Comparative Variation Analysis: Grammatical Alternations in World Englishes (Studies in Language Variation and Change) 1108491561, 9781108491563

Citation preview

Comparative Variation Analysis

Variation studies is an increasingly popular area in linguistics, becoming embedded in curriculum design, conferences, and research. However, the field is at risk of fragmenting into different research communities with different foci. This pioneering book addresses this by establishing a canon of state-of-the-art quantitative methods to analyze grammatical variation from a comparative perspective. It explains how to use these methods to investigate large datasets in a responsible fashion, providing a blueprint for applying techniques from corpus linguistic, variationist, and dialectometric traditions in novel ways. It specifically explores the scope and limits of syntactic variability in a global language such as English, and investigates three grammatical alternations in nine varieties of English, exploring what we can learn about the grammatical choices that people make based on both observational and experimental data. Comprehensive yet accessible, it will be of interest to academic researchers and students of sociolinguistics, corpus linguistics, and World Englishes. B E N E D I K T S Z M R E C S A N Y I is Professor of Linguistics at KU Leuven. His research interests include language variation and its interfaces with typology, geolinguistics, complexity theory, and psycholinguistics. J A S O N G R A F M I L L E R is Assistant Professor of Corpus-Based Sociolinguistics at the University of Birmingham. His research involves the application of various quantitative techniques to examine the forces shaping how language varies across regions, styles, and time.

S T U D I E S I N L A N G UAG E VA R I ATION AND CHANGE Series Editor Sali A. Tagliamonte, University of Toronto Studies in Language Variation and Change is dedicated to studies of systematic and inherent variation in language, including contemporary or historical sources. It is concerned with the impact of society, geography and culture in so far as they intersect with the structures and processes of language. Studies in Language Variation and Change is firmly situated in the variationist sociolinguistic enterprise with its roots in historical linguistics, dialectology, anthropology and importantly in the advancing quantitative methods of the field. The series concentrates on book length syntheses of research that engages with the details of linguistic structure in actual speech production and processing (or writing). It emphasizes replicability of findings, consistent reporting and building critical and substantive explanations out of empirical foundations. Published so far in the series: Sociolinguistic Variation in Children’s Language: Acquiring Community Norms by Jennifer Smith and Mercedes Durham Explanations in Sociosyntactic Variation, edited by Tanya Karoli Christensen and Torben Juel Jensen Meaning, Identity and Interaction: Sociolinguistic Variation and Change in Game-theoretic Pragmatics by Heather Burnett Forthcoming titles: Linguistic Variation and Language Change: Synchrony Meets Diachrony by Alexandra D’Arcy

Comparative Variation Analysis Grammatical Alternations in World Englishes

Benedikt Szmrecsanyi KU Leuven

Jason Grafmiller University of Birmingham

Shaftesbury Road, Cambridge CB2 8EA, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of Cambridge University Press & Assessment, a department of the University of Cambridge. We share the University’s mission to contribute to society through the pursuit of education, learning and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781108491563 DOI: 10.1017/9781108863742 c Benedikt Szmrecsanyi and Jason Grafmiller 2023

This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press & Assessment. First published 2023 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Szmrecsanyi, Benedikt, 1976– author. | Grafmiller, Jason, author. Title: Comparative variation analysis : grammatical alternations in world Englishes / Benedikt Szmrecsanyi, Catholic University Leuven ; Jason Grafmiller, University of Birmingham. Description: Cambridge ; New York, NY : Cambridge University Press, 2023. | Series: Studies in language variation and change | Includes bibliographical references and index. Identifiers: LCCN 2023025075 | ISBN 9781108491563 (hardback) | ISBN 9781108798471 (paperback) | ISBN 9781108863742 (ebook) Subjects: LCSH: English language – Variation. | English language – Syntax. | English language – Foreign countries. | English language – Data processing. Classification: LCC PE1074.7 .S96 2023 | DDC 427–dc23/eng/20230622 LC record available at https://lccn.loc.gov/2023025075 ISBN 978-1-108-49156-3 Hardback Cambridge University Press & Assessment has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

To Angela, Leo, and Alex.

Contents

List of Figures List of Tables Series Editor’s Preface Acknowledgments Data Availability Statement 1

2

3

Introduction 1.1 Variationist Sociolinguistics and Corpus-Based Variationist Linguistics 1.2 Comparative Linguistics and Comparative Variation Analysis 1.3 Dialectology, Dialectometry, and Dialect Typology 1.4 Probabilistic Linguistics and Probabilistic Grammar 1.5 Psycholinguistics 1.6 English as a World Language 1.7 Structure Grammatical and Syntactic Variation 2.1 Does Grammatical Variation Even Exist? 2.2 From Traditional Dialectology to Variationist Linguistics 2.3 Grammatical Variation in English 2.4 Comparative Perspectives on Grammatical Variation in English 2.5 Grammatical Alternations Subject to Study in This Book 2.6 Summary World Englishes and Dialect Typology 3.1 Theoretical Models of World Englishes 3.2 A View from Dialect Typology

page ix xi xiii xvi xvii 1 4 5 6 7 8 9 10 12 12 14 15 19 21 32 34 34 42 vii

viii

Contents

3.3 Varieties in This Study 3.4 Summary

46 55

4

The Data 4.1 Data Types in Variationist Linguistics 4.2 The Corpora 4.3 Defining the Variable Contexts 4.4 Annotating the Constraints 4.5 Summary

56 56 60 65 75 81

5

Alternation-by-Alternation Analysis 5.1 Methods 5.2 A Comparative Analysis of the Genitive Alternation 5.3 A Comparative Analysis of the Dative Alternation 5.4 A Comparative Analysis of the Particle Placement Alternation 5.5 Discussion

82 82 84 92 98 106

6

Distances, Similarities, and Coherence 6.1 Background 6.2 VADIS: An introduction 6.3 VADIS: The Empirical Pipeline 6.4 Quantification via Similarity Coefficients 6.5 Evaluating Coherence 6.6 Mapping Out (Dis)similarity Relationships 6.7 Discussion

112 112 114 115 121 123 128 135

7

Experimental Corroboration 7.1 Methods 7.2 Results 7.3 Discussion

141 143 148 157

8

Where Are We Now, and Where to Next? 8.1 Empirical Summary 8.2 Typological Universals 8.3 Forces Driving Probabilistic Indigenization 8.4 L1 Transfer and L2 Learning 8.5 Implications for Variationist Sociolinguistics 8.6 The Road Ahead

166 167 171 173 176 180 187

References Index

191 217

Figures

3.1 3.2

5.1 5.2

5.3 5.4 5.5 5.6

5.7 5.8 5.9

5.10 5.11 5.12 5.13

The Three Circles model (inspired by Kachru, 1985) Multidimensional scaling visualization derived from co-occurance matrix containing 76 features × 46 varieties (adapted from Szmrecsanyi and Kortmann, 2009, 1651). Conditional Random Forest permutation variable importance ranking for the pooled genitive dataset. Conditional Random Forest permutation variable importance ranking of constraints on the genitive alternation by variety of English. Partial effects plot of the interaction of VARIETY_OF_E with PORANIMACYBIN in the genitive regression model. Partial effects plot of the interaction of VARIETY_OF_E with PORFINALSIBILANCY in the genitive alternation. Conditional Random Forest permutation variable importance ranking for the pooled dative dataset. Conditional Random Forest permutation variable importance ranking of constraints on the dative alternation by variety of English. Partial effects plot of the interaction of VARIETY_OF_E with RECPRON in the dative alternation model. Conditional Random Forest permutation variable importance ranking for the pooled particle placement dataset. Conditional Random Forest permutation variable importance ranking of constraints on the particle placement alternation by variety of English. Intercept adjustments for the random effect VARIETY_OF_E. Partial effects plot of the interaction of CIRCLE with DIRECTIONAL PP in the particle placement model. Partial effects plot of the interaction of CIRCLE with SEMANTICS in the particle placement model. Partial effects plot of the interaction of CIRCLE with DIROBJLETT L ENGTH B IN in an alternative particle placement model.

36

45 86

87 91 92 95

96 99 101

102 104 105 106 107 ix

x

6.1

6.2 6.3

6.4

6.5 6.6

6.7

6.8 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9

8.1

List of Figures

Variation-Based Distance and Similarity Modeling (VADIS) distance matrix for the third line of evidence in the particle placement alternation (all data included, eight constraints considered). Multidimensional scaling representation of third line distances for the particle placement alternation (see Figure 6.1). Multidimensional scaling representation of compromise distances per alternation: a) genitive alternation; b) dative alternation; c) particle placement alternation. Multidimensional scaling representation of the 0-matrix (a single compromise distance matrix merged across all lines and alternations). Dendrogram: clustering the 0-matrix (a single compromise distance matrix merged across all lines and alternations). Visualizing aggregate similarities: NeighborNet diagram depicting the 0-matrix (a single compromise distance matrix merged across all lines and alternations). Multidimensional scaling representation of the fused distances from all three VADIS lines for seventy-five simulated datasets representing five hypothetical dialect varieties. Hierarchical clustering of simulated datasets based on VADIS Line 2. Experimental items versus corpus model predicted probabilities. Welcome page and instructions for the ratings experiments. Experimental ratings versus corpus model predicted log odds, with LOESS smooths. Experimental ratings across varieties, with median rating. Experimental ratings versus corpus model predictions, averaged by item and variety. Partial effects plots of interaction of VARIETY on participant ratings. Partial effects plots of interaction of VARIETY and CORPUS PREDICTION on participant ratings. Partial effects plots of interaction of VARIETY and DIRECT OBJECT LENGTH on participant ratings. Predicted probabilities by DIROBJLENGTH (in words), register and VARIETY obtained from random forest model predicting the split variant (adapted from Szmrecsanyi et al., 2016a, figure 3). Partial effects plot of the interaction of VARIETY_OF_E with RECPRON in the dative alternation model.

125 129

130

131 132

134

138 139 146 147 148 149 150 154 155 156

162 178

Tables

3.1 4.1 4.2 4.3 4.4 4.5 4.6 4.7 5.1 5.2 5.3 5.4 5.5 5.6 6.1 6.2 6.3 6.4 6.5 6.6

English varieties and their theoretical categorization. page 55 Design of the ICE corpora. Values reflect number of texts. 62 Design of the GloWbE corpus (from Davies and Fuchs 2015, 6). 63 Summary of genitive variants in ICE and GloWbE corpus data. 75 Summary of dative variants in ICE and GloWbE corpus data. 75 Summary of particle placement variants in ICE and GloWbE corpus data. 76 Annotation schema for the coding of ANIMACY across the three alternations. 77 Annotation scheme for the coding of NP TYPE across the three alternations. 78 Variant rates of genitive constructions across varieties of English. 85 The genitive alternation: fixed effect coefficients in mixed-effects logistic regression analysis. 89 Variant rates of dative constructions across varieties of English. 93 The dative alternation: fixed effect coefficients in mixed-effects logistic regression analysis. 97 Variant rates of particle placement constructions across varieties of English. 100 The particle placement alternation: fixed effect coefficients in mixed-effects logistic regression analysis. 104 Predictor sets used for the VADIS analysis. 116 Model estimates for three fictitious varieties A, B, and C. 118 Distance matrix for fictitious varieties A, B, and C (Euclidean distance). 118 Mean distances and mean similarities per variety. 119 Similarity coefficients across lines of evidence and alternations. Input dataset: all available data. 121 Core grammar scores (0) and hierarchies of stability for subsets of the data. 122

xi

xii

6.7

6.8 7.1 7.2

7.3

7.4

List of Tables

DBCalternation measurements: Mantel correlation coefficients between fused distance matrices (combining all lines of evidence and based on all available data). Mantel correlation coefficients between line-of-evidence-specific distance matrices. Average ratings (with standard deviations) by item across participants. Summary statistics, fixed effects coefficients, and random effects standard deviations for linear mixed-model fitting participant rating as function of DIRECT OBJECT LENGTH, CORPUS MODEL PREDICTION, and participant VARIETY, GENDER, and EDUCATION LEVEL . Agreement scores reflecting the percentage of experimental items in which participants’ preferred variant matched the variant predicted by corpus model. Agreement scores reflecting the percentage to which participants’ preferred variant matched the variant predicted by corpus model.

126 127 151

153

159 160

Series Editor’s Preface

As Language Variation and Change scholarship has expanded beyond variableby-variable analyses and North America, the scholars of the field have begun to broaden their academic relationships as well. I got to know Benedikt Szmrecsanyi through a collaborative research project funded by the National Science Foundation in the USA: Bresnan, Joan. 2010–2014. The Development of Syntactic Alternations. Research project funded by the National Science Foundation (NSF). BCS1025602. With collaborators Ford, Marilyn, Hay, Jen, Rosenbach, Anette, Szmrecsanyi, Benedikt, and Tagliamonte, Sali A. The project brought together multiple data sets with a plan to build parallel corpora in order to conduct a unified analysis of two well-studied syntactic variables, the dative and genitive alternation. What a great plan, right? It was a veritable epic journey, not only in expanding studies of variation and change, but also in setting the stage for combining data sets, advancing statistical modelling, and importantly, establishing best practice for a multiparty collaborative enterprise. While the grandiose aims of the project were to refine the variable contexts in consultation with other members of the team, we spent most of our time in deep discussion about which was the best way to code contexts of variation and analyze them “properly.” Everyone had their own backgrounds, training and predilections and we all had our own vocabulary. Looking back, I wish we had recorded our meetings and discussions. They would have been a gold mine for studying how academic disciplines operate in practice; data, coding, analysis, but oh, so many different ways of doing things and an insightful journey into how different people come to an understanding of each other’s point of view. During the same time frame statistical methods were changing quickly as mixed-effects models and new tools for analysis were introduced. This required us to adjust the analytic models and rework the statistical modeling techniques. One of Joan’s Ph.D. students at the time was Jason Grafmiller, who was instrumental in conducting the analyses of the big, amalgamated data set that was eventually published in Glossa.

xiii

xiv

Series Editor’s Preface

Szmrecsanyi, Benedikt, Grafmiller, Jason, Bresnan, Joan, Rosenbach, Anette, Tagliamonte, Sali A. and Todd, Simon. 2017. Spoken Syntax in a Comparative Perspective: The Dative and Genitive Alternation in Varieties of English. Glossa. 861–27. This backstory gives you the history from which developed the authorship team of Benedikt Szmrecsanyi and Jason Grafmiller. They were uniquely poised to take variationist work to a new level, not only across corpora and with cuttingedge methods, but also with established experience in functional collaboration. In this book, Benedikt and Jason have taken variationist work to the next level, extending the comparative endeavour to a global perspective, analyzing nine varieties of English from two compendia of data, the International Corpus of English (ICE) and the Global Corpus of Web-based English (GloWbE) – parallel, balanced corpora of the standard national varieties of each country, including both spoken and written registers. As Benedikt and Jason started presenting their work at conferences, I was consistently impressed with their methods, techniques and the importance of the findings arising from their studies. On July 9–10, 2018, some of the original Bresnan-lead team and a few others met for a reunion workshop at Annette Rosenbach’s beautiful farm Tanagra in McGregor, Western Cape, in South Africa. After yet another superb presentation of results about their research in a talk entitled “Measuring Variable Grammars,” I suggested to Benedikt and Jason that it was time to pull together their joint work in a single monograph to consolidate the new direction of comparative studies and advanced statistical techniques they had pioneered. Among their new techniques they had developed was a Variation Based Distance and Similarity Modelling (VADIS) method which quantifies the similarity between and coherence across datasets as a function of the correspondence in their patterns of choice between competing variants of a variable. In December 2018, Jason came to Toronto to teach me how to use the VADIS method. One cannot have too many tools with which to probe data! Comparative Variation Analysis: Syntactic Variation in World Englishes explores the stability and plasticity of probabilistic grammar(s) across (standard) varieties of English around the globe in three syntactic alternations: the genitive, dative, and particle placement alternations. In keeping with the mission statement of the series, Studies in Language Variation and Change, the book advances the field in several ways. First, it applies rigorous quantitative methods, not only using the standard methods of the field but augmenting the “sociolinguistic toolkit” to include new visualization methods. Second, it brings into the field a new level of cross-variety comparison, extending the comparative sociolinguistic enterprise to incorporate new methods (e.g. VADIS). Third, it advances the techniques brought to bear on cross-linguistic

Series Editor’s Preface

xv

analysis by using experimental methods. Szmrecsanyi and Grafmiller adhere to quantitative procedures meeting the standards of replicability, consistent reporting, and the embedding of research findings in sociolinguistic theory. Their efforts to develop and employ multiple “lines of evidence,” corpus, dialectal and experimental, to arrive at a fulsome explanation lends greater validity to the eventual explanations they propose. As the chapters in Benedikt and Jason’s book build from the introduction, to the variables, to the methods, the reader will notice they are entering new and exciting territory for variationist studies. Benedikt and Jason step beyond alternation-by-alternation analyses (Chapter 5), which is itself cutting-edge, into analyses of distance, similarity, and coherence (Chapter 6), where the reader will learn the VADIS method. But that is not all; Chapter 7 demonstrates experimental corroboration of corpus predictions that tap the grammatical knowledge of participants’ preferences. In the final chapter, “Where Are We Now, and Where to Next?” Benedikt and Jason pull together all the evidence and synthesize it insightfully, pointing analysts to “the road ahead.” They propose two explanations for the prevalence of probabilistic grammar universals: 1) shared histories, the “common ancestry” of the varieties under investigation, and 2) shared humanity, “the influence of theoretically universal biases in production and comprehension that influence language structure.” By the end of the book, Benedikt and Jason have taken readers on an adventure in language variation, deep into the largest corpora in the world, using the most cutting-edge methods and they have sorted out – I would judge – to be among the most complex set of findings ever. Now, there is an apocryphal story, passed down to me from late-night conversations with colleagues and students, that when faced with inscrutable findings, Benedikt’s advice and modus operandi is to partake in a glass of wine and deeply ponder the patterns. Whether this is true or not, the fact that Benedikt and Jason have been able to concisely and incisively interpret their complex and multi-faceted results (i.e. see the forest in the trees) is testimony to their scholarship and deep understanding of human language. The intricate design in World Englishes has sorted itself into a comprehensive understanding at a new level that that will leave readers equipped for advancing their own studies in the future. Sali A. Tagliamonte

Acknowledgments

First and foremost, we would like to thank Melanie Röthlisberger and Benedikt Stemmler (née Heller) for their extensive and inspired work on the dative and genitive alternations. Melanie and Benedikt were hardworking PhD students in the early stages of the project which this book summarizes, and we cite their work throughout this book. Thank you, Melanie and Benedikt, for the heavy lifting and for your commitment – this project is your project, really. We are likewise extremely grateful to Laura Rosseel, whose comparatively short stint as a project postdoc proved indispensable for getting the rating task experiments on their way, and for codesigning the VADIS method. There are countless other people who contributed directly or indirectly, in one way or another. Here we would like to mention in particular the participants of our 2016 workshop on Probabilistic Variation across Dialects and Varieties (Joan Bresnan, Anette Rosenbach, Marianne Hundt, Sali Tagliamonte, Tobias Bernaisch, Christoph Wolk, Lars Hinrichs, Natalia Levshina, Daniel Ezra Johnson, Magali Paquot, and Dominique Hess [née Bürki]), from whose inspiring feedback we profited enormously. We would also like to express our thanks to Hubert Cuyckens and Dirk Geeraerts (both KU Leuven), who were heavily and passionately involved during the design phase of the project (the first-named author learnt a lot about grant-proposal writing from Hubert and Dirk). On the institutional plane, we are grateful to KU Leuven for hosting the project, to the Quantitative Lexicology and Variational Linguistics (QLVL) research group at KU Leuven’s Department of Linguistics for intellectual stimulation, and to the Research Foundation Flanders (FWO, grant no. G.0C59.13N) for generous funding. Finally, our heartfelt thanks go to Matt Hunt Gardner, Thomas Van Hoey, and Yi Li for a thorough reading of some of the chapters. Needless to say, all remaining errors are our own.

xvi

Data Availability Statement

The data we report on in this study were compiled from two corpora: the International Corpus of English (http://ice-corpora.net/ice/index.html) and the Corpus of Global Web-Based English (https://www.english-corpora.org/ glowbe/). Restrictions apply to the availability of these corpora, which were used under license for this study. The annotated datasets that were collected from these corpora, along with the dataset of experimental ratings, are freely available on our Open Science Framework (OSF) repository at https://osf.io/ 5hvtw/. All statistical analyses were conducted using R statistical software (https://www.r-project.org/), and the corresponding R code for our analyses can also be found in our OSF repository. Tools for applying the VariationBased Distance and Similarity Method can be found in the VADIS R package (https://github.com/jasongraf1/VADIS) and accompanying vignettes.

xvii

1

Introduction

This book will establish a canon of state-of-the-art quantitative methods to responsibly and rigorously analyze variationist datasets from a comparative perspective. In this spirit, we will showcase various theoretically exciting intersections between variationist linguistics and related subfields, including dialectology and dialect typology, comparative linguistics, probabilistic linguistics, usage-based theoretical linguistics, psycholinguistics, and research on English as a world language. As a case study, we will distill key findings and methodological innovations from a five-year research project entitled “Exploring probabilistic grammar(s) in varieties of English around the world” about the scope and limits of grammatical variation in a global language such as English. In this book, we adopt the variationist methodology and take a particular interest in how people choose between “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188). In so doing, the book breaks new ground by marrying the spirit of Probabilistic Grammar research (which posits that grammatical knowledge is experience-based and partially probabilistic – see Grafmiller et al., 2018) to research along the lines of the English worldwide paradigm (which is concerned with the dialectology and sociolinguistics of postcolonial English-speaking communities around the world – see Schneider, 2007). The overarching objective, then, is to understand the plasticity of probabilistic knowledge of English grammar, on the part of language users with diverse regional and cultural backgrounds: how different are the ways a speaker of, say, British English chooses between different ways of saying the same thing (e.g. look up the word vs. look the word up) from how a speaker of, say, Canadian English chooses? To address this question, we investigate three grammatical alternations (see (1) to (3)) in some nine varieties of English around the world (British English, Canadian English, Irish English, New Zealand, English, Hong Kong English, Indian English, Jamaican English, Philippines English, and Singapore English).

1

2

(1)

Introduction

The genitive alternation in English: a.

b.

(2)

The dative alternation in English: a.

b.

(3)

Two other journalists who wrote a book criticising [the president] possessor ’s [brother]possessum were ordered to pay £6.3 million in fines (GloWbE AU B vexnews.com) (the s-genitive) Can you imagine a couple of years after WW2 the allies permitting [the brother]possessum of [the president]possessor bankrupting the central bank through embezzlement and getting away with it? (GloWbE GB G guardian.co.uk) (the of -genitive) A victim will be asked to giveverb [the police]recipient [a statement] theme explaining what has happened. (GloWbE CA G slsedmonton.com) (the ditransitive dative) Neither of them gaveverb [a statement]theme to [the police]recipient . (GloWbE JM G jamaicaobserver.com) (the prepositional dative)

The particle placement alternation in English: a.

b.

