic factors and sociophonetic variation: Speech rate and glottal

variants in Tyneside English”. In: Proceedings of the 16th International Congress of

Phonetic Sciences, Saarbruecken.

Docherty, G.J, and Foulkes, P. (2014). “Glottal variants of /t/ in the Tyneside variety of

English”. In: W. J. Hardcastle, J. M. Beck, ed. A Figure of Speech: A Festschrift for

John Laver. (pp.173-99). London: Lawrence Erlbaum,

Dodsworth, R. (2011). “Social class.” In R. Wodak et al. (eds.) The SAGE Handbook of

Sociolinguistics. (pp. 192-207). London: Sage.

Dymond, D. and Northeast, P. (1985). A History of Suffolk. Chichester: Phillimore & Co. Ltd.

Eddington, D. and Taylor, M. (2009). “T-glottalization in American English.” American

Speech, 84 (3): 298-314.

Ellis, A.J. (1869-89). On early English pronunciation. Parts 1-5. London: Philological

Society.

Erickson, B. H. & Nosanchuk, T. A. (1992) Understanding Data. Toronto: University of

Toronto Press.

Eckert, P. (1989), Jocks and burnouts: Social categories and identity in the high school. New

York: Teachers College Press.

Eckert, P. (1991). “Social polarization and the choice of linguistic variants”. In Eckert, P. (ed.).

New ways of analysing sound change. (pp. 213-232). New York: Academic Press.

Eckert, P. (1997). “Age as a Sociolinguistic Variable”, in Coulmas, F. (ed.) The Handbook of

Sociolinguistics. (pp. 151-167). Oxford: Blackwell Publishing.

274
Eckert, P. (2000). Linguistic Variation as Social Practice. Oxford: Blackwell Publishers Ltd.

Eckert, P. & Rickford, J. R. (2001). Style and Sociolinguistic Variation. Cambridge:

Cambridge University Press.

Eckert, P. (2008). “Variation and the indexical field”. Journal of Sociolinguistics, 12, 453-

476.

Eckert, P. (2012). ‘Three Waves of Variation Study: The Emergence of Meaning in the Study

of Sociolinguistic Variation’, Annual Review of Anthropology, 41, 87-100.

Ewen, C. and van der Hulst, H. (2001) The Phonological Structure of Words: An Introduction.

Cambridge: Cambridge University Press.

Fabricius, A. (2000). T-glottalling, between stigma and prestige: a sociolinguistic study of

modern RP. Unpublished Phd thesis. Copenhagen Business School.

Fasold, R. (1972). Tense marking in Black English: A linguistic and social analysis. Arlington,

Va: Center for Applied Linguistics.

Feagin, C. (2002). “Entering the community: fieldwork”. In Chambers, J.K., Trudgill, P. and

Schilling-Estes, Natalie (eds.), The Handbook of Language Variation and Change. (pp.

20–39). Malden, MA: Blackwell.

Fincham, P. (1976). The Suffolk We Live In. Woodbridge: Barbara Hopkinson Books.

Foley, J. (1977). Foundations of Theoretical Phonology. Cambridge: Cambridge University

Press.

Foulkes, P. and Docherty G. (2007). “Phonological Variation in England.” In Britain, D. (ed)

Language in the British Isles. (pp. 52-74). Cambridge: Cambridge University Press.

Foulkes, P. (2006). “Phonological Variation: A Global Perspective”, in Aarts, B. and

McMahon, A. (eds.) The Handbook of English Linguistics. (pp. 625-669). Oxford:

Blackwell Publishing Ltd.

275
Fors, K. L. (2015). Production and Perception of Pauses in Speech. Doctoral dissertation:

University of University of Gothenburg.

Fuchs, R. (2015). Word-Initial Glottal Stop Insertion, Hiatus Resolution and Linking in British

English. Paper presented at INTERSPEECH 2015, Dresden, September 6-10.

Fruehwald, J. (2012). “Redevelopment of a morphological class”. University of Pennsylvania

Working Papers in Linguistics 18 (1), Proceedings of the 35th Annual Penn Linguistics

Colloquium.

Fruehwald, J. (2016). “Filled pause choice as a sociolinguistic variable”. University of

Pennsylvania Working Papers in Linguistics 22, 39-49.

Giddens, A. (1989). Sociology. Cambridge: Polity.

Gimson, A.C. (1980) An Introduction to the Pronunciation of English. Third Edition. London:

Arnold.

Goldsmith, J. (1990). Austosegmental and Metrical Phonology. Oxford: Basil Blackwell.

Gordon, M. J. (2013). Labov: A Guide for the Perplexed. London: Bloomsbury Academic.

Gorman, K. and Johnson, D. E. (2013). “Quantitative analysis.” In Bayley, R. Cameron, R.

and Lucas, C. (eds.) The Oxford Handbook of Sociolinguistics. (pp. 214-240). Oxford:

Oxford University Press.

Gumperz, J. J. (1971). Language in Social Groups. Stanford, CA: Stanford University Press.