For all my second language readers: no need to lookverb [the word]NP upparticle in the dictionary . . . (GloWbE NZ B dedepuppets.com) (the split variant) Lookverb upparticle [the word]NP in a dictionary and write down its meaning in a vocabulary notebook. (GloWbE US G artofmanliness.com) (the continuous variant)

The analysis is mostly based upon observational corpus analysis but will be supplemented behaviorally by rating task experiments. Why do we need these two sources of evidence? On a practical level, corpora provide potentially massive amounts of data that can be analyzed at comparatively low cost (especially if the analysis relies on pre-existing corpora, as we do in this book). More substantially speaking though, corpora cover naturalistic language usage, not behavior in (more or less) artificial experiments, which is why corpus findings are ecologically valid in a way that experimental findings are not (see Campbell-Kibler, 2010 for discussion). On the other hand, experiments can directly target specific phenomena, variables, and constraints in a way that corpus analysis cannot. What is more, rating task experiments (the type of experiment we will be relying on) in particular explore metalinguistic judgments, a facet of linguistic competence that is not covered by corpora (which

Introduction

3

cover production and to some extent comprehension). Finally, it is always a good idea to strive for methodological pluralism (see Klavan and Divjak, 2016). Thus we aim to sketch a picture of probabilistic grammar variation across different native and nonnative varieties of English, and to develop a method for exploring indigenization patterns which builds upon established methods in comparative sociolinguistics while expanding our analytical toolkit to include methods common in dialectology and in psycholinguistics. The specific research questions that will guide our inquiry include the following: For a given alternation, how consistent are the probabilistic effects of the variable grammar’s constraints across varieties? Do some alternations vary more than others with respect to their probabilistic conditioning? Are there some (types of) constraints that are more variable than others? How and where to draw boundaries between distinct probabilistic variable grammars? To what extent can the patterns we observe in corpus data be replicated in rating task experiments? Do the crossvarietal patterns we find align with our current understanding of typological variation among varieties of English? In a nutshell, we may summarize the key findings to be discussed at length in the remainder of this book as follows. Probabilistic grammars across World Englishes are overall surprisingly stable: on a scale between 0 and 1, where 0 indicates total dissimilarity and 1 indicated total identity, the overall similarity of the alternation phenomena under study calculates as approximately 0.7. Effect directions are stable across varieties. If a particular constraint favors a particular grammatical outcome in a given variety, it will also do so in the other varieties. In contrast, strength of effects vary. For example, animacy may have strong effects on grammatical outcomes in variety A, but comparatively weaker effects in variety B . We will also see that different alternations are differentially hospitable to what we call probabilistic indigenization: for example, the particle placement alternation is (probably in function of its comparatively strong lexical anchoring) particularly malleable. On the interpretational plane, we often see a dialect-typological split between Inner Circle (ENL) and outer Circle (ESL) varieties. Finally, experiments and corpus analysis converge largely but not entirely. We note that variation studies of the kind presented in this book represent an increasingly popular area in linguistics – they are becoming increasingly entrenched in curriculum design; variation conferences are becoming ever larger and more numerous; and linguists more broadly are increasingly engaging with variation (see Nagy and Hoffman, 2017 for discussion). But this growing interest also means that the field is in danger of fragmenting into different research communities with different foci that do not necessarily talk as much to each other as they should. Against this backdrop, one of the aims of the book is to cross-pollinate different research tracks in variation studies, in one monograph with a coherent empirical focus. On the theoretical plane, the

4

Introduction

book prescribes a “balanced diet” (Guy, 2014, 65) to model and interpret variation as the association of conventional rules or constraints with probabilities learned from experience. In what follows, we briefly discuss some key concepts and research orientations that take center stage in the remaining chapters. 1.1

Variationist Sociolinguistics and Corpus-Based Variationist Linguistics

This book is an exercise in variation analysis. Specifically, we use the variationist method to study variation between grammatical variants that are in principle available to all members of the speech communities under study. The variationist method is designed to investigate quantitatively how speakers choose between “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188) as a function of properties of the linguistic contexts and of language-external factors. The variationist method is the cornerstone of the field of Variationist Sociolinguistics, also known as the Language Variation and Change (LVC) paradigm. Variationist Sociolinguistics is a research orientation in sociolinguistics pioneered by William Labov in the 1950s and 1960s (see e.g. Labov, 1963, 1966) dedicated to the rigorous and quantitative study of the interaction between linguistic variation and linguistic change based, typically, on observational data (for instance, corpora covering sociolinguistic interviews). Most work in variationist sociolinguists models the way language users choose between different ways of expressing the same meaning or grammatical function subject to both language-internal constraints (that is, properties of the linguistic context) and language-external constraints (such as age, gender, register, or geography). A key concept in variationist sociolinguistics is that of the linguistic variable (i.e. a particular meaning or function the expression of which is variable) and linguistic variants (particular forms which come under the remit of a particular variable). Over the last few decades, the variationist methodology has also become popular outside of Variationist Sociolinguistics proper. In particular, in corpus linguistics a new subfield has emerged that Szmrecsanyi (2017) calls corpus-based variationist linguistics (or CVL for short). Compared to other methodologies in corpus linguistics, the focus in CVL is on the conditioning of variation, and not so much on text frequencies. Compared to Variationist Sociolinguistics, CVL analysts tend to be more interested in language-internal constraints on variation than in language-external factors. Also, CVL analysts are more enthusiastic than variationist sociolinguists to consider other registers beside vernacular speech. What both CVL and variationist sociolinguistics share in common is that both orientations carefully define variables

1.2 Comparative Linguistics and Comparative Variation Analysis

5

and variants and follow the Principle of Accountability (Labov, 1969, 738) to understand the conditioning of variation. 1.2

Comparative Linguistics and Comparative Variation Analysis

Because we will be interested in the extent to which regional varieties of English are different or equivalents in terms of how language users make grammatical choices, this book also comes under the remit of comparative linguistics. Assessing the similarity or dissimilarity of language systems across varieties, dialects, or languages for that matter is an important topic of theoretical significance in comparative linguistics, including in crosslinguistic typology, dialectology, and sociolinguistics. There is sure enough a rich literature on how to assess such similarity. Much of this literature, however, is based on fairly decontextualized data that are more about competence than about usage, such as reference grammars, dialect atlases (such as e.g. The Survey of English Dialects; Orton and Dieth, 1962), or crosslinguistic surveys (such as e.g. The World Atlas of Language Structures; Dryer and Haspelmath, 2013). Data sources like these are valuable, but not of much use in variationist linguistics: they typically cover the inventory of forms and variants, but do not provide information about probabilistic variation patterns. Therefore, we will use comparative methods that are designed to investigate usage data about probabilistic variation patterns. We draw inspiration from two traditions of comparative variation analysis. The first tradition goes back to seminal work by Shana Poplack and Sali Tagliamonte (Poplack and Tagliamonte, 1989) and is now a subfield in variationist sociolinguistics known as comparative sociolinguistics (see Tagliamonte, 2001). The name of the game in comparative sociolinguistics is to investigate the conditioning of variation in a small number of varieties or dialects for the sake of determining if these varieties or dialects are historically related. The second comparative tradition that will inform the variation analyses presented here has been developed in corpus-based variationist linguistics (see Section 1.1) and builds on foundational work carried out by Joan Bresnan, Jennifer Hay, Lars Hinrichs and others in the 2000s (see Hinrichs and Szmrecsanyi, 2007; Bresnan and Hay, 2008). Here, the focus is not necessarily on historical relatedness – instead, the research questions addressed in this line of work are the following: are varieties diverging or converging? What are the constraints that are particularly stable or unstable across varieties? What can differences and similarities across varieties tell us about the nature of knowledge that language have about probabilistic grammars? Of note, comparative variation analysis in this spirit can be backed up by rating task experiments (Bresnan and Ford, 2010), and this is exactly what we are going to do in this book as well.

6

Introduction

1.3

Dialectology, Dialectometry, and Dialect Typology

Dialectology, dialectometry, and dialect typology are to some extent also comparative endeavors, and this book draws inspiration from all three (sub)fields. Dialectology is concerned with the study of regional varieties of language and has a long history that goes back at least to the nineteenth century (see Chambers and Trudgill, 1998). This book is inspired by work in dialectology in that what will take center stage is regional varieties of English. That said, we hasten to add that while traditional dialectology focusses on primarily phonological or lexical features of rural dialects spoken by nonmobile old rural males (NORMs), we will be interested in the grammar of more acrolectal international standard varieties of English spoken and written by all kinds of language users. We also deviate from standard practice in traditional dialectology in that our empirical analysis is not based on questionnaires or survey data (which are the customary datasources in traditional dialectology – see Anderwald and Szmrecsanyi, 2009), but on corpora and experiments. Dialect typology (also known as sociolinguistic typology) is a subfield in dialectology that explores the intersection between dialectology and typology. Typologists seek to categorize human languages based on their structural differences and similarities, and similarly dialect typologists take an interest in categorizing dialects and varieties of the same language. Sometimes dialect typology is used interchangeably with sociolinguistic typology, and in sociolinguistic typology in particular there is an interest in the “extent to which differences of linguistic structure, whether within or between languages, can be ascribed to or explained in terms of features of the society in which the dialects in question are spoken” (Trudgill, 1996, 3; see also Trudgill, 2011; Röthlisberger and Szmrecsanyi, 2019). Dialect typology will play a role in this book because we will systematically distinguish between native L1 varieties of English (such as Canadian English and New Zealand English) and indigenized L2 varieties of English (such as Indian English or Hong Kong English). Dialectometry is a subfield in dialectology that specializes in measuring, visualising, and analysing aggregate dialect similarities or distances as a function of properties of geographic space (seminal work includes Séguy, 1971, Goebl, 1982, and Nerbonne et al., 1999). Thus, whereas traditional dialectologists study in depth a typically small number of features deemed interesting in a typically correspondingly small number of dialects, dialectometricians explore relationships between a large amount of dialect locations based on a large amount of features. In this endeavor, dialectometrical analysis strongly relies on quantification, cartographic visualisation and exploratory data analysis for the sake of inferring patterns from feature aggregates. This book draws inspiration from dialectometry in that we study multiple grammatical alternations in multiple varieties of English, with one of our aims being the discovery of general patterns in a bird’s eye perspective.

1.4 Probabilistic Linguistics and Probabilistic Grammar

1.4

7

Probabilistic Linguistics and Probabilistic Grammar

Probabilistic Linguistics is a research orientation whose point of departure is that probabilistic patterns and gradience have been shown to be pervasive on all levels of language. Given this pervasiveness, Probabilistic Linguistics seeks to complement more traditional structural/categorical/generative theorizing by exploring the extent to which gradient rules, to be discovered through quantitative modeling using the mathematics of uncertainty, can predict (aspects of) linguistic knowledge and of linguistic usage (see the papers in Bod et al., 2003 for more discussion and case studies). It is clear that quantitative variationist (socio)linguistics (see Section 1.1) is essentially a variety of Probabilistic Linguistics. Variationist sociolinguists, after all, have been busy analyzing variation patterns probabilistically for decades (see e.g. Cedergren and Sankoff, 1974). Needless to say, this book is (among other things) engaging in Probabilistic Linguistics. And even more specifically, this book will engage in (comparative) Probabilistic Grammar analysis (see Grafmiller et al., 2018, which is summarized in the following discussion). As our literature review in Chapter 2 will show, it is amply documented that (morpho)syntactic variation within and across varieties of the same language is very systematic, and that the determinants of this variation are numerous, multifactorial, and probabilistic in nature. Against this backdrop, this book endeavors to systematically asses the scope and limits of differentiation between probabilistic grammars in a world language such as English. Crucially, this includes an experimentalist inquiry into the extent to which language users’ knowledge about probabilistic grammars differs as a function of geography and/or regional identity. Now, current theorizing about the nature of grammatical knowledge is often trapped in an opposition between fully usage-based (i.e. exemplar-based) approaches (e.g. Pierrehumbert, 2006) and fully rule-based approaches (e.g. Chomsky and Halle, 1968). We do not think that this dichotomy is productive. We are certainly committed to the usage-based notion that grammar is the “cognitive organization of one’s experience with language” (Bybee, 2006) – most probabilistic approaches to analyzing variation are actually or inherently usage-based, in that they capitalize on statistical regularities likely derived from experience, yet they associate these quantitative patterns not (only) with surface forms or lexical items (as in pure exemplar models), but with abstract features or constraints. But beyond this commitment, we submit that a hybrid model is necessary to account for variation in all its complexities (see Guy, 2014 for discussion). Of note, both usage- and rule-based models of grammar are mentalistic, in that they view language as a cognitive object, and this common ground is shared by most hybrid models. The work we report is especially inspired by the variation-centered, usageand experience-based Probabilistic Grammar approach developed by Joan

8

Introduction

Bresnan and collaborators (e.g. Bresnan, 2007 and follow-up work reviewed in Chapter 2). This work, and our work, makes two key assumptions (see Grafmiller et al., 2018, 2–3): First, grammatical knowledge is partially probabilistic in nature, and language users have demonstrably powerful predictive capabilities. Second, this probabilistic knowledge is acquired through language experience, and so is subtly, but dynamically (re)constructed throughout speakers’ lives. Needless to say, regional differentiation of the type that takes center stage here is predicted. Note also that this approach is hybrid in nature, as it assumes that conventional rules or constraints are associated with probabilities learned from experience. In other words, the approach we will adopt follows a “balanced diet” (Guy, 2014, 65) by modeling syntactic variation drawing on both qualitative and quantitative aspects. 1.5

Psycholinguistics

Psycholinguistics is a discipline situated at the intersection between psychology and linguistics. Psycholinguists investigate the psychological processes that enable language users to produce, comprehend, process, and acquire language, and the way in which language users store knowledge about language (see e.g. Kennison and Messer, 2014). This book builds on key insights and makes use of methodologies from psycholinguistics in a number of ways. For one thing, we will be interested in the probabilistic nature of (knowledge of) grammar, an issue that is needless to say inherently relevant to how language is produced and processed, and to how knowledge about language is stored. Secondly, a number of language-internal constraints on variation that we consider in this book relate to language processing. Consider, for example, priming effects, or surprisal effects. Priming is about the tendency that language users have a preference for reusing syntactic patterns that they have produced or have been exposed to in previous discourse (see e.g. Gries, 2005). Surprisal is about the extent to which material in different slots of constructions tends to co-occur, and about the consequences that these co-occurrence preferences have for syntactic placement preferences (see e.g. Levy and Jaeger, 2007). Third, we will supplement corpus-based analysis (the customary methodology in variationist (socio)linguistics) with rating task experiments to spot-check the psychological plausibility of our findings. The rating task experiments that we will conduct are inspired by Bresnan (2007, 76–84). That study used a scalar rating task based on corpus materials as stimuli to model subjects’ responses regarding the naturalness of syntactic variants in context. Responses were compared to the predictions of a parallel regression model fitted on corpus data. Analysis showed that subjects’ gradient naturalness ratings correlated with corpus-generated probabilities. This demonstrates that language users’

1.6 English as a World Language

9

implicit knowledge about language must be to some extent probabilistic in nature. Corpora are an observational data source that covers language production as well as, to some extent, language comprehension (because in naturalistic settings, whatever is spoken/written is designed to be also comprehended). Rating task experiments, on the other hand, tap into subjects’ intuitions about the naturalness of grammatical variants given a real-life context. We will thus cover language production, language comprehension, as well as the predictive capacities of language users. The methodological diversity that this book aims for is ultimately motivated by the quest for ecologically valid paradigms (see Klavan and Divjak, 2016 for discussion). In short, we fully agree with (Dabrowska, ˛ 2016a, 488): Corpus analysis is absolutely vital to usage-based approaches . . . . In the end, however, corpora can only provide information about frequency of items and frequency of cooccurrence of items. If we want to make claims about speakers’ mental representations, corpus data needs to be complemented with experimental research.

1.6

English as a World Language

In this book, we cover variation in multiple varieties of English around the world. Thanks to a history of colonial expansion and other favorable sociohistorical circumstances, English is fairly unique in having diversified into a language with a wide range of postcolonial varieties of English (a.k.a. “New Englishes”) around the world. This diversity is particularly interesting from the point of view of dialect typology (see Section 1.3), and so we cover both native mother-tongue, “Inner Circle” (Kachru, 1992) varieties (e.g. New Zealand English) as well as non-native indigenized second-language “Outer Circle” (Kachru, 1992) varieties (e.g. Hong Kong English). Research on the scope and limits of phonological and grammatical variation within and across varieties of English around the world has in recent years engendered a lively research field. Key results of research on the “English language complex” (see McArthur, 2003, 56; Mesthrie and Bhatt, 2008, 1–3) include the finding that the structural make-up of postcolonial varieties of English can be predicted by the particular communicative needs of the colonizers and the colonized (Schneider, 2007), and that there is a “World System of Englishes” in which differential associations with prestige shape the hierarchical structure of, and relationships between, World Englishes (Mair, 2013). Variationist research activity on World Englishes is clearly picking up (see Chapter 2 for a review), but a shortcoming of much previous research on the English language complex is an often primarily descriptive interest in the variable presence or absence of particular features in particular varieties, or in usage frequencies of grammatical patterns. But while feature inventories and

10

Introduction

usage frequencies are no doubt interesting, they do not necessary address the most interesting part of the story, which is: Do language users’ probabilistic grammars differ across varieties of English, and if so, to what extent? This is the central question that we are going to address in this book. 1.7

Structure

The remainder of this book is structured as follows: Chapter 2 surveys the literature on variation in general and on grammatical variables (a.k.a. “alternations”) in particular. Next, we review well-known grammatical variables/alternations in English as well as previous comparative investigations of grammatical alternations in English. Last but not least, we discuss in detail previous variationist work on the three alternations subject to study here: the genitive alternation, the dative alternation, and the particle placement alternation. Chapter 3 begins with a review of the World Englishes and dialect typology literature. Next, we introduce the nine regional varieties of English under study in the book with a brief summary of relevant aspects of their sociohistories and linguistic profiles: British English, Canadian English, Irish English, New Zealand English, Hong Kong English, Indian English, Jamaican English, Philippines English, and Singapore English. These varieties are a fairly representative sample covering both native (or “Inner Circle” – see Kachru, 1992) varieties and nonnative (or “Outer Circle”) varieties. Chapter 4 kicks off with a general discussion of the common data types used in variationist linguistics. Next, we present the primary data sources we use in the study. To study variation in production, we tap into corpus data from the International Corpus of English (ICE) (Greenbaum, 1991) and the Corpus of Global Web-based English (GloWbE) (Davies and Fuchs, 2015). After introducing the corpora, we describe the procedures for identifying and extracting interchangeable tokens of each alternation, and detail the annotation procedures for a wide range of constraints including, for example, the principle of end weight (longer constituents tend follow shorter constituents) and animacy effects (animate constituents tend to occur early). Chapter 5 interrogates corpus data to analyze the three alternations subject to study one-by-one using a battery of state-of-the art analysis techniques, including – in addition to customary descriptive statistics – conditional random forest modeling and mixed-effects logistic regression analysis. The goal of the chapter is to uncover qualitative generalizations: for example, we see that while effect directions of constraints on variation are generally stable across varieties of English, effects strengths can be significantly different.

1.7 Structure

11

Chapter 6 is inspired by work in comparative sociolinguistics and quantitative dialectometry. We use a corpus-based method (Variation-Based Distance and Similarity Modeling – VADIS for short) to quantify the similarity between, and coherence across, the varieties of English under study as a function of the correspondence of the ways in which language users choose between different ways of saying the same thing. Key findings include the result that probabilistic grammars are remarkably stable across varieties, but that coherence across alternations is not perfect. Chapter 7 examines the extent to which contrasts uncovered in the corpus analyses in Chapters 5 and 6 can be replicated in an experimental acceptability judgment task. To compare ratings to corpus model predictions, we use a variant preference rating task modeled after the work of Bresnan and Ford (2010) in which participants rate the naturalness of alternative syntactic forms by distributing points between two alternatives. This experimental paradigm is relatively new but increasingly popular. The hypothesis is that the ratings suggested by participants (who are provided with the surrounding context of the given corpus example) correlate significantly with the probabilities predicted by the corpus models. The results are in line with our expectation – the splits people suggest are typically in line with the splits predicted by corpus-based regression models. Our results provide further evidence that linguistic choices in both production and comprehension are sensitive to the quantitative distributions of various contextual cues, that is, that grammatical knowledge is to some extent probabilistic. However, there remain some subtle discrepancies between our ratings data and corpus models that raise important questions about the comparability of different observational and experimental methods. We consider a number of these questions in the discussion. Chapter 8 summarizes the study’s key findings, and discusses these findings against the backdrop of the various frameworks to which the book is relevant, including World Englishes research and (Labovian) variationist sociolinguistics, but also, for example, dialectometry (e.g. Nerbonne et al., 1999) and general usage, end experience-based linguistics (e.g. Bybee, 2010). We also highlight the application potential of the methodological innovations presented in the book, and conclude with some general reflection on where the road ahead may lead.

2

Grammatical and Syntactic Variation

This chapter reviews the literature on grammatical variables (a.k.a. “alternations”), including from the perspective of traditional dialectology, and of modern variationist linguistics. This leads us to a brief review of well-known grammatical variables/alternations in English, a survey of previous comparative investigations of grammatical alternations in English, and finally to the presentation of the three grammatical (and more specifically, syntactic) alternations that will take center stage in the empirical chapters of the book: the genitive alternation, the dative alternation, and the particle placement alternation.

2.1

Does Grammatical Variation Even Exist?

In this book we are interested in what van Hout and Muysken (2016, 250) call Type 3 variability, that is, “[v]ariability in the linguistic signal within a given language” – or, in Labovian parlance, “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188). Now, whether or not that kind of variability exists in the realm of grammar (a term that in what follows we use synonymously with morphosyntax) has been controversial for a long while. Consider that the study of variation (and change) in pronunciation has a long history in linguistics, going back to seminal work on sound changes in (proto)languages such as Indo-European and Germanic (e.g. Grimm’s law and Verner, 1877). Lexical variation has likewise concerned linguists for a long while. By contrast, grammatical variation in the sense that we are interested in – about choices languages users have between different grammatical constructions to express the same meaning or function – began receiving interest considerably later. There are several reasons why. For one thing, grammatical variation is arguably more abstract and thus less obvious to most people than pronunciational or lexical variation. Secondly, historically speaking there has been a gut feeling among some or even many descriptive linguists that grammatical variation – plain and simple – does not or should not exist, and that if it does exist it should be short-lived diachronically.1 In variationist circles, this gut feeling is 1 We hasten to add that prescriptive linguistics and language mavens have been commenting

on variation between “good” and “bad” ways of saying the same thing for centuries, also and

12

2.1 Does Grammatical Variation Even Exist?

13

also known as the “Doctrine of Form-Function Symmetry” (see Poplack, 2018, 7 for discussion). Take the well-known Principle of Isomorphism, formulated by Haiman citing authors such as Bloomfield and Bolinger, among others: . . . the commonly accepted axiom that no true synonyms exist, i.e. that different forms must have different meaning . . . (Haiman, 1980, 516)

This principle simply denies that different ways, lexical or grammatical,2 of saying the same thing exist. The same idea fuels the “Principle of No Synonymy” in Construction Grammar à la Adele Goldberg: If two constructions are syntactically distinct, they must be semantically or pragmatically distinct. (Goldberg, 1995, 67)

Thus again the “Principle of No Synonymy” precludes the existence of different grammatical or syntactic ways of saying the same thing, and its impact in Construction Grammar and Cognitive Linguistics circles should not be underestimated.3 What is more, axioms such as the Principle of Isomorphism and the Principle of No Synonymy have also shaped thinking in historical linguistics, as De Smet et al. critically observe: The relation between functionally similar forms is often described in terms of competition. This leads to the expectation that over time only one form can survive (substitution) or each form must find its unique niche in functional space (differentiation). (De Smet et al., 2018, 197)

However, as De Smet et al. (2018, 201) point out, “[i]f isomorphism were the only force shaping communicative codes, synonymy . . . should be consistently rooted out in diachrony.” But it is not the case that grammatical variation phenomena are necessarily short-lived diachronically: for example, all three alternations subject to study in the present book have been around for centuries (see Section 2.5 for discussion). But the Doctrine of Form–Function Symmetry appears to be on shaky empirical grounds also in synchrony: the underlying assumption that believers in form–function symmetry often (implicitly) make is that grammatical variation is suboptimal, (needlessly) complex, cognitively costly, and therefore inconvenient for language users. Against this backdrop particularly in the realm of grammar (see e.g. Anderwald, 2016 for discussion). We stress that when we diagnose a certain reluctance to accept the existence of grammatical variation, we exclusively refer to work in descriptive/theoretical linguistics. 2 Haiman clarifies that he is talking about “one-to-one correspondence[s] between the signans and the signatum, whether this be a single word or a grammatical construction” (Haiman, 1980, 515). 3 Consider, for example, that even recent work about grammatical variation published in journals such as Cognitive Linguistics regularly includes disclaimers such as the following: “Strictly speaking, there are no so-called syntactic alternations in natural languages. The semantics of alternative forms is never fully equivalent (Goldberg, 2002) . . . ” (Fang and Liu, 2021, 2, in a paper about syntactic variation in Mandarin Chinese).

14

Grammatical and Syntactic Variation

Gardner et al. (2021) investigate corpus data to check whether variable grammatical patterns attract production difficulties in the form of disfluencies (i.e. filled or unfilled pauses), as variable patterns must if the Doctrine of FormFunction Symmetry were to have a cognitive basis. However, Gardner et al. find no evidence whatsoever of a relationship between grammatical variation and disfluency. Therefore, it does not appear that grammatical variation is difficult to handle for language users. For us the question of whether or not grammatical variation exists is an empirical one – and the empirical record, which includes a massive body of literature in grammatically oriented variationist (socio)linguistics and neighboring fields, strongly suggests that there is, in fact, “[Form-Function] asymmetry in the form of robust variability subject to regular conditioning” (Poplack, 2018, 7). 2.2

From Traditional Dialectology to Variationist Linguistics

Reservations about grammatical variation of the type discussed in the previous section have led to the situation that even fields specializing in analyzing variation, such as dialectology and variationist sociolinguistics, have traditionally prioritized variation on other linguistic levels than grammar. Early dialectology was theoretically inspired by the Neogrammarian theorem of the “exceptionlessness of sound change” (Ausnahmslosigkeit der Lautgesetze – see Osthoff and Brugmann, 1878, XII–XIII), and so it is maybe not surprising that traditional dialect studies have relied mostly on questionnaires to elicit data about lexis and – in particular – pronunciation. Consider, for example, the Survey of English Dialects (SED) (Orton and Dieth, 1962): in this atlas project, only a small proportion of questions is dealing with morphosyntactic variation. The fact of the matter is that “modern dialectology has until recently largely neglected the field of syntax” (Görlach, 1999, 493). As Kortmann wrote in 2002, “Even today, when we look at current AngloAmerican dialect research, there is no denying that the study of dialect syntax still constitutes no more than a sideline” (Kortmann, 2002, 187). The qualifier “Anglo-American” is important, because in the generative community the emergence of the Principles and Parameters framework in the 1980s triggered much interest in syntactic microvariation between regional varieties of the same language (Brandner, 2012); the bulk of the relevant literature deals with Italian, Dutch and Flemish dialects. For non-generative dialectological circles, to the extent that the existence of grammatical variation is acknowledged, there continues to be a sense that grammatical variation exhibits relatively weak geographical patterns compared to other levels of language structure. For example, Lass (2004, 374) writes that “English regional phonology and lexis . . . are generally more salient and defining than regional morphosyntax” (see also Wolfram and Schilling-Estes, 1998, 161 and Löffler, 2003, 116).