Gussenhoven, C. & Jacobs, H. (2011). Understanding Phonology. Third Edition. Abingdon:

UK.

Guy, G. R. (1980). “Variation in the group and the individual: The case of final stop deletion.”

In William Labov, ed., Locating language in time and space. (pp. 1-36). New York:

Academic Press.

276
Guy, G. (1988). “Advanced VARBRUL Analysis”. In K. Ferrara, B. Brown, K. Walters & J.

Baugh eds. Linguistic change and contact. (pp. 124-136). NWAV-XVI, 124-136.

Austin: University of Texas.

Guy, G. R. & Boyd, S. (1990). “The development of a morphological class.” Language

Variation and Change 2, 1-18.

Guy, G. R., (1991). “Explanation in variable phonology: An exponential model of

morphological constraints.” Language Variation and Change 3, 1-22.

Guy, G. R., & Boberg, C. (1997). “Inherent variability and the obligatory contour principle.”

Language Variation and Change 9, 149-164.

Guy, G. R., Hay, J. and Walker, A. (2008). Phonological, lexical, and frequency factors in

coronal stop deletion in early New Zealand English. In Laboratory Phonology 11.

Wellington, New Zealand.

Guy, G. R. (2013). “The cognitive coherence of sociolects: How do speakers handle

sociolinguistic variables?” Journal of Pragmatics, 52, 63-71.

Guy, G. R. (2014). “Words and numbers: statistical analysis in sociolinguistics.” In Holmes J.

and Hazen K. (ed) Research Methods in Sociolinguistics: A Practical Guide. (pp. 192-

210). Oxford: John Wiley & Sons, Inc.

Guy, G. R. & Hinskens, F. (2016) Linguistic coherence: Systems, repertoires and speech

communities. Lingua, vol. 172-173, 1-9.

Guy, G. R., & Torres Cacoullos, R. (2018) Reporting statistical results for LVC. Paper

presented at New Ways of Analyzing Variation (NWAV) 47, 19 October, New York.

Harris, J. & Kaye, J. (1990). A tale of two cities: London glottalling and New York City

tapping. The Linguistic Review 7, 251–274.

Harris, J. (1994). English Sound Structure. Oxford: Blackwell.

277
Hartford, B. S. (1975). The English of Mexican-American adolescents in Gary, Indiana.

Doctoral dissertation, University of Texas, Austin.

Hartley, S. (1992). A study of the effects of sex and age on glottalisation patterns in the speech

of Tyneside schoolchildren. Unpublished undergraduate dissertation, University of

Newcastle upon Tyne.

Herman, D. (1999). “Toward a socionarratology: New ways of analyzing natural language

narratives”. In D. Herman (ed.), Narratologies: New Perspectives on Narrative

Analysis. (pp. 218-46). Columbus, OH: Ohio State University Press.

van Heuven, W. J. B., Mandera, P., Keuleers, E. and Brysbaert, M. (2014). Subtlex-uk: A

new and improved word frequency database for British English. The Quarterly Journal

of Experimental Psychology 67, 1176–1190.

Hickey, R. (2005). Dublin English. Evolution and Change. Amsterdam: John Benjamins.

Hickey, R. (2009b). “Weak segments in Irish English”, in Donka Minkova (ed.) Phonological

Weakness in English. From Old to Present-day English. (pp. 116-29). Basingstoke:

Palgrave Macmillan.

Hock, H. H. and Joseph, B. (1996). Language History, Language Change and Language

Relationship. Berlin: Mouton de Gruyter.

Holmes, J. (1995). Glottal stops in New Zealand English: an analysis of variants of word-final

/t/. Linguistics. 33, 433-463.

Holmes, J. (1997). Setting new standards: Sound changes and gender in New Zealand English.

English World-Wide 18, 107–142.

Holmes-Elliot, S. (2019). Calibrate to innovate: Variation and change between childhood and

adolescence. Paper presented at UK Language Variation and Change (UKLVC) 12,

Queen Mary University of London, 3-5 September 2019.

278
Honeybone, P. (2008). “Lenition, weakening and consonantal strength: tracing concepts

through the history of phonology.” In Brandão de Carvalho, J., Scheer, T. & Ségéral,

P. (eds) Lenition and Fortition. (pp. 9-93). Berlin: Mouton de Gruyter.

Honeybone, P. (2012). “Lenition in English.” In Nevalainen, Terttu & Traugott, Elizabeth

(eds.) The Oxford Handbook of the History of English. (pp. 773-787). New York:

Oxford University Press.

Horvath, B. M. (1985) Variation in Australian English. Cambridge: Cambridge University

Press.

Horvath, B. & Sankoff, D. (1987). “Delimiting the Sydney speech community”. Language in

Society, 16, 179-204.

Hudson, R. and Holloway, A. (1977). Variation in London English. London: UCL Press.

Hudson, R.A. (1996). Sociolinguistics. 2nd ed. Cambridge: Cambridge University Press.