2.3 Grammatical Variation in English

15

With all that being said, we also need to keep in mind that dialectologists tend to be primarily interested in Type 4 variability according to the categorization in van Hout and Muysken (2016, 250), that is, “inter-individual variability”: what are the forms that individuals in location X use, as opposed to those forms that individuals in location Y use? By contrast, Type 3 variability (“[v]ariability in the linguistic signal within a given language,” van Hout and Muysken, 2016, 250) – or, in Labovian parlance, variability between “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188) – is not something that traditional dialectologists much care about. The discipline in variation studies that does care both about Type 3 and Type 4 variability, is variationist sociolinguistics. Early work in variationist sociolinguistics was often focused on pronunciational variation. In the 1970s analysts started to take more interest in grammatical variation as well (see e.g. Weiner and Labov, 1983 for early work on the active/passive alternation). This sparked a debate in the variationist sociolinguistics community as to whether linguistic variables can be grammatical in nature: Lavandera (1978) argued that the concept of the linguistic variable cannot be easily extended to linguistic levels other than phonology because “difficulties arise from the fact that non-phonological variation involves referential meaning” (p. 181); Labov (1978) replied that while it may be hard to work out the envelope of variation, it is in principle not impossible to extend the linguistic variable to levels such as grammar. The consensus that emerged after this debate is, then, that the concept of the linguistic variable can indeed be applied to linguistic levels other than phonology because “distinctions in referential value or grammatical function among different surface forms can be neutralized in discourse” (Sankoff, 1988, 153). Work on grammatical variation has boomed especially in Canada (see e.g. Poplack and Dion, 2009; Tagliamonte, 2014) and in the United Kingdom (see e.g. the papers in Beaman et al., 2021) and in variationist sociolinguistic work on languages such as Spanish (see e.g. Travis and Torres Cacoullos, 2012). But the fact remains that phonology is going strong in the variationist sociolinguistics community – certainly stronger than grammar. By contrast, grammatical variation has been a focus from early on in corpus-based variationist linguistics (see e.g. Gries, 2005; Bresnan et al., 2007; Grondelaers and Speelman, 2007 for seminal work). In that community, grammatical “variables” (in variationist sociolinguistics parlance) are typically known as grammatical “alternations,” defined as “structurally and/or lexically different ways to say functionally very similar things” (Gries, 2017, 7).

2.3

Grammatical Variation in English

Concerns about variation in the domain of grammar notwithstanding, the upshot is that grammatical variation in languages such as English is now quite

16

Grammatical and Syntactic Variation

well documented and understood. In nonstandard English, well-known grammatical variation phenomena include the following (many of which are also covered in surveys such as Kortmann and Szmrecsanyi, 2004 and Kortmann and Lunkenheimer, 2013): • copula retention versus deletion (as in she is the first one vs. she the first one; Labov, 1969) • standard versus nonstandard reflexives (as in they didn’t go themselves vs. they didn’t go theirself ; Hernández, 2010) • demonstrative those versus demonstrative them (as in in those days vs. in them days; Britain, 2010) • be versus have as perfect auiliaries (as in I’m come down to pay the rent vs. I’ve come down to pay the rent; Tagliamonte, 2000) • a-prefixing on -ing-forms (as in he was a-waiting there versus he was awaiting there; Wolfram, 1976) • nonstandard verb forms (as in he knowed versus he knew; Szmrecsanyi, 2013a, features 28–30) • standard negators versus ain’t (as in people haven’t got no money vs. people ain’t got no money; Anderwald, 2003) • multiple negation versus standard negation (as in people haven’t got no money vs. people haven’t got any money; Anderwald, 2005) • the was–weren’t split (as in It wasn’t very dead, no it were just busy; Britain, 2002, 19) • was–were variation (as in Three of them was killed vs. three of them were killed; Tagliamonte, 2009) • standard versus non-standard verbal -s (as in I say vs. I says; Britain, 2010) • the kena-passive in SgE (as in John kena scolded vs. John was scolded; Bao and Wee, 1999) • unstressed preverbal did (as in I did always eat vs. I always ate; Jones and Tagliamonte, 2004). We now know that there is likewise plenty of grammatical variation in more acrolectal, standard English of the type that is investigated in this book. Gardner et al. (2021), in their comprehensive study of the (nonexisting) relationship between grammatical variation and disfluency in mainstream US American English, study the following catalogue of grammatical variables/alternations (note that a corresponding list for British English or other varieties of English would look fairly similar): • indefinite pronouns (as in everyone would know vs. everybody would know; D’Arcy et al., 2013) • case and order of coordinated pronouns (as in my husband and I vs. me and my husband; Angermeyer and Singler, 2003)

2.3 Grammatical Variation in English

17

• that versus zero complementation (as in I don’t think it’s a deterrent at all vs. I don’t think that it’s a deterrent at all; Tagliamonte and Smith, 2005) • infinitival versus gerundial complementation (as in I love to play racquetball vs. I love playing racquetball; Mair, 2003) • complementation after remember, regret, and deny (as in I don’t remember being that picky vs. I don’t remember that I was that picky; Cuyckens et al., 2014) • complementation after the verb to try (as in I try to get you some popcorn vs. I try and get you some popcorn; Brook and Tagliamonte, 2016) • particle placement (as in to pay off their credit cards vs. to pay their credit cards off ; Gries, 2003) • the dative alternation (as in someone’s given me one vs. someone’s given one to me; Bresnan et al., 2007) • the genitive alternation (as in they took someone else’s life vs. they took the life of someone else; Hinrichs and Szmrecsanyi, 2007) • relativization (as in the house that you have to have money for vs. the house which you have to have money for; Hinrichs et al., 2015) • analytic versus synthetic comparatives (as in I think in Dallas it’s a lot more scary vs. I think in Dallas it’s a lot scarier; Hilpert, 2008) • plural existentials (as in there’s some places vs. there are some places; Chambers, 2004) • future temporal reference (as in eventually it will happen vs. eventually it’s going to happen; Torres Cacoullos and Walker, 2009) • deontic modality (as in I must admit vs. I got to admit; Tagliamonte and Smith, 2006) • stative possession (as in you got a Louisiana accent vs. you have got a Louisiana accent; Tagliamonte et al., 2010) • quotatives (as in I was like, “No way!” and she goes, “Yeah!”; Gardner et al., 2020) • not versus no negation (as in There’s nobody to sit them down vs. There isn’t anybody to sit them down; Childs, 2017) • negative versus auxiliary contraction (as in he wouldn’t admit it to you vs. he’d not admit it to you; MacKenzie, 2013). We add that, for practical reasons, Gardner et al. (2021) do not include grammatical alternations for which potentially every utterance would constitute a variable context. Such alternations include the active/passive alternation (as in they broke into the liquor closet vs. the liquor closet was broken into; Weiner and Labov, 1983), the progressive alternation (as in I love it vs. I’m loving it; Hundt, 2004), and variation between omitted versus overt pronominal subjects (as in Tom says hi. Asks how you are doing vs. Tom says hi. He asks how you are doing; Torres Cacoullos and Travis, 2014). Other grammatical alternations in standard English that have been studied elsewhere include would versus

18

Grammatical and Syntactic Variation

used to as markers of habitual past (as in I used to dance vs. I would dance; Tagliamonte, 2000), and the mandative subjunctive alternation (as in the law requires that the doors be shut vs. the law requires that the doors must be shut; Kastronic and Poplack, 2014). We conclude our survey of grammatical variation in English by noting that grammatical variables/alternations can be categorized into three general types (see De Troij, in preparation): Permutation alternations Permutation alternations allow language users to manipulate constituent ordering. Consider (4). In (4-a), the pronoun follows the full NP while in (4-b), it precedes; constituent order is variable. (4)

The case and order of coordinated pronouns alternation: a. . . . you give Al Gore and I a chance to bring America back. (Angermeyer and Singler, 2003, ex. (1a)) (NP + pronoun order) b. in the interchanges between she and Chairman Fazio . . . (Angermeyer and Singler, 2003, ex. (1b)) (pronoun + NP order)

Insertion/deletion alternations. These alternations give language users the option of retaining or omitting functional markers or patterns. Consider (5). In (5-a), the complementizer that is omitted; in (5-b), it is retained. (5) The complementizer retention/omission alternation: a. To prove ∅ I could do it. . . . (Tagliamonte and Smith, 2005, ex. (1)) (complementizer omission) b. . . . Yes, I had to prove that I could do it. (Tagliamonte and Smith, 2005, ex. (1)) (complementizer retention) Substitution alternations. Substitution alternations replace one particular function word or functional construction by another function word or functional construction. Consider (6). In (6-a), the speaker uses the future marker will instead of be going to; vice versa for (6-a). (6) The future marker alternation: a. And he’ll probably live ’til a hundred. (Torres Cacoullos and Walker, 2009, ex. (1c)) (the future marker will) b. My doctor tells me I’m going to live ’til a hundred. (Torres Cacoullos and Walker, 2009, ex. (1d)) (the future marker be going to)

2.4 Comparative Perspectives on Grammatical Variation in English

2.4

19

Comparative Perspectives on Grammatical Variation in English

In this section, we review previous comparative work on grammatical variation in English (see Section 1.2): how do grammatical choice-making processes in location/variety/dialect A differ from those in location/variety/dialect B? The field of comparative sociolinguistics is too large to review exhaustively here, therefore our discussion will necessarily be somewhat selective. As to comparisons involving vernacular dialects in North America, for example, in a seminal study Poplack and Tagliamonte (1989) examine variation in verbal -s inflection in two corpora on early Black English from a historical and comparative perspective; Walker (2007) studies variable agreement in existentials with plural reference in varieties of Quebec English; and Gardner (2017) uses the comparative method across three grammatical variables to asses the genetic similarity between Cape Breton English (Canada) and its two potential progenitors, Scottish English and United Empire Loyalist English. As to comparisons involving vernacular dialects in the United Kingdom, Tagliamonte and Smith (2002) investigate negative/auxiliary contraction in eight communities (Tiverton, Henfield, York, Wheatley Hill, Maryport, Cullybackey, Cumnock, Buckie); Jones and Tagliamonte (2004) compare variation patterns between unstressed preverbal did and plain verb forms in Somerset to those in Samaná in the Dominican Republic, showing that the similarities are due to diffusion; and Childs (2017) studies variation between negative constructions in three UK locations (Glasgow, Tyneside, Salford). Beyond the United Kingdom and North America, we would like to mention Tagliamonte et al. (2016a), who study quotative variation in four speech communities around the world: Toronto (Canada), Victoria (Canada), Christchurch (New Zealand), and Perth (Australia). In less sociolinguistically and more theoretically oriented circles, analysts interested in the probabilistic nature of knowledge about grammar and grammatical variation have recently also taken an interest in the comparative analysis of variation data. In this spirit, Bresnan and Hay (2008) explore the dative alternation in US American English (based on the Switchboard Corpus) and New Zealand English (based on the Origins of New Zealand English corpora) and find that a US regression model generalizes remarkably well to the New Zealand data, although New Zealand English speakers are more sensitive to the animacy constraint. Szmrecsanyi et al. (2017) investigate two alternations – the dative as well as the genitive alternation – in four varieties of spoken English: US American English (based on materials from the Switchboard Corpus and the Corpus of Spoken American English), British English (materials from the Freiburg English Dialect Corpus, and materials collected in the United Kingdom between 1997 and 2010 by Tagliamonte), Canadian English (materials from the sociolinguistic interviews in the Ontario Dialects Archive), and New Zealand English (materials from the Canterbury Corpus of

20

Grammatical and Syntactic Variation

the Origins of New Zealand English). The study concludes that “while there are a number of subtle probabilistic contrasts between the regional varieties under study, there is overall a striking degree of cross-varietal homogeneity” (Szmrecsanyi et al., 2017, 1). Meanwhile, scholars interested in divergence/convergence patterns have endeavored to enrich comparative variation analysis with the addition of real time as an interacting language-external factor. Consider Hinrichs and Szmrecsanyi (2007), who study the genitive alternation in written-editedpublished corpus materials from the Brown family of corpora, contrasting both American English against British English and 1960s usage against 1990s usage; multivariate analysis indicates that the spread of the s-genitive in both British English and American English is primarily due to economy-related factors, while one of the reasons why the s-genitive is now more popular in American English than in British English is that the animacy constraint is weaker in American English than in British English. Hinrichs et al. (2015) investigate the same data source to study relativizer variation (which vs. that) in late twentieth-century written-edited-published English, likewise using comparative multivariate regression analysis. Results show that the welldocumented spread of relative that in restrictive relative clauses is what Hinrichs et al. (2015, 806) call a case of “colloquialization-cum-Americanization.” Going a bit farther back in real time, Hundt and Szmrecsanyi (2012) investigate the progressive alternation and the genitive alternation in nineteenth- and early twentieth-century written New Zealand English (tapping into A Corpus of Early New Zealand English), as well as in British English and American English of the same period (tapping into A Representative Corpus of Historical English Registers – ARCHER); it is shown that there are differences between earlier New Zealand English and the other varieties in terms of the effect that animacy has on grammatical variation, similar to what Bresnan and Hay (2008) report. Finally, tapping into ARCHER also, Wolk et al. (2013) study the development of the dative and the genitive alternations during the late Modern English period (sixteenth to twentieth century). Their dative alternation materials include data from both British English and American English. It turns out that in American English, increasing theme length has come to disfavor the prepositional dative more robustly than in British English. Lastly, within the World Englishes community comparative multivariate/probabilistic variation analysis is a relatively new addition to the analytical toolbox. Gries and Bernaisch (2016) model the dative alternation in six South Asian Englishes (English in Bangladesh, India, the Maldives, Nepal, Pakistan and Sri Lanka), based on the SAVE Corpus. Based on a regression-based technique – Multifactorial Prediction and Deviation Analysis with Regression (MuPDAR) – the study reveals that Indian English is the linguistic epicentre of South Asian English. In a similar spirit, Heller et al. (2017a) investigate the genitive alternation in six varieties of English (English in Great

2.5 Grammatical Alternations Subject to Study in This Book

21

Britain, Hong Kong, India, the Philippines, Singapore, and Sri Lanka) as represented in the International Corpus of English (ICE); based on a conditional random forest modeling-based approach (Multifactorial Prediction and Deviation Analysis Using Regression/Random Forests, or MuPDARF for short), Heller et al. argue, as Gries and Bernaisch (2016) do, that Indian English has epicenter status. Moving on, Levshina (2018) studies complementation patterns after the verb to help in seven varieties of English (English in Australia, Ghana, Great Britain, Hong Kong, India, Jamaica, and the United States) covered in the Corpus of Global Web-based English (GloWbE); what the study demonstrates via Bayesian regression is that universally, predictability of units in discourse is an important constraint on the variation, but that this being said “the strength, shape and directionality of predictability effects exhibit variation across the countries” (Levshina, 2018, 1). Davydova (2019) investigates quotative like in materials obtained from two student communities, one in Outer Circle India (Jawaharlal Nehru University), and the other in Expanding Circle Germany (University of Mannheim). Regression analysis shows that “the patterns of use of innovative be like are uniform in the speech of young adults with elevated levels of exposure to mass media products and apply across the Outer and Expanding Circle English board” (Davydova, 2019, 578). Finally, we would like to mention Tamaredo et al. (2019), who tap into the International Corpus of English (ICE) to explore the degree of regional variability concerning the probabilistic conditioning of four grammatical alternations in English (the genitive, dative, particle placement, and subject pronoun omission alternations) in British, Indian and Singapore English; Tamaredo et al. report that the genitive alternation is the most stable/homogeneous one across the three varieties, while the particle placement alternation is most amenable to probabilistic indigenization.

2.5

Grammatical Alternations Subject to Study in This Book

Of the many alternations in the grammar of English (see above), we investigate three: The genitive alternation, as in (1) (reprinted as (7) below); the dative alternation, as in (2) (reprinted as (8) below); and the particle placement alternation, as in (3) (reprinted as (9) below). (7) The genitive alternation in English: a. Two other journalists who wrote a book criticising [the president] possessor ’s [brother]possessum were ordered to pay £6.3 million in fines (GloWbE AU B vexnews.com) (the s-genitive)

22

Grammatical and Syntactic Variation

b. Can you imagine a couple of years after WW2 the allies permitting [the brother]possessum of [the president]possessor bankrupting the central bank through embezzlement and getting away with it? (GloWbE GB G guardian.co.uk) (the of -genitive) (8) The dative alternation in English: a. A victim will be asked to giveverb [the police]recipient [a statement] theme explaining what has happened. (GloWbE CA G slsedmonton.com) (the ditransitive dative) b. Neither of them gaveverb [a statement]theme to [the police]recipient . (GloWbE JM G jamaicaobserver.com) (the prepositional dative) (9) The particle placement alternation in English: a. For all my second language readers: no need to lookverb [the word]NP upparticle in the dictionary . . . (GloWbE NZ B dedepuppets.com) (V-DO-P, a.k.a. the split variant) b. Lookverb upparticle [the word]NP in a dictionary and write down its meaning in a vocabulary notebook. (GloWbE US G artofmanliness.com) (V-P-DO, a.k.a. the continuous variant) All three alternations are first and foremost permutation alternations (see Section 2.3): by switching between variants, language users can change the order of possessor and possessum (genitive alternation), of recipient and theme (dative alternation), or of direct object and particle (particle placement alternation). Hence. the alternations studied here are not only grammatical but more specifically syntactic. That said, note that the genitive alternation can additionally be classified as a substitution alternation, in that language users can replace the genitive clitic -s by the preposition of. The dative alternation, on the other hand, can also be considered an insertion/deletion alternations, as the prepositional dative variant allows language users to retain the preposition to, while they can omit it if they use the ditransitive dative variant. These alternations were chosen partly because the constraints governing each of them are well-known, and have been shown to exhibit regional variation in small-scale studies (see Bresnan and Hay, 2008; Wolk et al., 2013; Haddican and Johnson, 2012). However, the scope of variation on a larger scale is still poorly understood. Additionally, many constraints (animacy, phonology, semantic class, end-weight, information status, persistence) are shared across two or even all three alternations, which facilitates comparative analysis across alternations. The variation we find among these alternations is particularly illustrative of the often subtle, experienced-based nature of grammatical

2.5 Grammatical Alternations Subject to Study in This Book

23

knowledge, given that – to the best of our knowledge – none of these alternations have been shown to vary substantially along other social dimensions (e.g. age, gender, ethnicity, class), nor are they discussed as examples of regional indicators. At the outset we would like to stress that per alternation, we use identical variable context definitions across the varieties of English under study. The rationale is twofold. First, we do not have reason to believe, and have not come across data suggesting that the contexts in which the syntactic variants are interchangeable differ substantially across the varieties we study. Second, the assumption of identical variable contexts is a prerequisite for comparative variationist analysis. In the next sections we summarize the literature on each alternation. 2.5.1

The Genitive Alternation

The literature on the genitive alternation is vast; for a detailed literature review, we refer the reader to Rosenbach (2014) and Heller (2018). Variation between the s-genitive and the of -genitive is old. While the s-genitive goes back to the “properly” Germanic way of expressing genitive relations with inflectional means, the of -genitive is an innovation that appeared during the ninth century. Thomas (1931, 284) shows that frequency-wise, the inflected genitive greatly outnumbered the periphrasis with of up until the twelfth century. Then, starting in Middle English, we begin to see “a strong tendency to replace the inflectional genitive by periphrastic constructions, above all by periphrasis with the preposition of ” (Mustanoja, 1960, I:70). In texts from the Early Modern English period, however, we observe a revival of the s-genitive, “against all odds” (Rosenbach, 2002, 184). The s-genitive has continued to be on the rise during the Late Modern English period (see e.g. Wolk et al., 2013; Szmrecsanyi et al., 2014; Hinrichs and Szmrecsanyi, 2007). In Present-Day English s-genitive and of -genitive constructions at large can encode a “a grab-bag” (Givón, 1993, 264) of relationships between what is conventionally called the “possessor” NP (in (7), the president) and the “possessum” (in (7), brother). The point is that these labels do not necessarily mean that we are talking about possession: for the s-genitive alone, Quirk et al. (1985, 321–322) list eight different meanings: possessive, but also subjective, objective, the genitive of origin, descriptive, the genitive of measure, the genitive of attribute, and the partitive genitive. Crucially, not all of these can also be encoded by the of -genitive to the same extent, and vice versa. It is fair to say that there is no overwhelming consensus in the literature as to the importance of semantic distinctions: some analysts have claimed that possessive relations have a privileged status in the semantics of the s-genitive (Taylor, 1989), though others have argued that the two genitives convey generally the same meaning (e.g. Altenberg, 1982, 11), or – conversely – that the two genitive constructions are semantically empty (e.g. Hudson, 1984).

24

Grammatical and Syntactic Variation

This difficulty in defining the meaning of genitive constructions (cf. Strang, 1968, 109: “any attempt to sum up ‘the meaning’ of the s-genitive is doomed”) is why of the three alternations studied here, the genitive alternation is arguably the most tricky one when it comes to defining those contexts in which both variants are interchangeable: variationist analysis investigates variation between formally different but semantically/functionally equivalent constructions, but when the semantics of one or all variants is hard to pin down then the task becomes difficult. The way in which the recent variationist literature on the genitive alternations handles this difficulty is by distinguishing between choice contexts (i.e. contexts in which both variants can be used) and “knockout contexts” (see Rosenbach, 2014, Section 2.2.3 for detailed discussion), in which only one of the variants can be used. The trick here is to not define choice contexts directly, but to eliminate knock-out contexts until what is left is choice contexts only (see Chapter 4 for details). In choice contexts, we see a range of known probabilistic constraints that regulate variation between genitive variants. One the one hand, we see in the genitive alternation factors that regulate permutation alternations in general: constituent animacy, constituent definiteness, information status (old before new), and constituent length/weight. On the other hand, the alternation is governed by constraints that are specific to genitive variation, chiefly among them the semantic relation between possessor and possessum, and the phonological shape of the possessor (final sibilancy). We will discuss the constraints considered in this book on page 26. To conclude this section, we would like to mention in passing another complication: the vast majority of the literature restricts attention to binary variation between the s-genitive and the of -genitive. However, there is a third variant – the NN-genitive – which is sometimes interchangeable with either the s-genitive or the of -genitive or with both (as in the FBI director, which paraphrases both the FBI’s director and the director of the FBI – see Szmrecsanyi et al., 2016b for a detailed analysis). In our analysis, we take the liberty to set aside the NN-genitive. Previous Variationist Research on the Genitive Alternation In this section we review recent variationist research on the genitive alternation. We restrict attention to studies that employ modern multivariate analysis techniques. Szmrecsanyi (2006, Chapter 5) (see also Szmrecsanyi and Hinrichs, 2008; Szmrecsanyi et al., 2017 for reanalyses of the dataset) is to the best of our knowledge the first multivariate study of the genitive alternation in English; investigating corpora sampling spoken British English and American English, the study uses regression analysis to show that previous usage of an s-genitive primes future usage of an s-genitive (α-persistence), and that usage of the preposition of in non-genitive contexts primes usage of the of -genitive

2.5 Grammatical Alternations Subject to Study in This Book

25

(β-persistence). Hinrichs and Szmrecsanyi (2007) (see also Szmrecsanyi and Hinrichs, 2008; Szmrecsanyi, 2010 for reanalyses) investigate genitive variation in the Brown family of corpora (written-edited-published British English and American English, 1960s and 1990s) using regression analysis; their analysis shows that the spread of the s-genitive in written English is primarily due to economy, given that the s-genitive is the more economical coding option. Hundt and Szmrecsanyi (2012) conduct a study of the genitive alternation in nineteenth- and early twentieth-century written British and New Zealand English (based on the ARCHER and CENZE corpora) and demonstrate via binary logistic regression analysis that possessor animacy is a more powerful constraint in the New Zealand English material compared to the British English material. Wolk et al. (2013) (see also Szmrecsanyi, 2013b; Szmrecsanyi et al., 2014; Ehret et al., 2014 for follow-up research) study both the dative and the genitive alternation in ARCHER, a corpus that covers materials from the Late Modern English period; the regression analysis in Wolk et al. (2013) shows that both the possessor animacy constraint and the possessum length constraints are diachronically unstable. The regression analysis in Grafmiller (2014), based on materials from the Switchboard Corpus and the original Brown Corpus, reveals substantial variation across six spoken and written genres concerning the effect sizes of language-internal constraints governing the choice between the s-genitive and the of -genitive in American English. Jankowski and Tagliamonte (2014) (based on Jankowski, 2013; see also Szmrecsanyi et al., 2017 for a re-analysis of the dataset) undertake a sociolinguistic analysis of genitive variation based on sociolinguistic interview materials from the Toronto English Archive (TEA) and the Southeastern Ontario Archive; Varbrul analysis indicates, among other things. that the animacy constraint is near-categorical in the materials, and that nonhuman collective/organization possessors increasingly attract the s-genitive in apparent time. Based on a dataset drawn from the Switchboard Corpus of US American English (used in Grafmiller et al., 2016 and reanalyzed in Szmrecsanyi et al., 2017), Shih et al. (2015) utilize regression analysis to show that genitive choice is not only constrained by animacy, weight, and other wellknown language internal factors but also by eurhythmicity: all other things being equal, language users use that genitive variant which maximizes the extent to which stressed and unstressed syllables alternate in an utterance. Heller et al. (2017a) apply Multifactorial Prediction and Deviation Analysis Using Regression/Random Forests (MuPDARF) to a dataset of genitive variation in materials from six components of the International Corpus of English (Great Britain, Hong Kong, India, the Philippines, Singapore and Sri Lanka); analysis shows that Indian English has epicenter status. Hackert and Wengler (2022) conduct a diachronic analysis of the genitive alternation in five varieties of English based on newspaper materials from the Bahamas, Jamaica, India, Great Britain, and the US; conditional random forest and

26

Grammatical and Syntactic Variation

MuPDARF analysis shows that in particular Caribbean varieties are partaking in American-led global trends in grammar toward, for example, densification (via usage of the more compact s-genitive), without actually approximating American norms, while overall there is a widening gap between metropolitan and postcolonial Englishes. As to learner English, Dubois et al. (2022) regress genitive variants in the Trinity Lancaster Corpus and demonstrate that while native speakers and learners have remarkably similar probabilistic grammars, low-proficiency learners are less sensitive to possessor definiteness and possessor animacy than high-proficiency learners. There is also substantial work on the genitive alternation in varieties of English by Benedikt Heller and colleagues (Heller et al., 2017b; Heller, 2018; Heller and Szmrecsanyi, 2019; see also Szmrecsanyi et al., 2016a; Tamaredo et al., 2019), which we summarize below and throughout this book.

Probabilistic Constraints on the Genitive Alternation The genitive alternation is known to be sensitive to a number of probabilistic factors, the most important of which include the following: • Animacy of the possessor is one of the most important constraints on genitive choice (see e.g. Rosenbach, 2005; Szmrecsanyi et al., 2017): animate possessors favor the s-genitive, while inanimate possessors disfavor. • According the principle of end-weight (Behaghel, 1909; Wasow and Arnold, 2003), language users in VO-languages such as English have a preference for placing heavier constituents after less heavy constituents. This is why according to the literature decreasing possessor length favors the sgenitive (as the s-genitive places the possessor before the possessum) (see Rosenbach, 2014, table 2); the reverse pattern is true for the of -genitive. • The literature suggests that possessors high on the definiteness scale attract the s-genitive, all other things being equal (see e.g. Rosenbach, 2014, table 2). • Thanks to a haplology effect, the presence of a final sibilant in the possessor, as in President Bush’s speech, is known to disfavor usage of the s-genitive (see e.g. Zwicky, 1987). • Genitive choice is subject to priming effects (Szmrecsanyi, 2006; Hinrichs and Szmrecsanyi, 2007). • Prototypical semantic relations have been shown to favor the s-genitive (see e.g. Rosenbach, 2014, table 2). Prototypical relations include body part relations (Tom’s hand), kinship relations (Tom’s sister), legal ownership (Tom’s house), and part-whole relations (the book’s cover). More information on how we operationalized these factors is provided in Chapter 4.

2.5 Grammatical Alternations Subject to Study in This Book

2.5.2

27

The Dative Alternation

The dative alternation, like the genitive alternation, is an extremely wellresearched syntactic variable; an extensive literature review is provided in Röthlisberger (2018a, Chapter 2). Prepositional dative constructions are attested as early as in Old English, though in that period the construction was subject to lexical restrictions (Allen, 2006, 206; De Cuypere, 2015). Additionally, constituent order in Old English ditransitive constructions was variable: the recipient could precede the theme, as in Modern Standard English, or it could follow the theme (van Kemenade, 1987; Koopman, 1990). More robust variability between the ditransitive dative and the prepositional dative dates back to the Middle English period (McFadden, 2002, 112), when the prepositional dative expanded into “a fully productive alternative” (Fischer and van der Wurff, 2006, 166) to the ditransitive dative. The emergence of the prepositional dative as a more analytic/periphrastic construction is sometimes speculated to have been triggered by the loss of case distinctions (see e.g. McFadden, 2002; Fischer and van der Wurff, 2006); an alternative explanation is contact with French (Visser, 1963). After Middle English, constituent order of nominal (but not pronominal) constituents – see Gerwin (2013) – in the ditransitive dative construction became fixed along the lines of the Modern English pattern by the late-fourteenth century (Allen, 2006, 206). In this book, we restrict attention to prepositional dative constructions in which the theme precedes the recipient, and ditransitive dative constructions in which the recipient precedes the theme. The dative alternation has received attention in many subfields in linguistics, including e.g. by researchers interested in first/second language acquisition (e.g. Campbell and Tomasello, 2001), information status (e.g. Thompson, 1990), psycholinguistics (e.g. Bock, 1986), and collustructional preferences (e.g. Mukherjee and Gries, 2009). On a more general level, following Gerwin (2014, 19) the sizable body of literature on the dative alternation in English can be divided into two types: studies that adopt a single-meaning approach (popular in usage-based/variationist circles), as opposed to studies that adopt a multiple-meaning approach (which is dominant in the generative community, but also in Construction Grammar). The multiple-meaning approach is summarized by Bresnan et al. as follows: Advanced by Green (1974) and Oehrle (1976) [references omitted], this idea [the multiple meaning approach] was taken up in influential work on language learnability by Pinker and colleagues (Gropen et al., 1989; Pinker, 1989). They argued that there are two ways of viewing the same giving event: as causing a change of state (possession) or as causing a change of place (movement to a goal). They hypothesized that the different ways of conceptualizing the giving event are associated with different structures, the possession meaning with the double object structure and the movement meaning with the prepositional dative structure . . . (Bresnan et al., 2007, 71)

Subsequently Bresnan et al. present an extended empirical argument why the multiple-meaning approach is inadequate. Their summary is as follows:

28

Grammatical and Syntactic Variation

First, linguistic intuitions of ungrammaticality are a poor guide to the space of grammatical possibility. Second, usage data reveals generalizations which we are sometimes blind to. Third, English dative verbs have more syntactic flexibility than we thought, occurring more freely in alternative constructions. And fourth, we cannot predict the dative alternation from meaning alone. (Bresnan et al., 2007, 75)

The alternative view is the single-meaning approach, whereby the ditransitive and prepositional dative constructions have essentially the same meaning, and variant choice is a function of language-internal (contextual) constraints such as constituent weight/length, constituent pronominality, information status, and so on. It is this variationist perspective that informs our analysis. Previous Variationist Research on the Dative Alternation To our knowledge, the first study to use regression analysis to model the dative alternation is Williams (1994). The paper investigates a corpus of American English covering four spoken genres and eight written genres. The regression analysis tests a range of predictors: syntactic class of the verb, register, modality, givenness, prosodic length, definiteness, animacy, and specificity. Of these, prosodic length, syntactic class of verb, and register are identified as most important. Gries (2005) explores syntactic priming effects in the dative alternation (and also the particle placement alternation, based on the ICE-GB corpus). Regression analysis shows that there are significant priming effects comparable to those reported in the experimentalist literature on the dative alternation. That said, the strength of priming effects differs across dative verbs. Bresnan et al. (2007) present a regression model of the dative alternation in spoken American English sampled from the Switchboard Corpus. They demonstrate that the alternation is predictable from a series of language-internal probabilistic constraints: givenness of recipient/theme, pronominality of recipient/theme, definiteness of recipient/theme, animacy of recipient, person of recipient, number of recipient/theme, concreteness of theme, and syntactic parallelism (see Bresnan, 2007 and Bresnan and Ford, 2010 for follow-up investigations of the Bresnan et al., 2007 dataset adding an experimentalist twist via a rating task, and Szmrecsanyi et al., 2017 for a comparative re-analysis of the dataset). Bresnan and Hay (2008) compare US American and New Zealand English datives (see Section 2.4). Theijssen et al. (2013) is a methodological paper that investigates a dative alternation dataset extracted from the British National Corpus (BNC) using Bayesian Networks (Pearl et al., 1988), memory-based learning (Daelemans and van den Bosch, 2005), and logistic regression analysis. Theijssen et al. (2013) conclude from their multi-method approach that “most of the alternation is determined by the verb and the length of the two objects” (Theijssen et al., 2013, 227). De Cuypere and Verbeke (2013) analyze the dative alternation in a sample extracted from the Kolhapur corpus of Indian English. Regression analysis shows that dative verb, recipient pronominality and constituent length are significant predictors of dative choice in their

2.5 Grammatical Alternations Subject to Study in This Book

29

dataset. Using data from ARCHER (a corpus of materials from the Late Modern English period), Wolk et al. (2013) conduct a regression analysis of the dative alternation which indicates that the disfavoring effect of inanimate recipients towards the ditransitive dative variants weakened during the Late Modern English period. Results indicate that theme length has a stronger effect on variant choice in American English. Schilk et al. (2013) investigate variation between complementation patterns (including the ditransitive and prepositional dative variants) after the verb give in South Asian Englishes based on web-derived corpora; multinomial regression indicates that “Pakistani English is closer to British English with regard to relevant driving factors than Indian English” (Schilk et al., 2013, 187). Similarly, Bernaisch et al. (2014) explore the dative alternation in six East Asian varieties (Bangladeshi English, Indian English, Maldivian English, Nepali English, Pakistani English and Sri Lankan English) plus British English as the reference variety, based on materials from the South Asian Varieties of English (SAVE) Corpus and the BNC; conditional inference trees and conditional random forest modeling uncovers “variety-independent protostructions, that is, abstract combinations of crossvarietally stable features with high predictive power for a particular syntactic pattern” (Bernaisch et al., 2014, 7). A comparative sociolinguistics analysis of the dative analysis is presented in Tagliamonte (2014), based on vernacular spoken language data (sociolinguistic interviews) collected in the United Kingdom and in Canada (see Szmrecsanyi et al., 2017 for a reanalysis of the dataset); regression modeling and conditional inference tree analysis fails to uncover any major differences as to how the dative alternation works in the Canadian and UK materials. Engel et al. (2022) investigate the extent to which the effect of probabilistic constraints on the dative alternation in British English differs as a function of register. Analysis shows that register modulates the probabilistic effects of definiteness of the dative constituents. What will take center stage in this book is work on the dative alternation by Melanie Röthlisberger and colleagues (Röthlisberger et al., 2017; Röthlisberger, 2018a; see also Szmrecsanyi et al., 2016a; Tamaredo et al., 2019), which we summarize below and throughout this book. Probabilistic Constraints on the Dative Alternation The following factors are the “usual suspects” in the literature on the dative alternation: • The principle of end-weight (Behaghel, 1909; Wasow and Arnold, 2003) predicts that heavier themes should follow less heavy recipients, while heavier recipients should follow less heavy themes. Such end-weight effects are very well documented for the English dative alternation (see e.g. Bresnan et al., 2007; Szmrecsanyi et al., 2017).

30

Grammatical and Syntactic Variation

• According to the literature, pronominal recipients strongly favor the ditransitive dative variant (see e.g. Bresnan et al., 2007; Szmrecsanyi et al., 2017). • Bresnan and Ford (2010) show that definite dative constituents tend to be placed first. • Information status is a well-known determinant of constituent order variation, including the dative alternation (see e.g. Bresnan et al., 2007): language users tend to place discourse-given constituents before discoursenew constituents if they have a choice (see e.g. Arnold et al., 2000; Collins, 1995). • More animate recipients/themes tend to be placed before less animate constituents (again, see e.g. Bresnan et al., 2007; Szmrecsanyi et al., 2017). More information on how we operationalized the above factors is provided in Chapter 4.

2.5.3

The Particle Placement Alternation

After many transitive particle verbs in English, language users have the choice between placing the particle after the direct object (as in look the word up – the “split” variant) or before the direct object (as in look up the word – the “continuous” variant). This syntactic variable we refer to as the particle placement alternation. While there is a consensus that the two variants are semantically equivalent, previous qualitative and nonvariationist quantitative scholarship on the particle placement alternation has sought to uncover the pragmatic and discourse-functional differences between the two constituent order patterns (see e.g. Fraser, 1965; Bolinger, 1971; Fraser, 1976; Gries, 1999; Dehé, 2002; Lohse et al., 2004; Cappelle, 2005, 2009; Thim, 2012). This literature, which inspires the constraints that we model, agrees that particle placement is a probabilistic phenomenon, with no single factor categorically determining the choice that language users have. Given that we also cover a number of indigenized L2 varieties, we should also add that phrasal verbs are known to pose a challenge to learners of English, especially those whose L1 does not have phrasal verb constructions (see e.g. Liao and Fukuya, 2004; Siyanova and Schmitt, 2007; Alejo-González, 2010; see also Gries, 2011 for a discussion of L1 acquisition of particle verbs). The particle placement alternation in its modern form with postverbal variants is attested from the Middle English period onwards – in Old English, the particle could also be placed before the verb (Claridge, 2000, 85), as in modern Germanic languages such as Dutch or German. In the Middle English period, the preverbal variants disappeared (Hiltunen, 1983, 106–111; Thim, 2012, 103), leaving behind the variation between the two postverbal variants that we still see in Modern English. The conditioning of this variation in Middle

2.5 Grammatical Alternations Subject to Study in This Book

31

English was also similar to that in Modern English, in that the continuous variant was preferred with NP objects and the split variant with pronominal objects (Elenbaas, 2013). As to variant rates, analysts report a predominance of the continuous variant in Early, and Late-Modern English texts (see Elenbaas, 2013, 495; Rodríguez-Puente, 2016, 150), although substantial register differences exist.

Previous Variationist Research on the Particle Placement Alternation Gries (2003) (see also Gries, 2002 for a partial summary) is the first modern variationist analysis of particle placement. Gries investigates the British National Corpus to study the effect of a range of constraints discussed in the literature via discriminant analysis, a multifactorial technique. Results indicate that particle placement is primarily a function of discourse-functional factors. Gries (2005) explores syntactic priming effects on the particle placement alternation, beside the dative alternation (see Section 2.5.2), based on the ICE-GB corpus). Regression analysis uncovers significant priming effects comparable to those reported in the experimentalist literature while at the same time particle verbs differ in their amenability to priming effects. Priming effects (a.k.a. persistence effects) were also tested in Szmrecsanyi (2006 chapter 7) (see Szmrecsanyi, 2005 for a partial summary): here the particle placement study is based on materials from the Corpus of Spoken American English (CSAE) and the Freiburg Corpus of English dialects (FRED). A regression analysis indicates that usage of the split variant robustly decreases the odds that the continuous variant will be used at the next possible opportunity (αpersistence); at the same time, the more recently a generic non-separated verb-particle/preposition pattern (as in I look at the house) was used, the more likely the continuous variant becomes (β-persistence). Haddican and Johnson (2012) report a judgment experiment and a Twitter corpus study showing regional differences such that the split variant is more favored in British and Irish varieties of English than in North-American varieties. Paquot et al. (2019) analyze materials from the Louvain International Database of Spoken English Interlanguage to explore factors influencing particle placement choices for English as a Foreign Language (EFL) learners, compared to native speakers (as sampled in the Louvain Corpus of Native English Conversation). Conditional inference trees and conditional random forest analysis suggest that the probabilistic grammars of learners with Germanic L1s is similar to that of native speakers, while learners with non-Germanic L1s have significantly more simple grammars. Finally, Röthlisberger and Tagliamonte (2020) investigate particle placement in a corpus covering vernacular speech of six communities in Ontario. Phrasal verb tokens were annotated for two language-internal constraints and a number of language-external constraints. Regression analysis shows, among other things, that younger speakers use the continuous variant

32

Grammatical and Syntactic Variation

more than older speakers, which is interpreted as evidence of a change in progress. The analysis in this book is based on Grafmiller and Szmrecsanyi (2018) (see also Szmrecsanyi et al., 2016a; Tamaredo et al., 2019 for spin-off studies). Probabilistic Constraints on the Particle Placement Alternation The particle placement alternation has been reported to be probabilistically constrained by a number of factors, the most important of which include the following: • Consistent with the principle of end-weight (Behaghel, 1909; Wasow and Arnold, 2003), longer direct objects are reported to favor the continuous variant (Kennedy, 1920, 30; Quirk et al., 1985, 1154; Biber et al., 1999, 932–933). • According to the literature, given language users’ preference for placing old information before new information, if the direct object is discourse-given, the split variant is preferred. If it is discourse-new, there is a preference for the continuous variant (see e.g. Kruisinga and Erades, 1953; Chen, 1986). • The literature suggests that concrete (and hence more accessible) direct objects favor the split variant (see e.g. Gries, 2003; Haddican and Johnson, 2012). • We know that the split variant is preferred if the particle verb construction is followed by a directional prepositional phrase, as in send cattle off to the mainland (see e.g. Fraser, 1976; Gries, 2003). • According to e.g. Gries (1999), Biber et al. (1999, 933), and Quirk et al. (1985, 1155), particle verb constructions with an idiomatic meaning (as in carry out duties) prefer the continuous variant, while constructions where the particle has literal/spatial meaning (as in carry garbage out) favor the split variant. More information on how we operationalized the above factors is provided in Chapter 4. 2.6

Summary

This book is concerned with “[v]ariability in the linguistic signal within a given language” (van Hout and Muysken, 2016, 250), also known as the availability of “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188) in – crucially – the realm of grammar (that is, morphology and syntax). This chapter began with a discussion of how the existence of this type of grammatical variation has been and still is a controversial idea in various linguistic communities of practice. We pointed out that for us, this question is (and should be) an empirical

2.6 Summary

33

one, and the evidence shows that grammatical alternations are in fact plentiful in standard and nonstandard Englishes around the world. Against this backdrop, we reviewed the literature on the three grammatical alternations subject to study in this book: the genitive alternation, the dative alternation, and the particle placement alternation.

3

World Englishes and Dialect Typology

In this chapter we review some of the World Englishes and dialect typology literature, followed by brief introductions to each of the nine regional varieties of English under study: British English, Canadian English, Irish English, New Zealand English, Hong Kong English, Indian English, Jamaican English, Philippines English, and Singapore English. For the present study, we situate these varieties within the context of two of the most prominent and influential models of World Englishes, namely Kachru’s (1985; 1992) Three Circles model, and Schneider’s (2003; 2007) Dynamic Model. The varieties we examine are a fairly representative sample covering both native (or “Inner Circle”) varieties and nonnative (or “Outer Circle”) varieties (see Kachru, 1992), and all are argued to have progressed to the latter stages of the Dynamic Model. 3.1

Theoretical Models of World Englishes

Writing now in the third decade of the twenty-first century, we are faced with a rapidly proliferating body of work examining the ever-changing social, political, and linguistic dynamics of English around the globe. It is a true embarrassment of riches for English language scholars (see e.g. Seoane, 2016; Schreier et al., 2019), and one that we could not hope to cover exhaustively in one short chapter. In this section we will therefore seek to provide a concise overview of the current theoretical landscape in World Englishes research. For a more comprehensive breakdown of the varying trajectories in World Englishes theorizing over the past fifty years, see Buschfeld and Kautzsch (2019). For descriptive surveys of individual varieties, we point the reader to the works contained in various handbooks and collections such as Kortmann et al. (2004), Kirkpatrick (2010), Kortmann and Lunkenheimer (2012), and Nelson et al. (2019). Categorization has been a common theme in World Englishes research, and it thus seems appropriate and useful here to apply a similar mindset to summarizing the present state of the field. The major theoretical models can be sorted into broad approaches based on several key dimensions. These dimensions include, but are not necessarily limited to, a focus on variety typology and classification, an emphasis on identifying those aspects that distinguish or 34

3.1 Theoretical Models of World Englishes

35

unify different (perhaps all) varieties, and a concern with the norm orientations among (speakers of) respective varieties (Hundt, 2020, 2). Models also may differ with respect to the degree to which they take a primarily synchronic versus diachronic approach, with those in the latter camp often exploring various aspects of political history, dialect contact, acquisition, and sociolinguistic identity in considerable detail (e.g. Mufwene, 2001; Trudgill, 2004; Schneider, 2007; Buschfeld and Kautzsch, 2020). However, while theories may differ in their key foci, it is also true that all theoretical models are by necessity multifaceted, incorporating factors along multiple linguistic and extra-linguistic dimensions. Early models of World Englishes tended to focus primarily on classification of different variety types, with one of the most elemental distinctions being how speakers of a given region acquire English (Buschfeld and Kautzsch, 2019, 53). Varieties can be classified according to whether members of a country learn English as a native language (ENL), as a second language (ESL), or as a foreign language (EFL) (McArthur, 1998). While structural distinctions between ENL varieties and (indigenized) ESL varieties have considerable empirical support (see Section 3.2), these categories tend to be more descriptive than explanatory, and elide important differences and similarities among varieties. Subsequent models have tried to expand on this distinction to more adequately incorporate the complex sociolinguistic realities of the various communities they are meant to encapsulate (e.g. Kachru, 1985; McArthur, 1998; Mesthrie and Bhatt, 2008). We focus on two of the most significant such models in Sections 3.1.1 and 3.1.2. 3.1.1

The Three Circles Model

Of World English models, perhaps none has been more influential than Kachru’s (1985) Three Circles model, which organizes varieties of English into three broad types represented by concentric circles: the “Inner Circle,” the “Outer Circle,” and the “Expanding Circle” (Figure 3.1). The circles are designed to reflect key differences in the historical diffusion of English, its role in social, political, and educational settings across different countries, and its status as a first or second language among speakers in a given region. The Inner Circle encompasses those regions for which English is a first language for the vast majority of speakers, for example, the United Kingdom, United States, Canada, Australia, and New Zealand. These countries represent the first major diaspora of English outside the British Isles, and in some cases may themselves serve as the colonial input source for new varieties in the Outer Circle, such as American English in the Philippines. The Outer Circle encompasses postcolonial regions where English was brought by (mainly British) colonizers in what would become major trade or exploitation colonies (Mufwene, 2001), and was not learned as a first language by the majority of the local population.

36

World Englishes and Dialect Typology

Inner Circle

Outer Circle

Expanding Circle

Inner Circle English is used as a first, and often the only, language by most speakers, e.g. UK, USA, Canada, Australia, New Zealand. Outer Circle English is used nationally as a second language (ESL) in politics, education, and business, e.g. India, Singapore, Hong Kong, Nigeria, Kenya, Philippines.

Expanding Circle English is used as a foreign language (EFL) for communication in many domains, but primarily with speakers of other languages outside the country, e.g. most countries in Europe, Asia, South America, and Africa that were not former British colonies.

Figure 3.1 The Three Circles model (inspired by Kachru, 1985)

English often still functions as a second language for many in the Outer Circle, though it is used widely across the population, is often recognized as an official language, and commonly serves as one of the main languages for administration and education (Kachru, 1985, 13). Many Outer Circle varieties can thus be conceived of as “functionally native” in terms of the language’s “functional domains and range, and its depth in social penetration and resultant acculturation” (Kachru, 1997, 68). Outer Circle countries and territories include India, Singapore, Hong Kong, Nigeria, Kenya, and Jamaica, among many others. The Expanding Circle comprises almost everywhere else that English is used as an international language, with the key historical difference being that Expanding Circle countries were never British (or American) colonies. English does not have official status in most of the Expanding Circle, though it may be used as a lingua franca in institutional or educational settings. A second crucial difference is that Expanding Circle varieties tend to be dependent on other varieties, typically from the Inner Circle, for providing usage norms (Kachru, 1985, 17). This is in contrast to the situation in the “norm-developing” Outer Circle, in which there typically exists a more complex push-and-pull between internal and external norm orientations. From an academic perspective, one of the most consequential shifts that emerged from Kachru’s model, and others like it (e.g. McArthur, 1992), is the emphasis on the pluricentrality of English in the contemporary world (Buschfeld and Kautzsch, 2019, 55). Nonetheless, Kachru’s model has been criticized on several fronts (e.g. Bruthiaux, 2003; Mesthrie and Bhatt, 2008), and it is widely acknowledged that variety typologies, such as the Three Circles or the ENL–ESL–EFL distinction, tend to bring “as much controversy as clarity” (Mesthrie and Bhatt, 2008, 6). A commonly cited issue with the Three

3.1 Theoretical Models of World Englishes

37

Circles model is its classification of varieties according to primarily geographical or political factors, which results in a focus more on countries than on speakers. This tends to obscure important internal cultural and sociolinguistic complexities of these varieties, while at the same time overlooking commonalities among territories with quite different colonial histories (Bruthiaux, 2003, 166–167).1 Additionally, variety-type models tend to treat the boundaries between categories, such as between ESL and EFL or between Outer and Inner Circles, as clear-cut and static, when there is ample evidence that they are anything but (Buschfeld, 2014; Deshors, 2014; Edwards and Laporte, 2015; Gilquin, 2015a). Indeed there are some varieties, such as South African English, which seem to resist classification altogether (Schneider, 2007, 173– 174). Finally, there is increasing awareness that the very concepts of nativeness and second/foreign language themselves are not only linguistically hard to define but politically and ideologically loaded (Brutt-Griffler and Samimy, 2001; Mesthrie and Bhatt, 2008, 36–37; Dewaele, 2018).