Huges, A., Trudgill, P. & Watt, D. (2012). English Accents and Dialects. London: Hodder

Education.

Hyman, L. M. (1975). Phonology: theory and analysis. USA: Holt, Rinehart and Winston.

Ihalainen, O. (1994). “The dialects of England since 1776”. In R. Burchfield (ed.) The

Cambridge History of the English Language. Volume 5: English in Britain and

Overseas: Origins and Development. (pp. 197- 274). Cambridge: Cambridge

University Press.

Jeffries, E. (2011). The problem of glottaling in the head of a foot; a theoretical exploration

based on data from West Yorkshire. MA dissertation, University College London.

Jeffries, E. (2019). “Preschool children's categorization of speakers by regional accent”.

Language Variation and Change 31, 1-24.

279
Johnson, W. & Britain, D. (2007) L-vocalisation as a Natural Phenomenon, Language Science

29, 294-315.

Johnston, R.J. (1994). ‘Urban’. In Johnston, R.J., Gregory, D., and Smith, D.M. (eds.) The

Dictionary of Human Geography. 3rd ed. (pp. 651-653). Oxford: Blackwell Publishers.

Johnson, D. E. (2009). “Getting off the Goldvarb Standard: Introducing Rbrul for Mixed

Effects Variable Rule Analysis.” Language and Linguistics Compass, 3 (1): 359-383.

Johnstone, B. (2000). Qualitative Methods in Sociolinguistics. Oxford: Oxford University

Press.

Jurafsky, D., Alan, B., Gregory, M., & Raymond, W., D. (2001). “Probabilistic relations

between words: Evidence from reduction in lexical production”. In J. Bybee & P.

Hopper (eds.), Frequency and the emergence of linguistic structure. (pp. 229-254).

Amsterdam: Benjamins.

Kaye, J. (1990). ‘Coda’ Licensing. Phonology 7, 301-330.

Kirchner, R. (2001). An Effort Based Approach on Consonant Lenition. New York: Routledge.

Kirchner, R. (2004). “Consonant Lenition”. In Hayes, B. Kirchner, R. & Steriade, D.

Phonetically Based Phonology. (pp. 313- 345). Cambridge University Press (UK).

Kerswill, P. and Williams, A. (1992). Some principles of dialect contact: evidence from the

New Town of Milton Keynes. In Philippaki-Warburton, I. & Ingham, R. (eds.).

Working Papers 1992. Department of Linguistic Science, University of Reading. 68–

90.

Kerswill, P. and Williams, A. (1997). “Investigating social and linguistic identity in three

British schools”. In U. Kotsinas, A.-B. Stenstro¨ & A.M. Karlsson (eds.)

Ungdomssprak i Norden. Stockholm. (pp. 159-76). University of Stockholm Press.

280
Kerswill, P. E. (1987). “Levels of linguistic variation in Durham”, Journal of Linguistics.

23(1), 25-49.

Kerswill, P. & A. Williams. (2000). “Creating a new town koine: children and language change

in Milton Keynes”. Language in Society 29, 65–115.

Kerswill, P. (2007). “Standard and non-standard English”. In Britain, D. Language in the

British Isles. (pp. 34-51). Cambridge: Cambridge University Press.

Kerswill P. (2009). “Language and social class”. In Culpeper J, Katamba F, Kerswill P,

McEnery A, Wodak R, editors, English language: description, variation and context.

(pp. 358-37). Houndmills, Basingstoke: Palgrave.

Kiparsky, P. (1982) “Lexical morphology and phonology”. In I. S. (ed.) Linguistics in the

morning calm. Seoul: (pp. 3-91). Hanshin.

Kiparsky, P. (1985). Some consequences of lexical phonology. In C. J. Ewens & J. M.

Anderson (eds.), Phonology yearbook 2. (pp. 85-138). Cambridge: Cambridge

University Press.

Kiparsky, P. (1994). An OT Perspective on Phonological Variation. Handout from Rutgers

Optimality Workshop & NWAV 1994. URL http://www.stanford.edu/

kiparsky/Papers/nwave94.pdf.

Kirchner, R. (2004). ‘Consonant lenition’, in B. Hayes, R. Kirchner and D. Steriade (eds.),

Phonetically-based Phonology. (pp. 313-345). Cambridge: Cambridge University

Press.

Kokeritz, H. (1932). The Phonology of the Suffolk Dialect. Uppsala: A.-B. Lundequistka

Bokhandlen.

Labov, W. (1969). “Contraction, deletion, and inherent variability of the English copula”.

Language 45, 715-762.

281
Labov, W. (1966). The Social Stratification of English in New York City. Washington, DC:

Centre for Applied Linguistics.

Labov, W. & Cohen, P. (1967). Systematic relations of standard and non-standard rules in the

grammars of Negro speakers. Project Literacy Reports No. 8, 66-84. Ithaca NY:

Cornell University.