3.1.2

The Dynamic Model of Postcolonial Englishes

A particular concern with the variety-type approach was the emphasis on synchronic variation, and a lack of a sufficient diachronic perspective on the processes driving the development and emergence of new varieties of English. Thus a major theoretical refocusing came with Schneider’s (2003; 2007) Dynamic Model – itself heavily indebted to Mufwene’s (2001) “ecology” of language evolution model (Schneider, 2007, 21–25) – which heralded a “diachronic turn” (Buschfeld and Kautzsch, 2019, 59) in World Englishes modeling. In a nutshell, the Dynamic Model proposes that the development of new varieties of English can be “understood as a sequence of characteristic stages of identity rewritings and associated linguistic changes affecting the parties involved in a colonial-contact setting” (Schneider, 2007, 29). As with Kachru’s Three Circles model, the Dynamic Model has become one of the major inflection points in the conceptualization of World Englishes, and thus constitutes an important backdrop for contextualizing the varieties examined in our study. The Dynamic Model represents a large-scale attempt to consolidate various perspectives on language contact, language acquisition, and, most importantly, sociolinguistic identity construction into a unified theory of the development of new varieties across different postcolonial societies. At its core, the evolution of these new varieties is conceived of as “an identity-driven process of linguistic convergence” (Schneider, 2007, 30) between the members of a given 1 To be fair, the problem is perhaps more in the application of the model to specific circumstances

than in its overall conception. One could hardly accuse Kachru of being unaware of or indifferent to the sociolinguistic dynamics of English around the world.

38

World Englishes and Dialect Typology

society. Schneider places particular emphasis on the experiences of two major groups or “strands” (Schneider, 2007, 31), the settler (STL) and Indigenous (IDG) populations, and on the changing social, political, and ideological settings that shape, and are shaped by, their interactions. The model proposes that new varieties evolve along five distinct phases of development, beginning with the arrival of the English language to a new territory, typically as the first language of a small STL population. Gradually, new local linguistic norms emerge within the nascent variety as social and political forces give rise to both greater independence from the homeland, and more intimate intermingling of the STL and IDG strands. A sense of shared local identity grows among the population, the new variety stabilizes, and ultimately regional varieties take root. The five phases are summarized as follows: (i) Foundation: English-speaking settlers (STL) arrive in a new region, where English becomes the primary means of communication. In trade or exploitation colonies the STL group may be quite small, while in settlement colonies the founding groups tend to be larger, and grow rapidly with the influx of new migrants. Both the settler and Indigenous populations (IDG) see themselves as distinct from the other, and the STL group identifies strongly with the home country. Two main contexts of contact are operative at this stage: dialect contact within the STL population, and cross-linguistic contact between the STL and IDG peoples. Linguistically, the phase is marked by the gradual development of a “middle-of-the-road” variety (koine) within the STL population, while an incipient pidgin may develop as a lingua franca used for communication between the STL and IDG groups. (ii) Exonormative Stabilization: The new settlement becomes an official colony or territory, and English becomes the primary language of local administration and education. Members of the STL strand continue to see themselves as belonging to the home country, though the growing distance (both physical and psychological) between the new and the old worlds engenders a kind of hybrid “British-plus” identity (Schneider, 2007, 37) among many of the STL strand. Bi/multilingualism spreads among the IDG strand, as the practical and social value of knowing English becomes more apparent. Within the STL strand, the standards of the home variety (e.g. English as used in Britain) are still taken as the norms, and are maintained particularly among the elite. Structural features of the local IDG languages begin to exert their influence on the English of the IDG strand as more (often adults) learn English. Among the STL strand, spoken English in particular begins to gradually develop a local flavor of its own, though this is mainly reflected in the increasing borrowing of local words.

3.1 Theoretical Models of World Englishes

39

(iii) Nativization: This is the most dramatic stage of linguistic and social transformation (Schneider, 2007, 39). There is a large-scale reassessment of old identities among both the STL and IDG strands in light of the new local reality. Ties with the home country are weaker than before, and economic, political, and linguistic independence becomes a central concern within the STL strand. A stronger sense of shared identity also develops between the STL and IDG strands, as both come to see themselves as belonging to the same local territory. Contact between groups intensifies, both in the numbers of speakers interacting and in the frequency and regularity of their interactions. This results in a growing number of ESL speakers among the IDG population, and intensifying accommodation among L1 English speakers leads to the emergence of novel structural features in the STL strand. It is in this phase where we see some of the most dramatic restructuring of the local language and the birth of a formally distinct English variety (Schneider, 2007, 44). (iv) Endonormative Stabilization: This phase follows in the wake of political independence or a similar major event marking a tectonic shift in the cultural and political futures for the territory (Schneider, 2007, 48– 49). Local populations see themselves as members of a new nation, and cultural and ethnic divisions are typically downplayed for the sake of internal social harmony. Sociolinguistically, the new variety is widely accepted and embraced as a marker of local identity. New literary traditions begin to emerge, along with the development of local dictionaries and grammars. Formally, the new variety is characterized by its considerable homogeneity, resulting in part from earlier koineization processes as well as from an overt emphasis on construction of a singular national identity (Schneider, 2007, 48–49). (v) Differentiation: The stabilization of the new country’s independence opens up space for internal diversification. Identification with the new nation is a given, and members of the population begin to (re)orient themselves along regional, social, and ethnic lines. With these realignments comes the development of localized social networks and novel linguistic norms as language is recruited to establish and enforce these new local identities. This gives rise to ever greater geographical and social heterogeneity, and ultimately the birth of major dialects within the new country (Schneider, 2007, 54). The phases outlined above necessarily abstract from and generalize over many of the internal complexities of these varieties and communities (Schneider, 2007, 55). One of the advantages of the Dynamic Model is that it is, in a word, dynamic, in the sense that varieties are not fixed to any particular phase. Recent research tracking varieties along this model’s evolutionary pathway shows evidence of considerable ongoing movement, particularly in

40

World Englishes and Dialect Typology

the progression of varieties in Asia (Buschfeld et al., 2014; Schneider, 2014), though the strict linear progression implied by the model has been questioned for both Inner and Outer Circle varieties (see e.g. Mesthrie and Bhatt, 2008, 35; Onysko, 2016, 201). Others have challenged the appropriateness of the model as a whole, at least for certain varieties (e.g. Evans, 2014). Before moving on then, it is worth considering alternative and/or complementary approaches to the Dynamic Model. 3.1.3

Beyond the Dynamic Model

One point of contention with the Dynamic Model has been over the exact role of identity in the processes underlying new variety formation. Trudgill (2004, 2008), for instance, categorically denies the role of identity, and proposes a wholly functional and mechanical view of how language contact and accommodation drive language change in colonial settings. The driving force behind Trudgill’s model is the “seemingly universal . . . human tendency to ‘behavioral coordination’ ” (Trudgill, 2004, 27–28). Accommodation is merely an inevitable result of contact among speakers of different languages and dialects, and in Trudgill’s view social structure affects linguistic form “purely [as] a matter of who interacts most often with who” (Trudgill, 2004, 149; see also Labov, 2001, 503–506). The question of sociolinguistic identity in new dialect formation has prompted considerable discussion (see e.g. papers in Language in Society, vol. 32, issue 2), but there is general agreement about the importance of accommodation and the universal human tendency to talk like others do (Britain, 2018). All sides agree that while new varieties will differ in many details, because crucially the populations of speakers involved are different in each case, some broad similarities should also emerge due to common mechanisms of acquisition and accommodation. Mechanisms of selection – that is, mechanisms that favor one particular form being adopted by a community – are particularly important for new dialect formation in this view, and they drive processes such as borrowing and L1 transfer effects. Where Trudgill and Schneider seem to disagree is in the role of “identity” in this process. Trudgill categorically denies the role of identity at the level of the individual – speakers do not change their usage to symbolize some kind of local or national identity. Associations of identity emerge only after the new feature has developed. For Trudgill the most consequential period is during the earliest phases of (gradually intensifying) contact among speakers of maximally distinct languages/dialects. Schneider, on the other hand, sees the role of identity as operating more at the community/national level. He particularly focuses on its importance for later phases of development, in which attitudes towards the increasingly independent community begin to coalesce around a shared sense of nationhood and local identity distinct from the home country. These shifting attitudes and identities will change the kinds of interactions

3.1 Theoretical Models of World Englishes

41

members of the community may have with one another, and this is facilitated by the institutions (schools, governmental bodies, etc.) they set up and participate in. The “mechanical forces” do not change, but the patterns of interaction certainly do, and these changing patterns will have consequences for how the varieties continue to develop. Thus for Schneider, “identity-driven stabilization is strongest in phases 4 and 5” (Schneider, 2014, 15) as the population shifts from an exonormative to an endonormative sociolinguistic orientation. Another strand of criticism of the Dynamic Model is its limited focus on postcolonial varieties – that is, those of the Inner and Outer Circles – as recent work has raised important questions regarding its applicability to nonpostcolonial settings in the Expanding Circle (e.g. Onysko, 2016; Buschfeld and Kautzsch, 2017). Buschfeld and Kautzsch (2019, 61) point to numerous studies demonstrating the emergence of indigenized L2 varieties within countries that cannot be traced to a British colonial past, and Schneider himself questions the extent to which the Dynamic Model can capture the evolving dynamics of English as a global language (Schneider, 2014, 27–28). With this in mind, Buschfeld and Kautzsch (2017) take up the charge of adapting the model to more satisfactorily encompass the new social and political realities of English as a truly global language. They reconceptualize Schneider’s model in terms of a complex interplay between various intra- and extra-territorial forces that act on all varieties as they proceed through the various phases of evolution. These forces include familiar aspects of sociopolitical history (e.g. colonization), language attitudes, and identity construction, but also the changing dynamics of globalization, such as tourism, the Internet and social media, contemporary political relationships, and transnational migration. For more extensive coverage and discussion, see Buschfeld and Kautzsch (2020). Taking a different tack, Mair (2013) proposes a “World System of English” model which foregrounds the hierarchical relationship(s) between standard and nonstandard varieties across the globe. In Mair’s model, transnational influence among varieties cascades down a ladder of sociolinguistic dominance from the global “hyper-central” hub (i.e. standard American English) to increasingly less influential varieties at the periphery. Under this model, speakers of varieties at lower rungs are expected to be familiar with varieties above them, and the “hub” variety is expected to be a factor in the development of all others (Mair, 2013, 261). Like Buschfeld and Kautzsch’s model, Mair’s model is explicitly designed to address new mechanisms of communication and linguistic diffusion in the modern world (e.g. the Internet), however it remains an open question as to how best to operationalize suitable tests of the proposed sociolinguistic order (Hundt, 2020, 7–9). Lastly, there has been a recent revival of interest in Mufwene’s (2001) evolutionary model, which conceptually undergirds many of the sociolinguistic processes driving the Dynamic Model. For example, Hansen (2018), following Biewer (2015), expands upon Mufwene’s notion of competition and selection

42

World Englishes and Dialect Typology

among features within the “feature pool” to include not only pidgin and creole languages but ESL varieties as well (and, we presume, postcolonial ENL varieties; see Mufwene, 2001, 98–103). The focus here is not on the placement of varieties within a given phase or circle, but on the specific processes which guide the selection of features from within the pool (see also Biewer, 2015, 83–114). These include substrate transfer, L2 acquisition strategies, and general cognitive processes relating to perceptual saliency, markedness, or transparency. Cross-cutting these factors are sociocultural forces relating to the intensity of contact, language attitudes, and social accommodation (Hansen, 2018, 85–86). Specific processes may be more or less operative within a given community at a given time, leading to differences among varieties even at similar stages in their development cycles. At the same time, commonalities should be traceable to the extent that they reflect universal aspects of these processes, such as the tendency for high-contact languages to exhibit greater analycity and simplification (see Section 3.2). An appealing aspect of this approach is that it attempts to explain why specific varieties are structured the way they are (Hansen, 2018, 89). Unfortunately, arriving at such explanations requires a substantially more detailed investigation of the synchronic and diachronic sociolinguistic dynamics at work in each variety than we can provide with our data. In closing this section, we are reminded of a well-known aphorism in the statistics world, usually attributed to Box (1979), that “all models are wrong, but some are useful.” This is surely as true of scientific models as it is of statistical ones, and the utility of a given model will vary according to the needs of the user and the questions posed to it. With this in mind, we note that the present study was designed as an exploration of the scope of grammatical variation across diverse varieties, and not as a test of specific theoretical models of World Englishes. We view the proliferation of theoretical models as a sign of a healthy academic field, and although it seems unlikely that researchers will ever be able to build anything like a “grand unified theory” of World Englishes (Hundt, 2020, 15), we believe that different models can offer insight for understanding typological patterns we observe around the world. We turn to this topic in the next section. 3.2

A View from Dialect Typology

In the previous section we reviewed some of the dominant theoretical models of World Englishes. In this section we briefly discuss the state of the field from the perspective of dialect typology, which focuses on the empirical identification of systematic structural similarities and differences across varieties of English – that is to say, (socio)linguistic universals (Trudgill, 2009, 2011; Szmrecsanyi and Röthlisberger, 2019). We focus here on the two kinds of cross-varietal generalizations that are most relevant to our study: so-called angloversals, features

3.2 A View from Dialect Typology

43

that tend to recur in varieties of a specific language (English in this case), and varioversals, that is, features shared by language varieties with similar sociohistories, historical depth, and modes of acquisition (Szmrecsanyi and Kortmann, 2009, 33).2 Dialect typology is of particular interest to World Englishes research because most variety-centric approaches to World Englishes, which include Kachru’s and Schneider’s models, generally share the assumption that variety classifications should reflect important structural differences (Hundt, 2020, 3). Sets of varioversals should therefore be identifiable across World Englishes, and patterns among those varioversals should ideally correspond to theoretical variety taxonomies. Furthermore, it is expected that some degree of a “common core” (Quirk et al., 1985) should persist across World English varieties, given their common ancestry and often long-standing orientations towards certain norm-setting standard varieties (McArthur, 1998; Modiano, 1999; Mair, 2013). In other words, there should be some set of common grammatical features (that is, angloversals) that we can observe in all varieties (Szmrecsanyi and Röthlisberger, 2019, 543). Such a common core is often theorized to persist longest at the most abstract levels of grammatical structure, such as the probabilistic conditioning of morphosyntactic alternations (Tagliamonte and D’Arcy, 2009; Bernaisch et al., 2014), while evidence of localized indigenization effects is predicted to be most prevalent at more concrete levels of the grammar–lexis interface (Schneider, 2007, 46; Mukherjee and Gries, 2009; Schilk et al., 2012). The assumptions of variety-centric models have gained considerable empirical support in recent years through work on World Englishes dialectology (e.g. Kortmann and Szmrecsanyi, 2004; Szmrecsanyi and Kortmann, 2009; Kortmann and Wolk, 2012). In a recent survey of the dialect typology literature, Szmrecsanyi and Röthlisberger (2019) identify three extra-linguistic dimensions along which English dialects are commonly aligned: variety type, exposure to contact, and geographical proximity (areal patterns). To the extent that these dimensions overlap with (aspects of) theoretical models, we can view the observed typological patterns as confirming, or at least corroborating, evidence for different theoretical taxonomies. With respect to “variety type,” the most common classification used is a tripartite distinction roughly coextensive with the Inner and Outer Circles, but with the added inclusion of English-based pidgins and creoles 3.2.

2 A third kind of universal, vernacular universals – features common to spoken vernaculars – are

also widely studied in World Englishes research (Chambers, 2004; Trudgill, 2009). Because our study focuses on grammatical alternations observed across all registers (e.g. formal and informal language) and modalities (e.g. speech and writing), questions about the nature, prevalence, and putative origins of vernacular forms are less germane to our investigation (see Filppula et al., 2009 for further discussion).

44

(10)

World Englishes and Dialect Typology

a. Native L1 varieties (or ENL), for example, Canadian and New Zealand English, corresponding to Kachru’s Inner Circle b. Indigenized L2 varieties (ESL), for example, Indian, Singapore, and Jamaican English, corresponding to Kachru’s Outer Circle c. English-based pidgin and creole varieties, for example, Tok Pisin and Hawai’i Creole, which are harder to place within the Three Circles or ENL–ESL–EFL continuum.

Notably this classification may encompass more fine-grained regionally or socially defined dialectal varieties, such as, African American Vernacular English or dialects of the Southwest of England, which may or may not be subsumed within broader national varieties in theoretical models. It is also not obvious where Expanding Circle (EFL) varieties, for instance, English as used in Argentina, Belgium, China, or Dubai, fit into this system, but this question is not of central importance here, as we do not consider any such varieties in our study. Empirically, the search for varioversals often relies on data from descriptive surveys of lexical and grammatical features observed in Englishes around the world (see Chapter 2.3 for examples of alternations that have been studied). These features are then subjected to statistical analyses such as clustering and dimensional reduction techniques, which are aimed at detecting and summarizing covariation among the different features. Results can then be compared against different variety typologies. For example, Szmrecsanyi and Kortmann (2009) analyze the prevalence of seventy-six morphosyntactic features across forty-six varieties of English and find readily distinguishable groupings of L1 (ENL) varieties, indigenized L2 (ESL/EFL) varieties, and pidgin and creole varieties (Figure 3.2). Similar studies using different datasets and/or different features have revealed comparable patterns (e.g. Kortmann and Szmrecsanyi, 2009; Kortmann and Schröter, 2017). When in comes to patterning among surface-level morphosyntactic features, we thus have robust evidence that L1/ENL and L2/ESL variety types pattern alike, even when composed of widely dispersed varieties. On the more abstract end of linguistic structure, variationist studies of grammatical alternations have tended to reveal only relatively weak evidence of cross-varietal differences in conditioning constraints (Bernaisch et al., 2014; Heller et al., 2017a; Hundt et al., 2020, 2021; Röthlisberger et al., 2017, though cf. Szmrecsanyi et al., 2016b, 2017; Grafmiller and Szmrecsanyi, 2018). Thus the evidence for consistent differences in probabilistic indigenization – that is, the development of localized probabilistic variable grammars (see Chapter 5) – between these variety types is much less clear-cut. These findings would seem to lend support to the persistence of a common core of variable grammars shared across English varieties. Thus, for example, we might consider the strong preference for use of the s-genitive variant with animate

3.2 A View from Dialect Typology

45

Tob/TrnC BelC

JamC Gullah

NigP SolP

AusCs

CamP

HawC

Bislama

Urban AAVE AbE

TP

BahE Earlier AAVE

GhP Norfolk ButlE SgE

SurCs EAfE

GhE MalE

Variety type L1 L2 P/C

FijiE

AppE CollAusE East Anglia Southwest Southeast

CamE

BlSAfE

PakE

NfldE

ChcE

ScE InSAfE AusVE NZE WelE

OzE North CollAmE

IsSE

IrE

WhSAfE Orkney/Shetland

Figure 3.2 Multidimensional scaling visualization derived from co-occurance matrix containing 76 features × 46 varieties (adapted from Szmrecsanyi and Kortmann, 2009, 1651).

possessors or with prototypical genitive relations (kinship, ownership, bodypart) to be a genuine angloversal. We note however, that such studies tend to focus on one alternation at time, and thus a major desideratum of dialect typology is a method for simultaneously exploring patterns across multiple variables from a variationist perspective (see Szmrecsanyi et al., 2019). The second dimension, degree of language contact, is also of keen interest to typologists and World Englishes researchers. The categories in (3) also correlate with the degree of language and dialect contact present in their varieties’ formative stages. Trudgill (2001, 2004) places particular importance on the role of language and dialect contact in the development of new varieties, and proposes a broad typological distinction between “high” versus “low” contact varieties. The latter constitute long-established and relatively isolated mother tongue varieties, while the former cover all manner of varieties including all of the national standard varieties examined in this book. With respect to the third dimension, areal proximity, it seems to have relatively low explanatory power with respect to the distribution of morphosyntactic features among World Englishes (Szmrecsanyi and Röthlisberger, 2019, 538–540). Nonetheless, there has been a growing interest in the role

46

World Englishes and Dialect Typology

of regional “epicenters,” for example, Indian English in South Asia, as normproviding focal points for nearby less dominant varieties (see e.g. Mair, 2013) and targets of accommodation among speakers within the region (e.g. Hundt, 2013; Gries and Bernaisch, 2016; Heller et al., 2017a). When it comes to more recent theoretical approaches such as the Dynamic Model, empirical evaluation and testing has proven challenging, in no small part due to the complexity and scope of these models (Hundt, 2020). In terms of empirical predictions of the Dynamic Model, Gries et al. (2018) suggest a reasonable hypothesis, whereby: the structural distinctiveness of a postcolonial English will increase as it progresses through Schneider’s (2003, 2007) developmental cycle because sociohistorical and/or sociolinguistic advancement is assumed to be reflected in variety-specific/varietypreferential linguistic choices . . . In other words, the Dynamic Model rests on the assumption that more evolutionary progress means more structural difference from a historical input variety, which is British English in most cases. (Gries et al., 2018, 275)

As it turns out, a simple linear correlation between progression through the model’s phases and greater structural diversification appears to be insufficient, at least when it comes to variable grammars of morphosyntactic alternations (Hundt, 2020, 10–13). Results from Gries et al. (2018) comparing the genitive alternation in only two varieties, Singapore and British English, are rather subtle and complex, although they do suggest that constraints on genitive choices in Singapore English have diverged to a minor degree from British English. Similarly, studies comparing wider ranges of postcolonial ENL/Inner Circle varieties – all of which are (much) more advanced in the developmental cycle – with ESL/Outer Circle varieties have tended to find only minimal differences in probabilistic effects, which rarely sync up with the Dynamic Model phases. In sum, there is now a robust body of evidence attesting to widespread structural diversity across World Englishes, particularly at the level of surface lexico-grammatical features, and this diversity has been shown to correlate with variety typologies defined according to shared characteristics of social, historical, and linguistic settings. To what extent a similar picture emerges from comparison of varieties at the level of probabilistic indigenization of alternation specific constraints is the central question of this book.