Labov, W., Paul C., Robins, C. and Lewis, J. (1968). A study of the non-standard English of

Negro and Puerto Rican speakers in New York City. Final report, Cooperative Research

Project 3288. Vols. I and II.

Labov, W. (1972). Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.

Labov, W. (1982b). “Objectivity and commitment in linguistic science”. Language in Society,

11, 165–201.

Labov, W. (1984). “Field methods of the project on linguistic change and variation”. In J.

Baugh, and J. Sherzer (eds.), Language in Use: Readings in Sociolinguistics. (pp. 28-

66). Englewood Cliffs, NJ: Prentice Hall.

Labov, W. (1989). “The child as linguistic historian.” Language Variation and Change 1(1),

85-97.

Labov W. (1990). “The intersection of sex and social class in the course of linguistic change.”

Language Variation and Change 2(2), 205-254.

Labov, W. (1994). Principles of linguistic change: Internal factors. Oxford: Blackwell.

Labov, W. (2001). Principles of Linguistic Change, Volume 2: Social Factors. Oxford:

Blackwell Publishers Ltd.

Labov, W. and Rosenfelder, I. (2001). The Philadelphia Neighborhood Corpus.

Labov, W. (2010). “Narratives of Personal Experience”. In Patrick Hogan (ed.), Cambridge

Encyclopedia of the Language Sciences. (pp. 546–48). Cambridge: Cambridge U.P.

282
Lacoste, V. (2012). Phonological Variation in Rural Jamaican Schools. Amsterdam: John

Benjamins.

Ladd, D. R. (1996). Intonational Phonology. Cambridge: Cambridge University Press.

Ladefoged, P. (2005) Vowels and Consonants. Second Edition, Blackwell Publishing.

Lass, R. (1984). Phonology: an introduction to basic concepts. Cambridge: Cambridge

University Press.

Laver, J. (1994). Principles of Phonetics. Cambridge: Cambridge University Press.

Laver, John (1980). The Phonetic Description of Voice Quality. Cambridge University Press.

Leemann, A., Walker, E., Brown, G., Kirkham, S., Nance, C., Britain, D. and Blaxter, T.

(2019). Crowdsourcing variation and change of FOOT / STRUT and TRAP / BATH

across England. Paper presented at the International Conference on Language and

Variation in Europe (ICLaVE) 10, 26-28 June 2019.

Levinson, S. (1983). Pragmatics. Cambridge: Cambridge University Press.

Llamas, C. (2001). “The sociolinguistic profiling of (r) in Middlesbrough English”. Etudes et

Travaux, 123-139.

Llamas, C. (2006). “Shifting identities and orientations in a border town”. In. Omoniyi, T. &

White, G. (eds.). Sociolinguistics of Identity. (pp. 92-112). Continuum.

Luick, K. (1921). Historische grammatik der Englishen sprache, Ester Band, Ester Abteilung,

Leipzig: Chr. Herm. Tauchnitz.

Macafee, C. I. (1997). “Ongoing change in Modern Scots: The social dimension.” In Jones

(ed.): 514-48.

Macaulay, R. K. S. (1977). Language, social class and education: a Glasgow study.

Edinburgh: Edinburgh.

Malster, R. (2000). A History of Ipswich. Chichester: Phillimore & Co. Ltd.

283
Martin, E. (1999). “Suffolk and the East Anglian Kingdom”, in Dymond, D. and Martin, E.

(eds.) An Historical Atlas of Suffolk. 3rd ed. Ipswich. (pp. 22-23). The Archaeology

Service, Suffolk County Council.

Mascaró, J. (1983). “Phonological Levels and Assimilatory Processes.” MS, Universitat

Autònoma de Barcelona.

Mathisen, A. G. (1999). “Sandwell, West Midlands: ambiguous perspectives on gender

patterns and models of change.” In P. Foulkes and G. Docherty (eds) Urban Voices:

Accent Studies in the British Isles. (pp. 107-23). London: Edward Arnold.

Matthews, W. (1972) Cockney past and present. London: Routledge.

Mees, I. M. & Collins, B. (1999). Cardiff: real-time study of glottalisation. In: Foulkes, P.

Docherty, G. (eds.) Urban Voices: Accent Studies in the British Isle. (pp 185-202).

Arnold, London.

Milroy, L. (1980). Language and Social Networks. Oxford: Blackwell.

Meyerhoff, M. (2019). Introducing Sociolinguistics. New York: Routledge.

Milroy, J., L. Milroy, S, Hartley and Walshaw, D. (1994). “Glottal stops and Tyneside

glottalization: Competing patterns of variation and change in British English”.

Language Variation and Change 6, 327–57.

Milroy, L. and Gordon, M.J. (2003). Sociolinguistics: Method and Interpretation. Oxford:

Blackwell Publishing Ltd.

Milroy, J. (2007). “The history of English”. In Britain, D. Language in the British Isles.

Cambridge: Cambridge University Press, 9-33.

Mugglestone, L. (2003). Talking Proper: The Rise of Accent as a Social Symbol (2nd edn.)