3.3

Varieties in This Study

In this section we briefly introduce the nine varieties included in our study: British (BrE), Canadian (CanE), Hong Kong (HKE), Indian (IndE), Irish (IrE), Jamaican (JamE), New Zealand (NZE), Philippines (PhlE), and Singapore (SgE) English. These varieties were chosen primarily based on the availability of sufficient parallel corpus data when the project began, and we acknowledge that this set omits many important regional varieties, such as those of

3.3 Varieties in This Study

47

the United States and all of Africa. The aim of this book is not to provide a comprehensive picture of grammatical variation everywhere, but rather to examine the potential scope of stability and fluidity in probabilistic variable grammars across national varieties of (standard) English. We believe the range of varieties sampled here is sufficiently diverse to provide insight to this question, however a great deal more work on many other varieties is certainly needed. 3.3.1

British English (BrE)

In comparative studies of World Englishes British English (BrE) is often the natural reference point, as earlier forms of English(es) in Great Britain were the ultimate source from which all other varieties developed. With the exception of Philippines English, all the varieties considered in our study are the direct result of British colonization and (sometimes forced) migration of peoples from England as well as Asia and Africa. A full recounting of the history of English in the British Isles is of course beyond the scope of this book. As a locus for sociolinguistic and dialect research, Great Britain is unmatched among English speaking countries, with the United States being perhaps a close contender. Heller (2018) and Röthlisberger (2018a) provide concise reviews of key studies examining (respectively) the genitive and dative alternations in BrE, but until very recently there have been comparatively few studies exploring particle placement in BrE specifically (e.g. Gries, 2003), and we are only just beginning to examine the extent of internal regional variability. For example, Haddican et al. (2021), using judgment surveys and data from Twitter and other corpora, find little evidence of regional variation in probabilistic effects on particle placement within the United Kingdom, although they report a significant increase in the preference for and use of the split variant in the British Isles compared to Canada and the United States. The reasons for this discrepancy are not well understood, but the data from historical corpora suggest that the greater use of the split variant may in fact be an innovation particular to BrE, rather than part of a more general (global) trend towards colloquialization or Americanization (Haddican et al., 2021, 15). Their study highlights the danger in treating BrE as simply a static input source and not a living dialect in its own right (see also Gries et al., 2018, 201–202). With this in mind, we stress that the present-day BrE data we examine should not be viewed as representative of the historical source from which the other varieties developed (Röthlisberger, 2018a, 34). 3.3.2

Canadian English (CanE)

The foundation of what would become (mainland) Canadian English (CanE) was primarily laid in the late 1700s, as immigration from the British colonies

48

World Englishes and Dialect Typology

in New England was encouraged during and following the French and Indian War (1755–63) and American War of Independence (1775–83). English loyalists from America constituted the major founding population which established the early cultural and linguistic norms and “play[ed] a pivotal role in establishing the ‘bedrock’ of Canadian speech patterns” (Levey, 2010, 115). This period was followed by further waves of immigration from the British Isles, in some cases explicitly encouraged by the British Crown as a counter-balance to American political and cultural influence. The nineteenth and twentieth centuries saw increasing numbers of Scottish and Irish immigrants, and many more from other European countries. Except for a few specific enclave communities, for example, Newfoundland (Clarke, 2010), where many ancestral British features have been maintained, entrenchment of older American patterns has persisted into the twentieth century. However, the extent of this entrenchment – and its resistance to change – remains a matter of some debate (Levey, 2010, 115). Regardless of its features’ origins, CanE has often been held up as an exemplar of national grammatical homogeneity (Chambers, 2012), though it is apparent that standardizing and diversifying forces are at work in Canada today (Dollinger, 2019). Schneider (2007) places contemporary Canadian English squarely within the Differentiation stage (Phase 5), noting that “the sociolinguistic situation of present day Canada is marked by changes, realignments and diversification,” thus, “current changes of Canadian English are moving towards the birth of new dialects” (Schneider, 2007, 249). A growing body of work has documented a developing urban versus rural divide reflected in changes in progress involving a number of grammatical features (e.g. Franco and Tagliamonte, 2020; Jankowski and Tagliamonte, 2021), including the three alternations under study here (Jankowski and Tagliamonte, 2014; Tagliamonte, 2014; Röthlisberger and Tagliamonte, 2020). Grammatical diversification in Canadian English thus continues to be an active and exciting area of ongoing research. 3.3.3

Hong Kong English (HKE)

British control of Hong Kong began following the First Opium war in 1841– 42, although contact among traders from the British East India Company and local populations had started decades prior. Early contact with British traders and missionaries had resulted in the emergence of a so-called “Chinese Pidgin English,” but how much of an impact this pidgin English had on the development of modern Hong Kong English is uncertain (Schneider, 2007, 135; Setter et al., 2010, 104). English rapidly spread through schools as a second language, mainly among the middle and upper classes, and remained the only language of government function in Hong Kong until 1974, when Chinese was also recognized as an official language (Setter et al., 2010, 105). In the latter half of the twentieth century use of English became more democratic as English-medium

3.3 Varieties in This Study

49

secondary schools became more widespread (Schneider, 2007, 136). English has increasingly come to serve as a lingua franca for interethnic and international communication, and the instrumental role of English led to a growing dissociation between the English language and English or Western culture and identity (Setter et al., 2010, 106–107). Positive attitudes towards English and local identity construction furthered the emergence of a unique variety of Hong Kong English (Setter et al., 2010, 107). The importance of English has continued since the handover of control of Hong Kong to China in 1999, as more and more people use English for various functions in their daily lives. The sociolinguistic situation in Hong Kong at the beginning of the twenty-first century is thus one of “polyglossia with increasing bilingualism” (Li, 1999). Structurally, Hong Kong English has developed distinct linguistic features at all levels of linguistic structure, however there continues to be debate about the status of Hong Kong English as an independent variety (Setter et al., 2010, 112– 113). Schneider (2007, 136–137) places Hong Kong English within Phase 3 (nativization), noting the shift towards more localized identity construction and the rise of a local “complaint tradition.” The continued prevalence of exonormative orientations among many users towards British (and American) English standards suggests that HKE has not yet reached the phase of endonormative stabilization (see Setter et al., 2010, 112–116). 3.3.4

Indian English (IndE)

Sustained contact between British traders and missionaries and Indigenous peoples of the Indian subcontinent began in 1600, when a Royal Charter was granted to the East India Company (Schneider, 2007, 162). For the following 150 years, contact between locals and foreigners was fairly limited and the orientations of the colonizers and Indigenous groups remained strongly polarized (Mukherjee, 2010b, 168). It was only after control of India was ceded to the British Crown in the latter half of the eighteenth century that the British began to more aggressively consolidate power in the region. The push for greater control initiated a gradual process of Anglicization of India, as English was increasingly taught in schools and bilingualism spread particularly among the upper classes. Still, the sociolinguistic orientation among settlers and Indigenous English users at this stage was decidedly exonormative (Schneider, 2007; Mukherjee, 2010b). More extensive nativization (Dynamic Model Phase 3) began in the 1800s after colonial administration adopted a more extensive English medium school system designed to educate (i.e. “Westernize”) a growing class of civil servants from the local population, a policy that continued into the twentieth century. The spread of English accelerated following independence from Britain in 1947, and it continues to function as the language of various official domains in India, being used by circa 35– 50 million Indians daily (Mukherjee, 2010b, 167). Mukherjee (2007, 2010b)

50

World Englishes and Dialect Typology

places present-day Indian English at the stage of endonormative stabilization (Phase 4), albeit with echoes of continuing nativization (see also Schneider, 2007, 171–173). Contemporary Indian English (IndE) can thus be characterized as a semi-autonomous variety, reflecting a complex interplay of exonormative and endonormative orientations. In fact, one could argue that the centrifugal forces that move Indian English further away from native Englishes, on the one hand, and centripetal forces that keep the norms of Indian English close to native Englishes for the sake of international intelligibility, on the other, are in a state of equilibrium, determining a steady state of progressive forces of language change and conservative forces of (native) norm persistence . . . ,(Mukherjee, 2010b, 176–177; emphasis in original)

Linguistically, IndE gradually acquired a distinctive character during the later stages of nativization, as learners were/are exposed less to native (British) speakers of English and more to local Indian speakers and teachers (Schneider, 2007, 167). The distinctive morphosyntactic features of IndE have been widely studied (see e.g. Bhatt, 2004; Sailaja, 2009), but for our purposes, the most relevant work has been in the domain of (dative) verb complementation patterns. The main finding of this work is that the prepositional dative tends to be more frequent with certain verbs in IndE than with those same verbs in BrE (Olavarría de Ersson and Shaw, 2003; Mukherjee, 2005; Mukherjee and Gries, 2009; Mukherjee, 2010a; Schilk et al., 2013). As for the genitive and particle placement alternations, much less is known about the factors conditioning these variables in IndE (see Sections 2.5.1 and 2.5.3). 3.3.5

Irish English (IrE)

Of all the varieties examined here, Irish English has by far the longest, and arguably most complex, history of contact with its ancestral variety, dating back to the twelfth century when settlers from south-western England first arrived. Hickey (2010) divides the history of Irish English into two periods, with the first spanning from the earliest settlement to about 1600, during which the Anglo-Norman settlers were largely absorbed by the Irish. The seventeenth century saw renewed efforts by the English crown to Anglicize Ireland through a campaign plantation settlements largely reserved for migrants from Scotland and North and West Midlands of England. The large influx of English speaking settlers “tipped the linguistic balance” (Hickey, 2010, 77) in favor of English, especially in Ulster in the north. It is thought that a kind of functional bilingualism, largely via acquisition of English as adults, existed during the seventeenth and eighteenth centuries, but the increasing importance of English resulted in most of the population shifting completely from Irish to English by the end of the nineteenth century. Nonetheless, the long history of Irish-English bilingualism left its mark on the phonological and grammatical structure of Irish English, and a number of unique morphosyntactic features have been attributed

3.3 Varieties in This Study

51

to substrate influence from Irish (Filppula, 2004; Hickey, 2010). Despite its complex history, contemporary Irish English is usually grouped with the Inner Circle varieties, and we have sufficient evidence of regional diversification to place it squarely within Schneider’s Phase 5 (e.g. Kallen, 1997). Finally, it is worth noting that Irish immigrants themselves would come to settle in (or be deported to) other countries over the centuries, such as the United States, Canada, Australian, and New Zealand, and Irish English (at various stages of development) has itself contributed to the development of new Englishes in various postcolonial regions, for example, Newfoundland (Clarke, 2010). 3.3.6

Jamaican English (JamE)

The history of English in Jamaica begins in 1655 when the British Crown took control of the island from the Spanish (Deuber, 2014, 28). Sugar cultivation was introduced rapidly, and white settlers, primarily from other areas in the Caribbean arrived soon after, followed by large influx of non-white slaves and indentured servants from Africa and the Caribbean towards the end of the fifteenth century. Language contact in the early phases was surely ubiquitous, involving different British dialects, African languages (and possibly West African pidgins), and older forms of Caribbean English, yet little is known about the exact linguistic details (Schneider, 2007, 228). Slave populations increased drastically by the early 1700s, creating conditions for the emergence of a basilectal Jamaican Creole (Deuber, 2014, 28). Until emancipation in 1838, exposure to English among the slave population came almost entirely from fellow slaves, and even well after emancipation the spread of education among former slaves was slow, ineffective, and poorly attended (Christie, 2003, 12). For most of its history, Jamaica has been demographically divided, yet among all groups a “native” Jamaican identity seems to have emerged fairly early on (Schneider, 2007, 231). Lower class Whites and African slaves both saw themselves as Jamaicans, albeit in different ways (Lalla and D’Costa, 1990, 23). For obvious reasons, the Black (ex-)slave population did not desire to unify itself with its White Jamaican oppressors, and so looked to their shared African heritage to forge a sense of communion and even local pride among fellow slaves and hence fellow (Black) Jamaicans (Lalla and D’Costa, 1990, 25). These social divisions were reflected in the linguistic situation in Jamaica, where Black slaves at the lower end of the social stratum spoke mostly only Creole. Curiously, White settlers also used Creole widely, most having learned it during childhood contact with Black children (Lalla and D’Costa, 1990). Diglossia among White Jamaicans, was prevalent very early on, and key features of (what would become) Jamaican Creole developed in the eighteenth and nineteenth centuries (Lalla and D’Costa, 1990; Patrick, 2004). Schneider (2007, 234) places the start of Phase 4 in Jamaican English (JamE) in the mid-twentieth century, initiated by increasing democratization,

52

World Englishes and Dialect Typology

urbanization and socioeconomic diversification in the wake of the Second World War and Jamaican independence in 1962. English is the official language and standard English is taught widely in schools, although command of standard JamE still connotes an upper-class status and is acquired only by an elite few. The majority of Jamaicans today predominantly use a mesolectal form of Jamaican Creole, but most speakers make use of variation towards and away from both basilectal and acrolect ends of the linguistic continuum when the context requires it (Mair, 2002). Traditionally, standard JamE has been the language of formal domains (government, education) and contexts where written language is required (Schneider, 2007, 235), but in recent years there has been a shift towards a greater acceptance of Jamaican Creole in formal contexts (Deuber, 2014, 30– 31; Mair, 2002). Among younger generations, Jamaican Creole is accepted when used in newspapers, government business and the law courts, reflecting a shift away from an exonormative orientation towards Standard British English and RP in favor of an endonormative one in which the local “dialect” is seen as a symbol of Jamaican identity (Mair, 2002). Growth in the use of JamE in “serious” literature is yet another indicator of local acceptance and endonormative stabilization. 3.3.7

New Zealand English (NZE)

English settlement of New Zealand began in the late 1700s, and the Treaty of Waitangi between the British Crown and local Maori leaders established the official colony of New Zealand in 1840. The treaty was unusual for its time in that it recognized the Maori land ownership and control of traditional resources, and offered the Maori the rights and privileges of British citizenship (Macglagan, 2010). Nonetheless, following the signing of the treaty, successive waves of European settlers arrived in nineteenth century, attracted by the burgeoning gold-mining industry and direct offers of government assistance. The majority of these settlers were of English origin, chiefly from the south east, though substantial numbers of migrants arrived from Ireland and Scotland as well. Immigration from Australia was comparatively much less, although there remained intense commercial traffic between the two countries. It is thought that by the end of the 1800s, New Zealand-born English speakers had come to outnumber new immigrants (Macglagan, 2010), and it is at this point where true nativization of New Zealand English began. That is, the late nineteenth century is the point at which we see the emergence of distinctly New Zealand forms, largely as the result of extensive dialect contact (Trudgill, 2004, 24). Linguistically, the effects of nativization were most apparent in innovative aspects of phonology and lexis (Bauer and Warren, 2008), the latter involving loan words from Maori as well as semantic shifts and novel compounds (Schneider, 2007, 130). As with Canadian English, New Zealand English morphosyntax is often noted for its internal homogeneity, though there

3.3 Varieties in This Study

53

is growing evidence for the recent emergence of regional variation (Bauer and Bauer, 2002; Macglagan, 2010), placing New Zealand English at the threshold of Phase 5 in the Dynamic Model (Schneider, 2007, 132). 3.3.8

Philippines English (PhlE)

Philippines English (PhlE) is unique among the varieties discussed here in that it developed from American rather than British English input. After gaining control of the Philippines following the Spanish–American War (1898), the United States quickly adopted strong Anglo-centric policies for education and administration, facilitated by the importation of several hundred American English teachers in 1901 (Schneider, 2007, 140). English was promoted to the Indigenous population as a force for social and economic equalization, and the language spread rapidly during the first decades of the twentieth century. As the Philippines has gradually gained greater political sovereignty, there has been continual debate around the role of English and other Indigenous languages, particularly Tagalog, as official languages (Schneider, 2007, 141). Linguistically, Philippines English displays traces of its colonial history as a Spanish and then American territory as well as influence from the various local languages used across the country. Phonologically, PhlE retains an orientation towards American English norms, largely due to the implementation of American education systems and teaching methods (Pefianco Martin, 2010), while exhibiting a number of unique morphosyntactic features along with many others that are characteristic of Outer Circle varieties more generally (Pefianco Martin, 2014, 75). English continues to be the language of formal and public domains used in business, higher education, science and technology, politics, the (print) media, and government bureaucracy. In this regard, English is still valued largely for its instrumental role in affording greater social and economic mobility, rather than as a marker of shared Filipino identity (Pefianco Martin, 2014, 76). Schneider (2007, 143) notes a “foreshadowing” of endonormative stabilization (Phase 4) in PhlE, yet suggests that external forces may have stalled its progression. These external forces involved a simmering backlash against English, which arose in the 1980s after the end of the Marcos dictatorship, and renewed (re)orientation towards local languages and cultures. Debate about the theoretical position of PhlE continues (see Pefianco Martin, 2014, 78–81), although it is generally agreed that it falls somewhere on the cusp of Phase 3/4. 3.3.9

Singapore English (SgE)

As with many colonies in Southeast Asia, the settlement that would become Singapore began as a trading post for the British East India Company in 1819, and the linguistic and cultural environment from which Singapore English (SgE) evolved has always been one of intense cultural and linguistic contact. This environment included people of primarily Chinese, Malaysian, and Indian

54

World Englishes and Dialect Typology

ethnic heritage, who spoke a range of dialects and languages which formed the substrate from which SgE gradually evolved. These languages were chiefly Baba Malay (Malay spoken by those of mixed Malay and Chinese parentage) and Bazaar Malay (a pidgin variety of Malay), as well as numerous southern dialects of Chinese (e.g. Hokkien, Teochew and Cantonese) (Low, 2010, 231). Languages from southern India, most notably Tamil, have also been present as waves of migration from India continued through the nineteenth century (Leimgruber, 2013, 2–4). In the early nineteenth century, the British government set up Englishmedium schools for the purposes of creating an English-educated elite to fulfil local administrative roles, as in other colonies such as India. However, most national (non-British) schools used Malay, Chinese, or Tamil as the medium of instruction up until the 1950s, when two types of primary schools were established: “national” Malay-medium schools, and “national-type” schools, which used either English, Chinese or Tamil (Low, 2010, 231). After Singapore gained full independence in 1965, a bilingual education policy was promoted as a way of maintaining students’ connections with diverse ethnic traditions. Consequently, for non-English-medium schools, English became the chief second language on offer. The importance of English grew further as it was adopted as the medium of instruction at Nanyang University, which merged with the University of Singapore in 1972 (Low, 2010, 232). English is now taught in school as a “first language” to all children irrespective of their ethnicity, and thus serves as a common bond for the younger generations of Singaporeans (Schneider 2007: 156). Since the 1970s, rapid economic growth and expansive language policy gave rise to a modern state evincing a unique “Singaporean” identity that is “characterized by a blend of Western orientation in business and lifestyle with an emphasis on fundamentally Asian values” (Schneider, 2007, 156). Contemporary SgE exhibits ample evidence of structural nativization and endonormative stabilization (Phase 4). Looking ahead to Phase 5 (and beyond, see e.g. Wee, 2014), there is evidence of growing internal variation. For example, Singaporeans can accurately identify the ethnic background of SgE speakers (Deterding, 2010, 5), implying a recognizable degree of social, if not regional, differentiation. Moreover, the informal local variety known as Singlish continues to grow and diversify in spite of prescriptive efforts from the government to suppress its spread and use via the “Speak Good English” movement (Schneider, 2007, 160). In all, the means through which Singaporean identities are constructed and communicated have become increasingly sophisticated (Wee, 2014), and (standard) Singapore English and Singlish have become vital resources for Singaporeans to express solidarity and pride with their nation. In this regard, Singapore English has more in common with Inner Circle varieties than traditional Outer Circle ones (Leimgruber, 2013).

3.4 Summary

55

Table 3.1 English varieties and their theoretical categorization.

Variety

Date est.

BrE CanE NZE IreE SgE IndE JamE HKE PhlE

3.4

N.A. 1763 1790s 1600s 1819 1600s 1655 1841 1898

Kachru Circle

Dynamic Model

Learned as L1 or L2

Inner Inner Inner Inner Outer Outer Outer Outer Outer

Phase 5 Phase 5 Phase 5 Phase 5 Phase 4/5 Phase 4 Phase 4 Phase 3 Phase 3(4?)

L1 L1 L1 L1 L1/L2 L2 L2 L2 L2

Summary

In this chapter we briefly reviewed the current picture of modeling of World Englishes as well as the view of global variation in English from the perspective of dialect typology. We then provided a brief introduction to each of the nine regional varieties of English under study in this book. We placed these varieties within the context of two of the most prominent and influential models of World Englishes, namely Kachru’s (1985; 1992) Three Circles model, and Schneider’s (2007) Dynamic Model. The varieties we examine are a fairly representative sample covering both native (or “Inner Circle”) varieties and nonnative (or “Outer Circle”) varieties, and all have been shown to have progressed to the latter stages of the Dynamic Model (see Table 3.1). It should be kept in mind that our aims here are not to test or validate the position of any given variety within a specific theoretical category, but rather to use these models as generalization to which we can appeal when trying to understand our findings. With this in mind, our review of the sociohistorical backgrounds of these varieties has tried to draw attention to the diversity of sociolinguistic settings from which these varieties have evolved, as well as the commonalities among them that (may) give rise to the linguistic patterns scholars observe.

4

The Data

We begin this chapter with a general discussion of data types in variationist linguistics, focusing in particular on the role of corpora as sources of naturalistic production data. To study variation in production data, we tap into the International Corpus of English (ICE) and the Corpus of Global Web-Based English (GloWbE). Here we present the datasets of the three alternations collected for the present study, and review some advantages and disadvantages of the particular ICE and GloWbE corpora. Despite their acknowledged limitations, we argue that these corpora offer some of the best sources available for the kind of parallel comparisons we are trying to make here. To create our datasets, relevant observations of the competing syntactic variants were identified as extracted from the corpora, and then annotated for a wide range of probabilistic constraints briefly reviewed in Chapter 2. The details of this data collection and annotation are described in depth from Section 4.2 onwards. 4.1

Data Types in Variationist Linguistics

When it comes to empirical research in the variationist tradition, one could broadly sort the most common strategies used by (socio)linguists into those that rely on observational versus experimental approaches to primary data collection. In the context of the present study, we also find it instructive to draw a further contrast in observational approaches between the Language Variation and Change (LVC) paradigm of Labovian variationist sociolinguistics (Tagliamonte, 2012, 1), and what Szmrecsanyi (2017, 686) refers to as the “corpus-based variationist linguistics” (CVL) paradigm. In this section we outline some basic aspects of the LVC and CVL approaches, as well as a growing interest in more experimental approaches to studying variation. All three can be seen as flavours of variationist research provided that they are a) concerned with “alternate ways of saying ‘the same’ thing” (Labov, 1972, 188), and b) adhere to the Principle of Accountability (Labov, 1969, 737– 738), that is, they consider all possible variants that could have been used in a given variable context. Despite sharing these key assumptions, each of these approaches have their respective strengths and limitations, which we discuss below. 56

4.1 Data Types in Variationist Linguistics

4.1.1

57

Small and Specialized Data in LVC Research

A defining aspect of the LVC tradition is the use of naturalistic production data from sociolinguistic interviews and ethnographic observation, which have been the bread and butter of sociolinguistic methodology for decades (Meyerhoff, 2016). As Szmrecsanyi (2017, 686) notes, to the extent that the datasets in this approach represent carefully compiled collections of naturally occurring language data, such work can be seen (perhaps provocatively) as a type of corpus-based approach, albeit one focused on the analysis of specialized corpora (see also Tagliamonte, 2012). Perhaps the most obvious advantage to the use of such data lies in the level of insight into the socio-cultural backgrounds of speakers and communities that it can provide. Close ethnographic approaches offer unparalleled access to language users and their lived experiences, yielding a richer understanding of the sociolinguistic dynamics at play within their communities. Indeed it is hard to see how such understanding could be achieved through any other means. Another key aspect of the LVC approach is the emphasis on vernacular speech, where “variation in language is most readily observed” (Tagliamonte, 2012, 2). Vernacular styles are thought to involve minimal self-monitoring on the part of speakers, making them (arguably) the optimal domain for investigating variation unimpeded by speakers’ conscious manipulations (Labov, 1972). Concomitant with its focus on vernacular speech, LVC research has predominantly tended to focus on phonetic variables, which are particularly useful both for their conveniently high degree of frequency in even short bits of speech, as well as their important role in variation and change below the level of consciousness (see Preston and Niedzielski, 2010a for discussion). Nonetheless there has been, and continues to be, a great deal of work on lexical and morphosyntactic variation within the LVC tradition (e.g Cheshire, 2005; Labov, 1969; Sankoff, 1988; Poplack and Tagliamonte, 1999; Tagliamonte et al., 2016a; Tagliamonte and Denis, 2010; Weiner and Labov, 1983), though again, most of the LVC work beyond the phonetic/phonological level still targets variation in spoken vernacular registers (D’Arcy, 2020). Due to the high resource demands of sociolinguistic interviews and ethnographic methods, by modern standards LVC datasets (corpora) tend to be small and narrowly focused. However, this small size allows for detailed manual annotation and a much closer connection between the researcher and the data. In contrast to recent trends in corpus linguistics and data science toward the use of “big data” (e.g. Hiltunen et al., 2017), traditional LVC research could thus be said to represent a kind of “small data” paradigm in empirical linguistics. The label “small” here is in no way intended as a comment on the overall value of such work, whose contribution to our understanding of language is immeasurable. We use the term as a metaphorical contrast to corpus-based approaches that make use of massive corpora and/or other

58

The Data

large-scale data sources (e.g. Twitter) on the order of hundreds of millions or billions of words. By comparison, the datasets compiled through traditional LVC research are typically on the order of hundreds of thousands or a few million words. Examples of such corpora include The Buckeye Corpus,1 the North Carolina Sociolinguistic Archive, and Analysis Project,2 and the Corpus of Regional African American Language.3 4.1.2

Big(ger) and Generalized Data in Corpus-Based Research

With the increasing availability of massive amounts of electronic language data and rapid advances in computing power, corpus-based variationists have turned to using data sources whose size and breadth far outstrip those used by more traditional LVC sociolinguists. Readily available online corpora now range in the size of tens of millions to tens of billions of words, making it possible to examine infrequent lexical and grammatical phenomena while at the same time considering more familiar variables within a broader range of linguistic contexts. The present study makes extensive use of such corpora, described in section 4.2, and thus is situated firmly within the corpus-based variationist (CVL) paradigm. Corpus-based variationist research does not necessarily involve the use of massive datasets – much variationist work has been done using relatively small (1–2 million words) corpora (e.g. Gries, 2003; Hinrichs and Szmrecsanyi, 2007; Bernaisch et al., 2014; Shih et al., 2015), and the value, even “beauty” (Hundt and Leech, 2012), of such sources is hard to deny. What sets CVL approaches apart is chiefly their reliance on existing corpora and/or data sources not directly compiled or elicited by researchers (e.g. news or academic articles, Twitter, blogs or websites). Such sources often contain a broad selection of users and/or registers rather than carefully sampled data from specific communities, as many such corpora were designed to be representative of language at a fairly abstract macro-social or regional level, such as the British National Corpus or the Corpus of Contemporary American English. Therefore CVL researchers typically analyze data that has been aggregated from many speakers, registers, and/or varieties, about which there may be relatively scant social or demographic information in the corpus. This certainly limits our ability to probe the sociolinguistic details of individual users and features, but we argue the CVL approach nonetheless possesses a great deal of potential for contributing to linguistic theory and description. 1 https://buckeyecorpus.osu.edu/ 2 https://www.lib.ncsu.edu/projects/north-carolina-sociolinguistic-archive-and-analysis-project

-ncslaap 3 https://oraal.uoregon.edu/coraal

4.1 Data Types in Variationist Linguistics

59

Aside from data size, another notable contrast between LVC and CVL research is the latter’s enthusiasm for exploring variation beyond the vernacular. CVL researchers tend to be much more open to the study of both spoken and written data, and in particular there is growing interest in the intersection(s) of register, variety, and language change (Grafmiller, 2014; Hinrichs and Szmrecsanyi, 2007; Hundt et al., 2021). Such variation across registers and modes has largely been ignored by LVC researchers, with some exceptions (Tagliamonte et al., 2016b), though we believe this is a area of considerable theoretical relevance (see e.g. Guy, 2015). The development of more sophisticated analytical techniques is another domain in which CVL studies are leading the way. While LVC researchers were (very) early adopters of what would become essential quantitative methods, for example, binomial logistic regression (Cedergren and Sankoff, 1974), the mantle of methodological innovation has largely been taken up by corpus linguists in recent years (though see Tagliamonte and Baayen, 2012; Tamminga et al., 2016b). Such methods include multinomial regression (Levshina et al., 2013; Grafmiller et al., 2016), multimodel inference (Barth and Kapatsinski, 2017), Bayesian regression and networks (Grafmiller et al., 2016; Theijssen et al., 2013), memory-based learning models (Theijssen et al., 2013), discriminative learning models (Baayen et al., 2013), as well as more sui generis approaches (e.g. Gries and Deshors, 2014; Szmrecsanyi et al., 2019). Moreover, these methodological advances have contributed considerably to a wide variety of subfields including cognitive linguistics (e.g. Claes, 2016; Pijpops and Van de Velde, 2018), psycholinguistics (e.g. Arnold et al., 2000; Levy and Jaeger, 2007), historical linguistics (e.g. Wolk et al., 2013), L2 acquisition (e.g. Wulff et al., 2014; Paquot et al., 2019), and of particular relevance to this book, World Englishes research (e.g. Bernaisch et al., 2014; Grafmiller and Szmrecsanyi, 2018). It should be added that these advancements are in no small part a response to demands of dealing with larger and more complex (and perhaps messier) datasets. Finally, perhaps the most unassailable argument for the value of CVL methods is in the study of language change, where corpora are necessary for investigating patterns over periods longer than an individual human lifespan (e.g. Wolk et al., 2013). Apparent time studies have been a bedrock of LVC research into language change, yet the method has a hard limit in terms of the time depth it can reach. As imperfect as they may be, historical corpora are the only sources of data we have for investigating fine-grained linguistic changes (as opposed to methods for historical reconstruction) beyond the limits of a single human lifespan. For these reasons and more (see e.g. Szmrecsanyi, 2017 and Hansen, 2018 for further discussion), we argue that the plethora of theoretical and methodological perspectives encompassed by the CVL community is a substantial

60

The Data

contribution in its own right, and offers tremendous potential for crosspollination with variationist sociolinguistics. 4.1.3

Experimental Data in Variationist Research

Experimental approaches involve studies and frameworks whose primary methods are designed to directly elicit linguistic data, and increasingly, variationist researchers are turning to experimental and survey-based approaches as a means of exploring variation at various levels of linguistic and social structure. The direct elicitation of data via, for example, reading word lists and/or short judgment surveys, has been a component of variationist sociolinguistics since its inception (Meyerhoff, 2016), but modern technology is providing means to massively scale up our data collection methods, which is incredibly appealing for sociolinguistic research (see e.g. Kim et al., 2019; Leemann et al., 2016). However, experimental approaches are somewhat limited in terms of the social information they can provide/obtain, and there are always concerns about ecological validity which must be considered (Campbell-Kibler, 2010). But the benefits very often outweigh the concerns, and we mention a few advantages of such methods here. A particularly attractive advantage of survey methods is that they can directly target specific phenomena and variables, especially rare ones for which naturally occurring examples may be hard to come by. At the same time, more carefully controlled experimental settings allow researchers to investigate questions about variation, social meaning, and cognition that are difficult to test via purely observational methods (e.g. Walker and Campbell-Kibler, 2015; Walker, 2019). Even when corpus-based methods are available, experiments can nonetheless provide independent corroboration of observational (corpus-based) evidence and support for theoretical models of language structure and variation (e.g. Bresnan and Ford, 2010; Hoffmann, 2011; Divjak et al., 2016). In recent years a variety of experimental techniques have been used to complement observational studies of linguistic variation, including reformulation or sentence completion tasks (Ford and Bresnan, 2015), lexical decision tasks (Ford and Bresnan, 2013), direct/indirect grammaticality judgment tasks and/or forced-choice tasks (Buchstaller and Corrigan, 2011; Divjak et al., 2016; Rosenbach, 2005), and magnitude estimation and rating tasks (Bresnan and Ford, 2010). In Chapter 7 we introduce our own contribution to this body of work with an experiment designed to compare results from our corpus data to those of human participants in an offline judgment task. 4.2

The Corpora

In this section we describe our corpus-derived datasets, beginning with a brief overview of the corpora used, followed by the details of how we defined the

4.2 The Corpora

61

variable contexts for identifying and extracting examples of our three alternations. Lastly we detail the procedures for annotating the linguistic features and constraints included in our analyses. Our corpus analyses presented in this book are based on data collected from two corpora of international Englishes: the International Corpus of English (Greenbaum, 1991; 1996), and the Corpus of Global Web-based English (Davies, 2013). These corpora are well-known to, and widely used by researchers of World Englishes and English corpus linguistics, and the relative merits of both have been the topic of much discussion in recent years (e.g. Davies and Fuchs, 2015; Kirk and Nelson, 2018; Loureiro-Porto, 2017). Although we cannot cover every detail of these debates here, we briefly introduce each corpus in this section and lay out the reasons why we believe these corpora are useful sources for variationist research. Despite its name, the International Corpus of English (ICE) does not constitute a single unified corpus, but rather a collection of region-specific individual component corpora, each of which was compiled by different teams across different countries. Compilation of the first ICE components began in the early 1990s, with the goal of creating a set of parallel corpora representing a wide range of English registers in regions around the world. ICE compilation continues in various regions up to the present day. Each ICE component is made up of of 500 texts (60 percent spoken; 40 percent written) of roughly 2000 words each, resulting in approximately one million words per component. Together, the nine ICE components used our study total over nine million words of running text. The spoken ICE material consists of dialogues from face-to-face and telephone conversations, as well as scripted and unscripted monologues. The written material consists of texts from a variety of printed and nonprinted sources, including letters, student essays, academic and popular nonfiction, creative writing, and news reportage (Table 4.1, Nelson, 1996). To be included in the corpus, speakers/writers must have received formal education and completed secondary school or have an appropriate public status, such as as politicians, writers, or educators (Greenbaum, 1996, 6). Thus the language in each ICE component is primarily representative of what could be considered the educated, standard or acrolectal variety of English in its respective region. Comparisons of ICE with other “standard” English corpora of different designs (e.g. the British National Corpus and FLOB) find a high degree of consistency with regard to various lexical and morphosyntactic patterns (Loureiro-Porto, 2017, 466–467), providing independent validation for the use of ICE as a representative corpus of standard English(es). At 1.9 billion words, the Corpus of Global Web-based English (GloWbE) is several orders of magnitude larger than ICE, and comprises tens of millions of words from websites in twenty different countries around the world (six Inner Circle and fourteen Outer Circle countries, Table 4.2). As with ICE, GloWbE

Table 4.1 Design of the ICE corpora. Values reflect number of texts. Mode Spoken 300

Genre Dialogues

Monologues

Written 200

Nonprinted

Printed

Subgenre 180

120

50

150

Text Type

Private

100

Public

80

Unscripted

70

Scripted

50

Student writing

20

Letters

30

Academic writing

40

Popular writing

40

Reportage Instructional writing

20 20

Persuasive writing Creative writing

10 20

Face-to-face conversations (90) Phone calls (10) Classroom lessons (20) Broadcast discussions (20) Broadcast interviews (10) Parliamentary debates (10) Legal cross-examinations (10) Business transactions (10) Spontaneous commentaries (20) Unscripted speeches (30) Demonstrations (10) Legal presentations (10) Broadcast news (20) Broadcast talks (20) Nonbroadcast talks (10) Student essays (10) Exam scripts (10) Social letters (15) Business letters (15) Humanities (10) Social sciences (10) Natural sciences (10) Technology (10) Humanities (10) Social sciences (10) Natural sciences (10) Technology (10) Press news reports Administrative writing (10) Skills/Hobbies (10) Press editorials Novels and short stories

Corpus code s1a s1b

s2a

s2b

w1a w1b w2a

w2b

w2c w2d w2e w2f

4.2 The Corpora

63

Table 4.2 Design of the GloWbE corpus (from Davies and Fuchs 2015, 6). Code

Websites

Webpages

United States Canada Great Britain Ireland Australia

US CA GB IE AU

82260 33776 64351 15840 28881

275156 135692 381841 102147 129244

386809355 134765381 387615074 101029231 148208169

New Zealand India Sri Lanka Pakistan Bangladesh

NZ IN LK PK BD

14053 18618 4208 4955 5712

82679 113765 38389 42769 45059

81390476 96430888 46583115 51367152 39658255

Singapore Malaysia Philippines Hong Kong South Africa

SG MY PH HK ZA

8339 8966 10224 8740 10308

45459 45601 46342 43936 45264

42974705 42420168 43250093 40450291 45364498

Nigeria Ghana Kenya Tanzania Jamaica

NG GH KE TZ JM

4516 3616 5193 4575 3488

37285 47351 45962 41356 46748

42646098 38768231 41069085 35169042 39663666

340619

1792045

Total

Words

1885632973

was designed to capture a range of (web-based) registers and genres, with the balance tipped more toward informal blogs (60 percent), in parallel with ICE’s greater preponderance of spoken data. The remaining “general” component involves various other genres and text types of an often more formal style (e.g. news sites). Being a massive corpus of online texts collected from websites around the world, GloWbE necessarily exhibits a relatively high degree of variability and heterogeneity in its data sources, and considerable effort was made to verify the quality and provenance of the data sources (Davies and Fuchs, 2015).4 In order to keep our GloWbE dataset to a manageable size, we created a subsample of the full corpus (2014 version) by selecting texts at random from each of the nine varieties, up to a total of 500 thousand words per variety. This resulted in a custom GloWbE subcorpus of roughly 4.5 million words, approximately half the size of the combined ICE components. 4 Details of the methods can be found online here: https://www.english-corpora.org/glowbe/help/

textsm.asp

64

The Data

Both ICE and GloWbE have been questioned regarding their cross-variety comparability (e.g. how parallel are the various ICE components?), their internal heterogeneity (e.g. what kinds of genres and registers do the GloWbE texts truly encompass?), and the accuracy of the data itself (e.g. how can we be sure whether a text/writer in GloWbE truly comes from a given region?). These are important questions which are actively being investigated by corpus compilers and linguists (Davies and Fuchs, 2015; Edwards, 2017; Kirk and Nelson, 2018), and we take a moment to briefly address them here. First, regarding the cross-varietal comparability of the ICE corponents, we note that some degree of variability is inevitable in corpora of this size and scope. Despite the best efforts of ICE compilation teams, the sampling, and hence representativeness, of registers in each component will vary somewhat depending on the availability of English texts/speakers in a given region. We are thus aware of potential differences among registers of certain varieties in the ICE corpora. Still, the present study compares these varieties at a relatively abstract level of grammatical structure using statistical methods such as multilevel modeling, which are designed to smooth over and account for (to a certain degree) some of the register-specific variability in the data. We aim to capture broad patterns of variation at the scale of what could be considered standard national varieties of English, and given this aim, we believe the ICE corpora are the best resources currently available for parallel analyses of regional varieties of this kind. Second, we add that while the precise proportions of blog data versus general data in GloWbE are at best a rough estimate, their relative balance is not a concern for us given the size of the corpus – there is more than enough data to work with. Of greater potential concern is the representativeness of these webbased registers, and their comparability to the data in ICE. In particular, we do not know to what extent the informal blog data in GloWbE should be seen as comparable to the ICE spoken data. Research directly comparing linguistic features across web-based and “off-line” registers suggests that blogs are indeed distinct from spoken registers, though perhaps not exactly like other written registers either (e.g. Biber and Egbert, 2018; Grieve et al., 2010; Sand, 2013). For example, in a comparison of twenty-one “oral” language features in ICE and GloWbE, Loureiro-Porto (2017, 457–460) found that these features were substantially less frequent in GloWbE blogs compared to ICE spoken registers, and that the GloWbE patterns were far more consistent with those of ICE written registers. Moreover, differences between the informal blog and (arguably) more formal “general” components of GloWbE were nonsignificant, suggesting that the language in GloWbE may be more akin to other (informal) written registers than spoken language. It is still unclear where exactly to place blogs on a continuum (or continua) of register variation vis-a-vis spoken language, however, what does seem certain is that the ICE and GloWbE should not be seen as equivalent alternatives (Loureiro-Porto, 2017, 468). With this in mind,

4.3 Defining the Variable Contexts

65

we primarily analyze our ICE and GloWbE data as separate datasets rather than simply lump data from both sources into a single analysis. It is also worth noting that the source and URL information for each text is available in the corpus, which allows for more detailed annotation and/or examination of specific texts and registers if desired in the future. Finally, it is likely inevitable that some of the writers in the GloWbE corpus (or any web-based corpus of even moderate size) are not users of the dialect in question. Again, GloWbE does provide researchers with the URLs and links to the original web pages for each of the 1.8 million web pages in the corpus, and as Davies and Fuchs (2015) point out, when the number of web pages one is working with is manageable, one can examine the pages individually. Inspecting each page is certainly not feasible for larger studies such as the present one, however, we add that informal inspection of our GloWbE texts during the manual filtering and annotation phases did not reveal any obvious inconsistencies or systematic issues regarding the representativeness of the data. We therefore have no reason to believe that the GloWbE data is systematically biased in this particular way, though we cannot say with absolute certainty that every one of our tokens comes from a user of the dialect specified in the corpus. To sum up, we are aware of the potential issues surrounding the representativeness of these corpora, and we have implemented methods to address them as much as possible within the context of our overall research goals. Nevertheless, we acknowledge the limitations of these data sources, and stress caution about generalizing beyond our findings to more focused comparisons of individual registers and/or regions. Still, we maintain that even with the differences in compilation and design, parallel and comparative study of crossvarietal patterns in ICE and GloWbE can be fruitful. Finding large differences between the corpora raises numerous potential questions about the relationship between register, time and geography (among other things), which can be further investigated through closer examination of the specific varieties and linguistic patterns. On the other hand, finding a high degree of similarity between the two corpora would be all the more striking in light of the considerable differences in their composition and origins. We thus believe that ICE and GloWbE provide appropriate and useful data sources for comparative grammatical analysis at this scale. For further exploration of these and other issues, we point interested readers to the discussion in Hansen (2018, chapter 11). 4.3

Defining the Variable Contexts

In this section we describe in detail the methods for identifying and extracting interchangeable tokens of the three alternations. An essential aspect of variationist analysis is identifying the range of environments in which variation can occur, the so-called “envelope of variation” or “variable context” (Tagliamonte, 2012, 10). At the morphosyntactic level, this can be particularly challenging

66

The Data

(Lavandera, 1978; Szmrecsanyi et al., 2016b), and part of the reason for focusing on the genitive, dative, and particle placement alternations is that we can rely on a great deal of scholarship to guide our data extraction. Procedures for collecting, filtering, and annotating the datasets described below were developed and carried out by the second author along with two PhD students, Benedikt Heller and Melanie Röthlisberger. The procedures followed methods outlined in the literature for the respective alternations, each of which are discussed below. Detailed extraction and annotation manuals can be found in the project OSF repository. 4.3.1

The Genitive Alternation

For identifying and extracting interchangeable genitives we used a two-step semi-automated process. First, all potential s- and of -genitive tokens were automatically identified and classified for interchangeability via scripts. Second, classifications made by the script were manually checked and false positives and negatives were corrected and/or removed. Identification and selection of valid tokens followed well-established procedures in the literature for identifying interchangeable genitive tokens (e.g. Rosenbach, 2002, 2014; Heller, 2018; Hinrichs and Szmrecsanyi, 2007; Grafmiller, 2014). Any potential token falling into one of the categories described below was excluded from the dataset. s-genitives lacking an explicit possessum. These include independent genitives where the possessum was omitted, as in (11). (11)

a. We’ll meet at Benedikt’s. b. That’s not my car, that’s Jason’s.

Descriptive s-genitives. Descriptive s-genitives (12) are not interchangeable with the of -genitive, and thus were excluded. (12)

a. Get in touch with your creative side and bring your stories to life by writing a children’s book. b. glancing at himself in the mirror and seeing how silly he looked in his hipster’s suit and no hat c. A Boomer’s Guide to Calgary d. Their influence as justices’ aides debated

Descriptive s-genitives can be distinguished from “true” s-genitives based on a number of semantic and formal properties (Rosenbach, 2006). Semantically, the possessor in a true s-genitive functions as a determiner, specifying and “anchoring” (Taylor, 1996) the referent of the NP. In contrast, the dependent (possessor) in descriptive genitives lacks specific reference, functioning more as a classifier than a specifier or determiner. Structurally, the dependent

4.3 Defining the Variable Contexts

67

noun of a descriptive genitive does not form a constituent with the determiner, if one is present, unlike in true genitives. Descriptive genitives (sans determiner) behave more like noun + noun compounds in this regard. This is evident from the fact that the number of the determiner agrees with the head (possessum) in descriptive genitives (13-a), but with the dependent (possessor) in true genitives (13-b). (13)

a. a children’s book = [a [children’s book]], descriptive reading only b. a child’s toys = [[a child]’s toys], possessive reading only c. a child’s toy = [a [child’s toy]] ∼ [[a child]’s toy], descriptive or possessive reading

Plural determiners likewise force a possessive genitive reading if the head is singular, for example, those children’s toy. Measure s-genitives. Measure s-genitives typically indicate temporal duration (14) or value (15) (see Biber et al., 1999, 296; Huddleston and Pullum, 2002, 470). These constitute a case of special s-genitive construction, which behaves similar to descriptive genitives and often involves semi-idiomatic expressions, such as arm’s length, a stone’s throw. (14)

(15)

a. Make sure you’re not dishevelled after a full day’s work. b. When you arrive in New Zealand, have at least a month’s supply of your medications on hand. I’ve just bought thirty-thousand pounds’ worth of shares.

As with descriptive s-genitives, measure s-genitives are not interchangeable with the of -genitive. Indefinite of -genitives. With true s-genitives the whole genitive NP is rendered definite by the presence of the referential possessor. A corresponding of -genitive construction must therefore also be definite and begin with the definite article, as s-genitives cannot have another determiner. Indefinite of genitives such as those in (16) do not have corresponding s-genitives, and therefore were also excluded. (16)

a. A picture of their cat b. Another child of the pre-irony days

Appositive of -genitives. These are tokens where the possessor and possessum are o-referential, as in (17). Generally these can be paraphrased by omitting the possessum entirely. (17)

a. We live in (the city of) Birmingham. b. Spain is an option but (the idea of) tasting a new culture and learning another language excites me the most.

Appositive of -genitives do not have equivalent s-genitive counterparts.

68

The Data

Double genitives. Tokens where possession is marked twice were excluded. (18)

A painting of Pete’s also forms part of this type

Noninterchangeable semantic relations. There are number of semantic relations that can be expressed with the of -genitive yet do not allow semantically equivalent s-genitive alternatives (Payne et al., 2013, 816). These include, but are not necessarily limited to, of -genitives expressing the relations in (19). (19)

a. Abstract or physical CONTENT: the study of linguistics, the question of law, the bottles of wine b. COMPOSITION or QUALITY: streaks of paint, a ring of gold, a person of honour c. SIZE, VALUE or AMOUNT: farm of 100 acres, 200 grams of rice d. MEMBERSHIP or COLLECTIONS: the team of scientists, the collection of books

We were careful to exclude all of -genitives that lack semantically equivalent s-genitive counterparts. 4.3.2

The Dative Alternation

Extraction of interchangeable datives proceeded in a step-wise fashion similar to that used for the genitives. To create the dative alternation dataset, we first compiled a list of 225 dative verb lemmas identified in previous research. Consulted works included reference texts (Levin, 1993) as well as diachronic studies (Wolk et al., 2013) and studies examining different varieties of spoken and written English (e.g. Mukherjee and Hoffmann, 2006; Bresnan et al., 2007; Theijssen, 2012; De Cuypere and Verbeke, 2013; Schilk et al., 2013). All potential ditransitive or prepositional dative tokens involving any of these verbs were automatically extracted from the corpora. The final list of dative verbs found across our nine ICE components is provided in (20). (20)

accord, advise, allocate, allot, allow, answer, appoint, ask, assign, assure, award, bequeath, bid, bring, call, carry, cause, cede, charge, concede, convey, cost, deal, deliver, demonstrate, deny, develop, drop, entrust, explain, extend, feed, flick, flip, forward, get, gift, give, grant, guarantee, hand, impart, inform, issue, keep, lease, leave, lend, loan, lose, mail, name, offer, owe, pass, pay, permit, play, pose, post, prescribe, present, promise, propose, provide, quote, read, recommend, refuse, render, sell, send, serve, set, show, sing, slip, submit, suggest, supply, take, teach, tell, throw, toss, vote, wish, write, yield

Once identified, potential tokens were semi-automatically classified for interchangeability, again following procedures set down by previous work on

4.3 Defining the Variable Contexts

69

English datives (e.g. Bresnan et al., 2007; Theijssen, 2012; De Cuypere and Verbeke, 2013; Wolk et al., 2013). As with the other alternations, an interchangeable observation was defined as any dative token not falling into any of the categories described below. See Röthlisberger (2014, 2018b) for extensive details of the extraction and selection procedure. Datives with noncanonical structure. Instances involving heavy NP shift (21), or with a missing to and/or nonstandard order (22) were excluded. (21)

a. And there’s a very recent decision of the Supreme Court of Canada and I’ll read to you two paragraphs to give you a little more detail as to what is maybe a word meant by a sexual assault. b. We consider it vital to bring to the attention of the Council members the present business and financial position of the HKAC.

(22)

a. Now which brings me my second point. b. And I find I gave it some of my M.Phil. students who had problems with usage and all that and they said it has been useful. c. She gave it him. (Hughes et al., 2012, 20)

Dative particle verbs. We excluded all tokens involving dative particle verbs, since there are in principle more than two variants (Röthlisberger, 2018b, 56). (23)

a. Then he gave the keys back to his mother. b. Then he gave back the keys to his mother. c. Then he gave his mother back the keys.

Datives with clausal constituents. Tokens where the Theme constituent is a clause were excluded (Theijssen, 2012). (24)

a. Call Jean and Denille and tell them [we’re coming]. b. Mark I was telling Rachel [the deaconess introduced you to Jean]. c. So I asked him [what treatment does is he going to give].

Benefactive constructions. Tokens where the non-Theme argument is a beneficiary rather than a recipient were excluded. These differ from the dative alternation in that the double object construction alternates with a for PP rather than a to PP (Levin, 1993, 49). (25)

a. With some great difficulty last year we could get a couple of books for some students. b. I had the rare extreme delight of meeting the president of Iceland, who promptly asked me to sing a song for her.

70

The Data

4.3.3

The Particle Placement Alternation

As with the datives data, we relied partly on specific lexical items to guide our search for examples of interchangeable particle verb tokens. For this dataset, we collected all instances of transitive particle verbs containing one of the 10 particles in (26). (26)

around, away, back, down, in, off, on, out, over, up

This list is based on the method used by Gries (2003, 67–68) who selected the ten verbal heads and ten particles with the greatest combinatorial potential in a list of 1357 transitive particle verbs found in British English (Gries, 2003, 203–210). Since the present study is based on data from numerous varieties, searches were not restricted by verbal head in order to capture uses that may be idiosyncratic to specific varieties. All potential particle verb tokens were extracted from the part-of-speech tagged ICE corpora using regular expression searches for tokens matching any verb followed within a ten-word window by one of these ten particles (see the OSF repository for details). Two concerns motivated us to restrict the data to only these particles. First, while part-of-speech tagging of verbs tends to be reliable, tagging of particles in these corpora is much less so – particles are often (unpredictably) tagged as either prepositions or adverbs. Early extraction tests using partof-speech tags fared poorly in terms of both precision and recall, forcing us to rely on much less restrictive simple string searches.5 The results of such searches require extensive manual postfiltering due to very many false positives, and the considerable time demands of such manual filtering was our second concern. Hence we opted to focus on only the most frequent particles. We believe this method is sufficient to capture the bulk of particle verb tokens in the data, but we acknowledge that it could potentially overlook novel and/or region-specific usages of other particles. We leave a more fine-grained investigation of individual particle verbs within specific varieties to future research. Following the automatic extraction using the method above, observations were manually filtered from the dataset. Again, an acceptable observation was negatively defined as an interchangeable transitive particle verb not meeting any of the criteria below. 5 More sophisticated methods for automatic extraction of particle verbs have been developed

(see e.g. Baldwin and Villavicencio, 2002; Kim and Baldwin, 2010), yet these methods are not always straightforward for other researchers to implement. More important, until recently such NLP tools were usually trained on datasets of written standard varieties of British or US English, such as the Wall Street Journal Corpus. The success of these algorithms on particle verb extraction in other regional and/or non-standard varieties remains to be seen. Preliminary tests of readily available tools such as Python’s spaCy parser (https://spacy.io/) suggest that it performs quite well, at least on American English data, but this was not available to us at the time of our initial data extraction.

4.3 Defining the Variable Contexts

71

Passive tokens and extracted direct objects. Tokens where the verb is passive (27), or the direct object is either an extracted wh- element or a relative pronoun (28) were excluded. In these cases we cannot be sure what the order would have been had the object been in its “basic” position. (27)

But the idea is that the crude birth rate should be brought down from thirty point four per thousand

(28)

a. Whati did Sansa pick i up i? b. The beliefs about god thati this survey throws

i

up

i

Doubled particles. In rare cases the particle is repeated before and after the direct object, and these tokens were excluded. Such tokens tend to show up mainly in spoken data. (29)

a. And Brian my brother take off the crocodiles off them b. I don’t think they appreciated it cos they turned up theirs up louder which wasn’t very helpful really.