Oxford: Oxford University Press.

Nathan, G. S. (2008). Phonology. A cognitive grammar introduction. Amsterdam: John

Benjamins.

284
Neu, H. (1980) “Ranking of constraints on /t,d/ deletion in America English: a statistical

analysis.” In W. Labov (ed.) Locating Language in Time and Space. (pp. 37-54). New

York: Academic Press.

Nomis. (2013). 2011 Census Data for England and Wales on Nomis. Available at:

https://www.nomisweb.co.uk/census/2011 (Last accessed: December 2018).

O’Connor, J.D. (1952). “RP and the reinforcing glottal stop”. English Studies 33, 214-18.

Odden, D. (2005). Introducing Phonology. Cambridge: Cambridge University Press.

Orton, H. (1933) The Phonology of a South Durham dialect. London: Paul, Trench, Trubner.

Orton, H. & Tillling, P. M. (eds.) (1969-1971), Survey of English Dialects. The East Midland

Counties and East Anglia. Leeds: Arnold.

Orton, H., Sanderson, S. and Widdowson, J. (1978). The Linguistic Atlas of England. London:

Croom Helm.

Parker, S. (2002). Quantifying the Sonority Hierarchy. PhD dissertation. U Mass, Amherst.

Patrick, P. L. (1991). “Creoles at the intersection of variable processes: -t, d deletion and past-

marking in the Jamaican mesolect”. Language Variation and Change 3, 171–189.

Patrick, P. L., Beall, H., Castillo-Ayometzi, C., Kuo, C., Mileva, R.Miller, J. Roberts, G.,

Takakusaki, Y., & Wake, V., Y. (1996) One hundred years of (TD)-deletion in African

American English. Paper presented at NWAVE-XXV in Las Vegas, Nevada.

Patrick, P. L. (1997). “Style and Register in Jamaican Patwa”. In Schneider, E. (eds.) Englishes

Around the World: Studies in Honour of Manfred Görlach. Vol. 2: Caribbean, Africa,

Asia, Australasia. (pp. 41-56). Amsterdam, Philadelphia: John Benjamins.

Patrick, P. L. (1999). Urban Jamaican Creole: Variation in the mesolect. [Varieties of English

around the World G17]. Amsterdam: John Benjamins.

285
Patrick, P.L. (2002). “The Speech Community”. In Chambers, J. K., Trudgill, P., Schilling-

Estes, N. (eds.) The Handbook of Language Variation and Change. (pp. 573-597).

Oxford: Blackwell.

Patrick, P.L. (2004) “British Creole: phonology.” In: Kortmann, B and Schneider, EW, (eds.)

A Handbook of Varieties of English. Phonology. De Gruyter, 231 - 243.

Pavlík, R. (2017). Some New Ways of Modeling T/D Deletion in English. Journal of English

Linguistics, 1 –34.

Payne, A. C. (1980). Factors controlling the acquisition of the Philadelphia dialect by out-of-

state children. In W. Labov (ed) Locating Language in Time and Space. (pp. 143-78).

New York: Academic Press.

Pennington, M. C. (1996). Phonology in English language teaching: an international

approach. USA: Addison Wesley Longman.

Pointer, F. (1996) Cockney Glottaling: A Study on the Phonetics of Contemporary London

Speech. Essen: Verlag Die Blaue Eule.

Potter, R. S. (2018). A variationist multilocality study of unstressed vowels and verbal -s

marking in the peripheral dialect of east Suffolk. Unpublished PhD thesis, University

of Essex.

Prince, A.S. (1983). “Relating to the grid”. Linguistic Inquiry 14, 19-100.

Prokosch, E. (1939). A comparative Germanic grammar. Baltimore: Linguistic Society of

America.

Purse, R., and Turk. A. (2016). ‘/t, d/deletion’: Articulatory gradience in variable phonology.

Paper presented at LabPhon.

Purse, R. and Tamminga, M. (2019). Evaluating lexical frequency measures for sociolinguistic

variation. Paper presented at UK Language Variation and Change (UKLVC) 12,

Queen Mary University of London, 3-5 September 2019.

286
Purse, R. (2019). Task effects in the articulation of coronal stop deletion. Paper presented at

New Ways of Analyzing Variation (NWAV) 48, University of Oregon, 10-12 October

2019.

Roberts, J. (2006). “As old becomes new: glottalization in Vermont.” American Speech, 81

(3): 227-249.

Renwick, M. E. L, Temple, R., Baghai-Ravary, L., and Coleman, J. S. (2014), (t,d) Deletion

in Everyday Speech. Paper presented at Labphon 14, Tokyo, July 25-27.

Reynolds, W., T (1997). Variation and phonological theory. PhD thesis, University of

Pennsylvania.

Rice, K. (1992). “On deriving sonority: a structural account of sonority relationships”,

Phonology 9, 61-99.

Rickford, J. R., Ball, A., Blake, R., Jackson, R. and Martin, N. (1991). “Rappin’ on the copula

coffin: Theoretical and methodological issues in the analysis of copula variation in

African-American Vernacular English”. Language Variation and Change 3, 103-132.

Rickford, J. R. & McNair-Knox, F. (1994). “Addressee- and topic-influenced style shift: A

quantitative sociolinguistic study”. In D. Biber & E. Finegan (eds.) Sociolinguistic

Perspectives on Register. (pp. 235-76). Oxford: Oxford University Press.

Rickford, J. R. (1986), “The Need for New Approaches to Social Class Analysis in

Sociolinguistics.” Language and Communication 6 (3), 215-221.

Roberts, J. (1994). “Acquisition of variable rules: A study of (-t, d) deletion in preschool

children.” Child Language 24, 351-372.

Roberts, J. (1997). “Acquisition of variable rules: A study of (-t, d) deletion in preschool

children.” Child Language 24, 351-372.

Roberts, J. (2006). “As old becomes new: Glottalization in Vermont.” American Speech, 81

(3): 227-249.

287
Roca, I. & Johnson, W. (1999). A Course in Phonology. Oxford: Blackwell.

Romaine, S. (1975). Linguistic variability in the speech of some Edinburgh school-children.

Unpublished M.A. thesis, University of Edinburgh.

Romaine, S. (1984). “The sociolinguistic history of t/d deletion”. Folia Linguistica Historica

2, 221-255.

Rose, D. and Harrison, E. (2010). Social Class in Europe: An Introduction to the European

Socio-economic Classification. New York: Routledge.

Rosewarne, D. (1984). Estuary English. Times Educational Supplement 42, 19 October.

Sankoff, G. (1980) “A quantitative paradigm for the study of communicative competence”. In

G. Sankoff (ed.), The Social Life of Language. (pp. 47-79). Philadelphia: University of

Pennsylvania Press.

Sankoff, G. (1988). “Variable rules”. In Ammon, U., Dittmar, N., and Mattheier K. J. (eds.)

Sociolinguistics: An International Handbook of the Science of Language and Society.

Vol 2. (pp. 984-97). Belin: de Gruyter.

Sankoff, D., Tagliamonte, S. & Smith, E. (2005). Goldvarb X: A variable rule application for

Macintosh and Windows. Department of Linguistics, University of Toronto.

Santa Ana A. O. (1991). Phonetic simplification processes in the English of the barrio: A cross-

generational sociolinguistic study of the Chicanos of Los Angeles. Ph.D. dissertation,

University of Pennsylvania.

Santa Ana, A. O. (1992), “Chicano English Evidence for the Exponential Hypothesis: A

Variable Rule Pervades Lexical Phonology”, Language Variation and Change (4),

275-288.

Santa Ana, A. O. (1996) Sonority and syllable structure in Chicano English. Language

Variation and Change 8(1), 63-89.

288
Silverstein, M. (1976). Shifters, linguistic categories, and cultural description. Meaning in

anthropology. K.H. Basso and H. A. Selby. Albuquerque, University of New Mexico

Press: 1155.

Schilling, N. (2013). Sociolinguistic Fieldwork. Cambridge: Cambridge University Press.

Sharma, D. (2018). “Style dominance: Attention, audience, and the ‘real me.’” Language

Variation and Change, 47(1), 1-31.

Schleef, E. (2013). Glottal replacement of /t/ in two British capitals: Effects of word frequency

and morphological compositionality. Language Variation and Change 25(2), 201–23.

Schreier, D. (2013). “Collecting ethnographic and sociolinguistic data.” In Krug, M. &

Schlüter, (eds.) Reasearch Methods in Language Variation and Change (Studies in

English Language Series). Cambridge: Cambridge University Press, pp. 1-23.

Schuchardt, H. (1885). Über die Lautgesetze: gegen die Junggrammatiker. Berlin: Oppenheim.

(Reprinted and translated in T. Vennemann & T. Wilbur. [1972]. Schuchardt, the

Neogrammarians and the Transformational Theory of Phonological Change.

Frankfurt: Athenaum).

Sivertsen, E. (1960) Cockney Phonology. Bergen: Oslo University Press.

Sonderegger, M., Beltrama, A., Chatzikonstantinou, T., Franklin, E., Kirken, B., Lee, J,,

Nelson, M., Nicoletto, K., Penslar, T., Provenza, H., Rothfels, N., Bane, M., Graff, P.,

& Riggle, J. (2011). Coronal stop deletion on reality TV. Paper presented at NWAV-

40, Washington, DC, October 2011.

Smith, J. & Holmes-Elliott, S. (2017). “The unstoppable glottal: tracking rapid change in an

iconic British variable”. English Language and Linguistics, 1-33.

Speelman, D. (2014). “Logistic regression: a confirmatory technique for comparisons in

corpus linguistics.” In Glynn, D. and Robinson, J. A. (ed) Corpus Methods for

289
 Semantics: Quantitative Studies in Polysemy and Synonymy. (pp. 487-533).

Amsterdam: John Benjamins.

Straw, M. & Patrick, P. L. (2007). Dialect acquisition of glottal variation in /t/: Barbadians in

Ipswich. Language Sciences 29, 385–407.

Stuart-Smith, J. (1999). “Glottals past and present: a study on T-glottaling in Glaswegian”.

Leeds Studies in English 30, 181-204.

Stuart-Smith, J., Timmins, C. and Tweedie, F. (2007). “Talkin’ Jockney? Variation and change

in Glaswegian accent.” Journal of Sociolinguistics, 11 (2), 221-260.

Tamminga, M & Fruehwald J. (2013). Deconstructing TD deletion. Paper presented NWAV-

42, University of Pittsburgh & CMU.

Tamminga, M. (2016). “Persistence in phonological and morphological variation.” Language

Variation and Change 28, 335–356.

Tamminga, M. (2018). “Modulation of the following segment effect on English coronal stop

deletion by syntactic boundaries” Glossa, 3(1), 86.

Tamminga, M. (2019). “Interspeaker covariation in Philadelphia vowel change.” Language

Variation and Change, 31, 119–133.

Tanner, J., Sonderegger, M., & Wagner, M. (2017). “Production planning and coronal stop

deletion in spontaneous speech.” Laboratory Phonology, 8(1), 15.

Tagliamonte, S. (1998). Was/were variation across the generations: View from the city of

York. Language Variation and Change 10, 153–191.

Tagliamonte, S. and Temple, R. (2005) New perspectives on an ol’ variable: (t,d) in British

English. Language Variation and Change 17, 281–302.

Tagliamonte, S. (2006). Analysing Sociolinguistic Variation. Cambridge: Cambridge

University Press.

290
Temple, R. (2009). (t,d): the variable status of a variable rule. In Oiwi Parker Jones and Elinor

Payne, (eds) Papers in phonetics and computational linguistics, (Oxford University

Working Papers in Linguistics, Philology, and Phonetics 12). (pp. 145-70). Oxford:

Faculty of Linguistics, Philology, and Phonetics.

Temple, R. (2014). “Where and what is (t, d)? a case study in taking a step back in order to

advance sociophonetics”. In. C. Celata and S. Calamai (eds.), Advances in

Sociophonetics, (pp. 97-134). Amsterdam: John Benjamins.

Temple, R. (2017). “Phonetic detail and variationist phonology: the case of (t,d).”

Dialectologia 18, 129-155.

Thurneysen, R. (1898). Review of Pedersen (1897). Anzeiger für indogermanische Sprach-

und Altertumskunde. Beiblatt zu den Indogermanischen Forschungen 9: 42-48.

Tollfree, L. (1999). “South East London English: discrete versus continuous modelling of

consonantal reduction.” In P. Foulkes and G. Docherty (eds.) Urban Voices: accent

studies in the British Isles. London: Arnold. 163-184.

Trechter, S. (2013). ‘Social Ethics for Sociolinguistics’. In Mallinson, C., Childs, B., and Van

Herk, G. (eds.) Data Collection in Sociolinguistics: Methods and Applications. (pp. 33-

45). Abingdon: Routledge.

Trudgill, P. (1972). “Sex, Covert Prestige and Linguistic Change in the Urban British English

of Norwich”, Language in Society, 1 (2), pp. 179-195.

Trudgill, P., & Foxcroft, T. (1978). “On the Sociolinguistics of Vocalic Mergers: Transfer and

Approximation in East Anglia”. In P. Trudgill (Ed.), Sociolinguistic Patterns in British

English. (pp.69-79). London: Edward Arnold.

Trudgill, P. (1974). The Social Differentiation of English in Norwich. Cambridge: Cambridge

University Press.

Trudgill, P. (1986). Dialects in Contact. Oxford: Basil Blackwell Ltd.

291
Trudgill, P. (1988). “Norwich revisited: recent linguistic changes in an English urban dialect.”

English World-Wide 9(1), 33-49.

Trudgill, P. (1995). Sociolinguistics: An Introduction to Language and Society. London:

Penguin, 1995.

Trudgill, P. (1999). “Norwich: endogenous and exogenous language change.” In P. Foulkes

and G. Docherty (eds.) Urban Voices: accent studies in the British Isles. London:

Arnold. 124-140.

Trudgill, P. (2000). Sociolinguistics: An Introduction to Language and Society. 4th edn.

London: Penguin.

Trudgill, P. (2001a). ‘Modern East Anglia as a dialect area’, in Fisiak, J. and Trudgill, P. (eds.)

East Anglian English. (pp. 1-12). Cambridge: D.S. Brewer.

Trudgill, P. (2003). The Norfolk Dialect. Cromer: Poppyland Publishing.

Trudgill, P. (2004a). ‘The dialect of East Anglia: phonology’. In Schneider, E.W., Burridge,

K., Kortmann, B., Mesthrie, R. and Upton, C. (eds.) A Handbook of Varieties of

English, Volume 1: Phonology. (pp. 163-177). Berlin: Mouton de Gruyter.

Trudgill, P. (2018). I’ll git the milk time you bile the kittle do you oon’t get no tea yit no coffee

more oon’t I. Phonetic erosion and grammaticalization in English dialects. Paper

presented at the International Society for the Linguistics of English (ISLE) 5, London,

17-20 July 2018.

Upton, C. and Widdowson, J.D.A. (1999) An atlas of English dialects. Oxford: Oxford

University Press.

Walker, J. A. (2012). “Form, function, and frequency in phonological variation”. Language

Variation and Change 24, 397–415.

Ward, I.C. (1929). The phonetics of English. Cambridge: Heffer.

292
Watson, K. (2007). “Liverpool English”, Journal of the International Phonetics Association

37, 351-360.

Watt, D. and Milroy, L. (1999). “Variation in three Tyneside vowels: is this dialect

levelling?” In: P. Foulkes and G. Docherty, (eds.) Urban Voices: Accent studies in the

British Isles. (pp. 25-46). London: Hodder Education.

Watt, D. and Allen, W. (2003). Tyneside English. Journal of the International Phonetic

Association, 33(2), 267-271.

Watt, D., Llamas, C., French, P., Braun, A., Robertson, D., and Kendall, T. (2019).

Correspondences between vowel phoneme boundary locations in production and

perception in dialects of North-East England. Paper presented at the International

Conference on Language and Variation in Europe (ICLaVE) 10, 26-28 June 2019.

Weinreich, U., Labov, W. and Herzog, M. I., (1968). “Empirical foundations for a theory of

language change”. In Winfred. P., Lehmann, P. and Malkiel, Y. (eds.) Directions for

Historical Linguistics: A Symposium. (pp. 95-188). Austin TX: University of Texas

Press.

Wells, J. C. (1982) Accents of English. Cambridge: Cambridge University Press, vol. I.

Wells, J. C. (1982) Accents of English. Cambridge: Cambridge University Press, vol. II.

Wells, J.C. (1999a). Questions and Answers about Estuary English.

http://www.phon.ucl.ac.uk/home/estuary/ee-faqs-jcw.htm

Williams, A. & Kerswill, P. (1997). Investigating dialect change in an English new town. In

A. Thomas (ed.) Issues and methods in dialectology. Bangor: Department of

Linguistics, University of Wales, Bangor, 46-54.

Williams, A. & P. Kerswill (1999). “Dialect levelling: Change and continuity in Milton

Keynes, Reading and Hull”. In Foulkes & Docherty (eds.), 141–62.

Wilson, A. (1977). Writers of East Anglia. London: Secker and Warburg.

293
Wolfram, W. (1969). A sociolinguistic description of Detroit Negro speech. Washington DC:

Center for Applied Linguistics.

Wolfram, W. (1993). “Identifying and interpreting variables.” In Preston, D., R, (ed.)

American Dialect Research. (pp. 193-221). Amsterdam, Philadelphia: Benjamins.

Wolfram, W. and Schilling-Estes, N. (1998) American English. Oxford: Blackwell.

Woolard, K. (1985). “Language variation and cultural hegemony: Towards an integration of

linguistic and sociolinguistic theory.” American Ethnologist 12, 738-48.

Woolford, K. (2018). An old variable with new tricks: t/d deletion in the northeast of England.

Poster presented at the 8th Northern English Workshop, Newcastle University, March

27– 28.

Wright, J. (1905) The English dialect grammar. London: Henry Frowde.

Wyld, H.C. (1936). History of Modern Colloquial English (3rd edn). Oxford: Blackwell.

Wyld, H. C. (1927). A short history of English, with a bibliography of recent books on the

subject, and lists of texts and editions. London: John Murray.

Zuraw, K. (2009). “Treatments of weakness in phonological theory.” In Donka Minkova (ed.)

Phonological Weakness in English. (pp. 9-28). Palgrave Macmillan.

294
Appendix I
Non-significant predictors for (t,d) deletion – overall results.

Constraints % Tokens
Syllable stress
unstressed 45 773
stressed 22 4106

Sex
Males 27 2617
Females 25 2262
Age
middle-aged 26 1631
Young 26 1595
Old 25 1653
Class
working-class 27 2388
middle-class 25 2491
Location
Colchester 26 1658
Ipswich 26 1615
Norwich 25 1606

295
Appendix II
Non-significant results for word-final /t/ glottaling – overall results.

Constraints % Tokens

Sex
Males 72 1561
Females 69 1490
Age
young 75 1039
middle 74 992
old 65 1020
Social class
working class 72 1464
middle class 69 1587
Location
Ipswich 73 1114
Colchester 73 1097
Norwich 64 840

Lexical frequency
high 70 2882
low 58 169

Non-significant predictors for word-medial /t/ - overall results

Constraints % Tokens
Social class
WC 41 887
MC 32 985
Location
Ipswich 39 694
Colchester 39 578
Norwich 30 600

296