Modified particles and adverbials. Tokens where there is an adverb immediately before the particle were excluded (e.g. Bolinger, 1971; Fraser, 1976; Gries, 2003). Most often these are adverbs such as right or straight, but such examples may involve other adverbs (31). (30)

a. Threw it right down b. Like hodds had like put the window half up c. My daughter really enjoys dunking, and the chute that sends the ball straight back to you is a great feature.

(31)

a. He loosened his fingers carefully so that they might not snag the sheer chiffon and laid it gently down on the corner of the bureau. b. Brought his hand slowly down upon the knob. c. She reaches up, takes off his hat, and tosses it casually away. d. He tilted his head slowly back to stare at the hot, hard sky.

Phrasal-prepositional verbs. We further excluded cases of “Phrasalprepositional verbs” (32), which, according to Thim (2012, 28–29), take prepositional phrases rather than noun phrases as their complements. There are also some verbs that take direct objects as well as a PP complement, such as let X in on Y (33). (32)

a. I just can’t keep up with the language, Mrs Sangster said. b. Now, if you will excuse me, I must look in on father. c. The cancer caught up with him in 1980 . . .

(33)

But some doctors don’t even let parents in on the whole truth.

72

The Data

Both cases above were excluded. However, there is a smaller subset of phrasal-prepositional verbs, such as mix up with and fix up with, which are interchangeable (34). (34)

a. Until you mix up science with spirituality, you will never understand the truth. b. The “Gospel of Prosperity” has mixed Jesus up with money. c. But if you’re one of these taxi drivers I’ve heard about who fixes foreigners up with prostitutes d. Get her admitted to a rehabilitation centre or fix up appointments with a psychologist.

These tokens are not common, but any such tokens we found were examined closely for interchangeability. Prepositional verbs. Cases involving “Prepositional verbs” (Thim, 2012, 28– 29), where words such as on or up function as a preposition (35), were treated with extra care to distinguish them from interchangeable particle verbs (36). (35)

He called on the United States to end its economic embargo of Cuba. [Prepositional verb]

(36)

She switched on the powerful docking lamps. [Particle verb]

In (35), on is part of the prepositional verb call on, while in (36) on forms part of the particle verb switch on (see also Quirk et al., 1985, chapter 16 and Biber et al., 1999, chapter 15). Despite their surface similarities with “true” particle verbs, prepositional verbs are not interchangeable. To identify prepositional verbs, we relied on a number of heuristics that have been proposed to distinguish them from particle verbs (Bolinger, 1971; Cappelle, 2005; Rodríguez-Puente, 2013). Because these tests are (arguably) independent of interchangeability, they were the primary criteria for including or excluding a given token. (37)

Intervening adverb test: Only prepositional verbs allow adverbs between the head and preposition in the continuous (joined) variant. a. He called repeatedly on the United States to end its economic embargo of Cuba. b. *She switched repeatedly on the powerful docking lamps.

(38)

Preposed PP test: Only prepositional verbs allow pied-piping of the preposition and its argument. a. Is the US the country on which he called to end its economic embargo? b. *Are these the lamps on which she switched?

(39)

It-cleft test: Only prepositional verbs allow it clefts with focused PPs.

4.3 Defining the Variable Contexts

73

a. It was on the United States that he called to end its economic embargo. b. *It was on the docking lamps that she switched. Finally there is the pronoun test, which is a reliable indication of a prepositional verb. It is unquestionably acceptable for a prepositional verb to take a pronominal direct object (40). (40)

Even Turkey, once one of Syria’s closest allies, has called on it to institute reforms.

This is generally not the case with continuous particle verbs (41).6 (41)

a. She walked over to the fan and switched it on. b. *She walked over to the fan and switched on it.

More important, the split order is never acceptable with prepositional verbs (42). (42)

4.3.4

*Even Turkey, once one of Syria’s closest allies, has called it on to institute reforms. General Exclusions

In addition to the alternation specific criteria just described, there were a few general criteria for selection that we applied to all three alternations. These included the following. Tokens with missing material. For some tokens, particularly spoken ones, information for one or more of the constituents was sometimes missing or unable to be unambiguously transcribed. These were naturally omitted from the dataset. Names, titles, and quoted text. Any tokens that were names, titles, or quoted text were excluded since they do not alternate (e.g. Harper’s Ferry, Kepler’s law, Chamber of Commerce). These tend to be much more common with genitives than the other alternations. Fixed expressions. Multiword expressions that are seemingly fixed in one variant are particularly vexing for studies of morphosyntactic variation, as intuitions about interchangeability are not always reliable (see e.g. Bresnan et al., 2007). The interchangeability of specific instances can be difficult to determine even for native speakers of English, and ideally, reliance on subjective assessments of interchangeability should be minimized in the selection 6 This test is essentially a modification of Bolinger’s (1971, 112) “definite NP test,” and there are

arguably instances in which the continuous order is acceptable with a pronominal direct object.

74

The Data

process. When manually filtering the data, coders were careful to keep in mind that what is categorical in one variety of English, may be variable in others, and that it can be difficult to distinguish between what is grammatically “possible” and what is merely very unlikely in a given context. To illustrate, one might question the interchangeability of a particle verb token such as (43), which involves the semi-fixed expression take on the job, which tends to heavily favor the continuous variant. (43)

a. It was Britain’s ambassador Sir David Hannay who took on [the thankless job of explaining why Washington and London would have nothing to do with the proposal from Paris]DirObj .

b. *It was Britain’s ambassador Sir David Hannay who took [the thankless job of explaining why Washington and London would have nothing to do with the proposal from Paris]DirObj on.

In uncertain cases like the one above, coders checked for alternate instances of a token using online corpora, for example, BNC, COCA, GloWbE, or Google Advanced searches restricted to English language sites from particular regional domains (e.g. google.in for India). In some cases this involved searching for a slightly altered or simplified version in the alternative variant. For instance, GloWbE searches for “[take] the job on,” found thirty-seven hits, for example, (44). (44)

a. It was Britain’s ambassador Sir David Hannay who took the job on.

b. I doubt they’d find anyone to take the job on. c. When I took the job on, I said I wanted to bring the thunder back . . . Searches in Google return still more, suggesting that take on in this context is indeed interchangeable, at least in principle. As a general rule of thumb, if five or more valid hits could be found for an alternative variant, a given token was accepted as interchangeable. The following examples illustrate a few expressions that were deemed to be consistently “fixed” in one variant.7 (45)

a. s-genitives: Hell’s bells, Pete’s sake, the mind’s eye, no man’s land b. of genitives: kingdom of heaven, skin of one’s teeth, the law of the land, son of a bitch

(46)

a. Prepositional datives: bring tears to X’s eyes/cheeks, bring X to a halt, keep X to herself/himself/. . . , pose a danger to X, take X to court b. Ditransitives: tell X the truth

7 See Röthlisberger (2014) for an extensive list of dative idioms found in the ICE corpora.

4.4 Annotating the Constraints

75

Table 4.3 Summary of genitive variants in ICE and GloWbE corpus data. ICE

GloWbE

of

%

s

%

Total

of

%

s

%

Total

BrE CanE IrE NZE JamE

787 631 629 770 744

74.0 63.2 70.8 70.6 81.0

276 368 259 321 174

26.0 36.8 29.2 29.4 19.0

1063 999 888 1091 918

388 207 294 525 265

70.7 57.3 69.2 67.7 68.1

161 154 131 251 124

29.3 42.7 30.8 32.3 31.9

549 361 425 776 389

SinE IndE HKE PhlE

609 904 719 898

67.1 79.5 69.8 74.8

299 233 311 302

32.9 20.5 30.2 25.2

908 1137 1030 1200

195 477 320 360

61.5 68.2 64.1 65.6

122 222 179 189

38.5 31.8 35.9 34.4

317 699 499 549

Total

6691

9234

3031

2543

1533

4564

Table 4.4 Summary of dative variants in ICE and GloWbE corpus data. ICE

GloWbE

DO

%

PD

%

Total

DO

%

PD

%

Total

BrE CanE IrE NZE JamE

642 671 652 738 684

73.0 72.7 74.3 71.2 72.3

237 252 226 298 262

27.0 27.3 25.7 28.8 27.7

879 923 878 1036 946

292 298 280 321 297

65.8 67.6 69.0 70.4 68.1

152 143 126 135 139

34.2 32.4 31.0 29.6 31.9

444 441 406 456 436

SinE IndE HKE PhlE

772 612 841 668

72.8 56.2 65.9 65.6

289 476 436 350

27.2 43.8 34.1 34.4

1061 1088 1277 1018

323 306 289 335

66.6 64.6 58.4 67.3

162 168 206 163

33.4 35.4 41.6 32.7

485 474 495 498

Total

6280

9106

2741

4.3.5

2826

1394

4135

The Corpus Datasets

Summaries of the final genitives (N = 13798), datives (N = 13241), and particle placement (N = 11340) datasets are shown in Table 4.3, Table 4.4, and Table 4.5, respectively. 4.4

Annotating the Constraints

Following the data collection and filtering process, all three datasets were annotated for a number of internal linguistic constraints known or hypothesized to influence the choice of variant in each alternation. Many of these constraints, such as animacy, length, and frequency, have been shown to have significant

76

The Data

Table 4.5 Summary of particle placement variants in ICE and GloWbE corpus data. ICE

GloWbE

Joined

%

Split

%

Total

Joined

%

Split

%

Total

BrE CanE IrE NZE JamE

580 655 621 705 618

63.6 63.2 65.7 62.6 82.7

332 381 324 422 129

36.4 36.8 34.3 37.4 17.3

912 1036 945 1127 747

359 309 368 354 326

87.1 86.8 91.5 85.1 90.8

53 47 34 62 33

12.9 13.2 8.5 14.9 9.2

412 356 402 416 359

SinE IndE HKE PhlE

699 593 710 554

80.8 88.4 83.1 85.1

166 78 144 97

19.2 11.6 16.9 14.9

865 671 854 651

431 384 318 314

92.1 93.2 89.8 89.0

37 28 36 39

7.9 6.8 10.2 11.0

468 412 354 353

Total

5735

7808

3163

2073

369

3532

effects in similar ways in all three alternations (Chapter 2), suggesting that morphosyntactic variation is driven in no small degree by deep-seated cognitive processes (MacDonald, 2013). In this section we describe the annotation schemata and procedures for each linguistic feature included in our analyses. Note that for a given alternation, not every feature we annotated for was included in the final analyses. See Chapter 6.5.3 for details of which features were used. 4.4.1

Animacy

A five-level ANIMACY classification (Table 4.6) was coded for the relevant constituents in each alternation (e.g. Bresnan et al., 2007; Rosenbach, 2014; Wolk et al., 2013), which was based on the coding scheme developed by Zaenen et al. (2004). In addition to the five-level classification, a simpler binary classification was created by collapsing the collective, locative, temporal and inanimate levels into a single “inanimate” class. ANIMACY annotations for the three alternations included: • Genitives: Animacy of the possessor; Animacy of the possessum • Datives: Animacy of the recipient; Animacy of the theme • Particle placement: Animacy of the direct object.

4.4.2

Definiteness

A two-level DEFINITENESS classification was coded for the relevant constituents in each alternation (e.g. Bresnan et al., 2007; Grafmiller and Szmrecsanyi, 2018; Heller et al., 2017b; Rosenbach, 2014; Röthlisberger

4.4 Annotating the Constraints

77

Table 4.6 Annotation schema for the coding of alternations.

ANIMACY

Class

Description

Examples

Animate

Humans and higher animals or beings (not e.g. fish or bugs)

Collective

Organizations or political states/bodies when seen as having collective purpose, agenda or will, OR . . . Group of animate individuals with potential variable anaphoric reference (it/they) Places qua places, not groups of inhabitants or members, including states/empires; not referable by they Noun or adverb with time reference

Shakespeare, engineers, the horse, a sixteen-year-old girl, God the House of Lords, the church, parliament, another country

Locative

Temporal Inanimate

Nontemporal, nonlocative inanimates: concrete and abstract, all gerunds, participles, and infinitives

across the three

family, multitudes, the public, a convoy, the majority the sea, the playground, China, the Earth yesterday, last week, March, 1986, this morning the table, oxygen, other topics, drinking

et al., 2017). Definite constituents included proper nouns, personal and definite impersonal pronouns (e.g. everyone, nothing), superlatives, temporal nouns with specific reference (yesterday, tomorrow, June 10th), and nouns with definite determiners. Indefinite constituents included indefinite pronouns (e.g. someone, anything, lots), bare plural NPs (e.g. people, some elderflower cordial), and nouns with indefinite determiners. Coding of (in)definite items followed the guidelines in Garretson et al. (2004). DEFINITENESS annotations for the three alternations included: • Genitives: Definiteness of the possessor • Datives: Definiteness of the recipient; Definiteness of the theme • Particle placement: Definiteness of the direct object. 4.4.3

NP Type

The syntactic category of the relevant constituent noun phrases was coded according to the scheme in (Table 4.7). In subsequent analyses, these categories were simplified and/or some excluded. 4.4.4

Information Status (Givenness)

While there are many possible degrees of discourse accessibility or givenness that could be explored, for the present study only two levels of GIVENNESS were coded, following previous work (Bresnan and Hay, 2008; Gries, 2003;

78

The Data

Table 4.7 Annotation scheme for the coding of NP TYPE across the three alternations. Class

Description

Examples

Proper noun

Proper nouns with unique reference, including years and days of the week Common nouns and proper nouns when used as common nouns Personal pronouns, including possessives and reflexives Any definite or indefinite pronoun, including wh pronouns Bare demonstratives

Japan, Tuesday

Nominal -ing forms

give up drinking, hunting’s purpose, Give your writing a break

Common noun Personal pronoun Impersonal pronouns Demonstrative pronouns Gerunds

birds, this year, She wants to be a Shakespeare me, ours, them, it everyone, something, nobody, most this, that, these, those

Hinrichs and Szmrecsanyi, 2007; Rosenbach, 2014). A constituent is coded as “given” if its referent is mentioned at any time in the 100 words preceding the token in the discourse, or if it is a personal pronoun. Any constituent that did not refer to a speech participant and is not referred to in the preceding 100 words was coded as “new.” GIVENNESS annotations for the three alternations included: • Genitives: Givenness of the possessor • Datives: Givenness of the recipient; Givenness of the theme • Particle placement: Givenness of the direct object. 4.4.5

Constituent Length

The length or syntactic weight of the relevant constituents was coded automatically for all three alternations. Two measures of length were used for our analyses: length in orthographic words, and length in orthographic letters. LENGTH annotations for the three alternations included: • Genitives: Length (in words and letters) of the possessor; Length (in words and graphemes) of the possessum • Datives: Length (in words and letters) of the recipient; Length (in words and graphemes) of the theme • Particle placement: Length (in words and letters) of the direct object. 4.4.6

Structural Persistence/Priming

For each observation of the three alternations, we coded the variant used in the previous choice context with the 100 words preceding the target observation

4.4 Annotating the Constraints

79

(Szmrecsanyi, 2005). If there was no preceding construction in the prior 100 words, the token was coded as “none.” For spoken dialogues, persistence is coded within and across turns, and within and across speakers. The first construction in each conversation or text is automatically to be coded as “none.” For the purpose of this study we counted only preceding tokens found in genuine choice contexts. 4.4.7

Frequency Measures

To capture potential effects of lexical frequencies, we measured the frequency of a lexical item in its specific text (“thematicity”) and the frequency of that item in its variety overall. Both frequency measures were calculated for the head noun of each constituent in each of the three alternations, that is, the possessor and possessum heads in the genitive, the recipient and theme heads in the dative, and the direct object in the particle placement alternations. Frequencies were calculated as follows. Frequency (overall frequency). Recent research suggests that frequency measures obtained from larger corpora (+50 million words) more accurately predict frequency effects in various psycholinguistic tasks (Brysbaert and New, 2009), therefore overall lexical frequencies for each variety were obtained from the GloWbE corpus, which is substantially larger than the ICE corpora. For each head noun in the relevant constituents, we calculated the overall frequency of that word in the given variety in the GloWbE corpus, normalized per million words. Thematicity (localized frequency). Thematicity (Osselton, 1988) is used as a measure of how topical a given entity is in the local discourse, here construed as a corpus text. Intuitively, the more topical or thematic something is in a given text, the more often it is likely to be referred to in that text. For example, the word nucleus may be quite infrequent overall, but in a text pertaining to chemistry or quantum physics we might expect it to occur more often than the global frequency would predict. From a language processing perspective, thematicity can be construed as an additional measure of the accessibility of a given entity/word within the local discourse context. Thematicity is operationalized as the number of uses of the constituent head word in its text divided by the total number of words in that text (Hinrichs and Szmrecsanyi, 2007, 450–451). 4.4.8

Alternation-Specific Features

In addition to the general features shared by the three alternations, we annotated for several features that are unique to the respective alternations.

80

The Data

FINAL SIBILANCY (genitive alternation). There is a well-established tendency for users to avoid the s-genitive with possessors ending in a sibilant, as in President Bush’s speech (e.g. Hinrichs and Szmrecsanyi, 2007; Heller et al., 2017b). The preference for the of -genitive in such cases is thought to be a kind of repair, similar to other haplology effects (e.g. Zwicky, 1987), used to avoid having to produce two adjacent sibilant sounds. We coded for the presence or absence of a sibilant on the final word of the possessor phrase. Sibilant sounds included [s], [z], [S], [Z], [tS], [dZ]. See Heller (2018, 71–72) for further details. SEMANTIC RELATION (genitive alternation). The semantic relation between the possessor and possessum also has an effect on the genitive alternation, with “prototypical” relations favoring the s-genitive (Rosenbach, 2014). Here we follow the binary “prototypical” versus “non-prototypical” coding scheme included in recent variationist studies (e.g. Grafmiller, 2014; Heller et al., 2017b). Prototypical tokens included those expressing a body-part (a moth’s wing), kinship (Mary’s brother), ownership (the guard’s rifle), or part-whole (the door’s hinges) relation (Rosenbach, 2002). All other tokens were coded as nonprototypical. See Heller (2018, 75–77) for further details. Post-modifying DIRECTIONAL PP (particle placement). The presence of a postmodifying directional PP, as in (47), has been shown to influence particle placement in Present Day English (Fraser, 1976; Gries, 2003; RodríguezPuente, 2016). (47)

Well, as we have said, applying VAT on government contracts is taking money away from one government pocket and putting it in another.

According to Gries (2003), the split variant foregrounds the spatial meaning of the verb, thus the presence of additional material spelling out the direction or endpoint of spatial movement is to be expected with this variant. Thus we included a binary factor indicating the presence of a directional PP modifier following the verb phrase. Semantic COMPOSITIONALITY (particle placement). Prior work suggests that particle verbs with idiomatic (i.e. noncompositional) meanings tend to favor the continuous variant (Fraser, 1976; Quirk et al., 1985; Gries, 2003; Lohse et al., 2004). To annotate for semantic compositionality, we followed the heuristic of Lohse et al. (2004, 244–246), who employed two entailment tests to determine semantic compositionality. If [X V (P) NP (P)] entails both [X V NP] (verb entailment) and [NP PredV P] (particle entailment) – where PredV represents a predication verb of the type be, become, come, go, stay – then the token was coded as “compositional.” If either entailment test failed, the token was coded as “noncompositional.” Semantic annotation was conducted

4.5 Summary

81

by the second author, after which a random sample of 200 tokens was coded independently by a trained research assistant and the ratings were compared. Agreement between coders was reasonably good given the task (90 percent, Cohen’s κ = .75). SURPRISAL (particle placement). As an additional measure of verb-particle dependency, we calculated the SURPRISAL of the verb and the particle as components of a combined particle verb, where surprisal is defined in information-theoretic terms (Shannon, 1948) as the log inverse of the conditional probability of an item given a particular context. Informally, the less predictable a word is in a given context, the more surprising or informative it is. For a given verb-particle pair, we calculated the surprisal of the particle given the verb (SURPRISAL.P), and the surprisal of the verb given the particle (SURPRISAL.V). The formula below illustrates how we calculate the surprisal of the particle up given the verb pick (SURPRISAL.P). Surprisal(up|pick) = − log2 P(up|pick) = − log2

Freq(pickup) Freq(pick)

We interpret surprisal as a measure of the relative syntactic/semantic (in)dependence of the verb and particle. From an information-theoretic perspective, we associate higher surprisal (lower predictability) with greater independence, since more informative verbs/particles are those that occur more often in contexts without their corresponding verb or particle. For instance, particle verbs with especially low SURPRISAL.P scores are those for which, once we know the verb, we can predict that a particle will follow with a high degree of confidence. We hypothesize such cases will strongly favor the continuous variant. 4.5

Summary

This chapter presented a brief discussion of the main data types in variationist linguistics, with an eye towards the role that large-scale corpora can play to complement and extend traditional variationist research. Next, we presented the corpus datasets of the three alternations collected for the present study, and discussed the relative advantages and disadvantages of the particular ICE and GloWbE corpora. We argue that despite some potential limitations of these corpora, they nonetheless offer some of the best sources available for the kind of parallel comparisons we are trying to make here. Finally, the details of data collection and annotation were described in depth in the final sections. In the next two chapters we move on to present and discuss the statistical analysis of these corpus-derived datasets, and follow up with an experimental study aimed at corroborating our corpus findings in Chapter 7.

5

Alternation-by-Alternation Analysis

With a view towards qualitative generalization, this chapter investigates corpus data to analyze the three alternations subject to study in this book using a battery of state-of-the art analysis techniques (in addition to customary descriptive statistics, conditional random forest modeling and mixed-effects logistic regression analysis). Analysis shows that, first, probabilistic grammars on the whole are stable in a cross-variety perspective. Specifically, we very rarely see reversals in effect directions. Constraints tend to have the same qualitative effect across varieties, which points to a fairly stable “common core” (Quirk et al., 1985, 33) shared by the grammar of global English. However, there are interesting quantitative differences in effect sizes: the strength with which particular constraints favor or disfavor particular variants differs across varieties of English. Second, our results find typological clustering of varieties. Corpora from communities where English is mostly acquired as a first language (Inner Circle) tend to have parallel patterns of variation distinct from the (also parallel) patterns of variation found in corpora from communities where English is mostly or was historically learned as a second language (Outer Circle). Third, this chapter also shows that the three alternations under investigation differ in how amenable they are to “probabilistic indigenization,” which we characterize as the process whereby numerous social, cognitive, and/or linguistic forces shape the development of regionally specific probabilistic grammars. 5.1

Methods

This chapter adopts an alternation-centric (or variable-centric) perspective. Based on corpus data from the Corpus of Global Web-based English (GloWbE) and the International Corpus of English (ICE) (see Chapter 4 for discussion), we will investigate the three alternations subject to study one by one, for the sake of learning more about how variant selection differs across varieties of English. The analysis methods that we will utilize in each case are the following: Variant Rates.Variant rates (also known as relative frequencies in corpus linguistics) are percentage figures that reveal how often language users 82

5.1 Methods

83

select a certain variant vis-à-vis all other variants. Variant rates can be established by simple counting and do not – due to the lack of multivariate control – speak to the issue of probabilistic choice-making. For example, we know that animate possessors favor the s-genitive rather than the of -genitive (the president’s speech, but the consequences of inflation) (see Rosenbach, 2005). Variant rates of the s-genitive must therefore not be overinterpreted: it may be that the s-genitive is particularly frequent in a particular lect or sample because animate NPs are also particularly frequent in that lect or sample, not because the underlying probabilistic grammar is different (see Szmrecsanyi, 2016 for discussion). Nevertheless, a proper discussion of variant rates is often considered a good stepping stone towards a responsible variationist/probabilistic analysis, if not actually a crucial component of a responsible analysis. Variable importance.What are the most important constraints on variation? Are some constraints more or less important in particular varieties of English? To address these questions, we will use Conditional Random Forest (CRF) modeling (see Tagliamonte and Baayen, 2012 for an accessible introduction). CRF can be used to rank individual constraints per variety and per alternation according to the constraints’ overall explanatory importance. CRF models are robust to problems of predictor correlation, as their predictions are derived from ensembles of conditional inference trees grown on randomly sampled subsets of the data (Breiman, 2001; Strobl et al., 2009). Each tree is built on a random subset of predictors whose explanatory power is assessed through random permutation of their values, which breaks the predictor’s association with the outcome. A “permutation variable importance” measure for each predictor can then be calculated by measuring the decrease in accuracy of the model with the permuted predictor compared to the accuracy of the model with the nonpermuted predictor, and used to rank predictors. We use the cforest() function in R’s party package (Hothorn et al., 2006; Strobl et al., 2008). Effect directions and effect sizes.To determine the extent to which effect directions and effect sizes are stable (i.e. invariant) or unstable (i.e. fluctuating) across varieties of English, we turn to mixed-effects binary logistic regression analysis (Gelman and Hill, 2007; Zuur et al., 2009) as implemented in the lme4 package in R (R Core Team, 2014; Bates et al., 2015). As the workhorse multivariate analysis tool in variation studies, the technique estimates models that predict binary linguistic outcomes based on fixed effects (including interaction effects between fixed effects) as well as random (i.e. nonrepeatable) effects. Fitting and digesting mixed-effects regression models

84

Alternation-by-Alternation Analysis

involving random effects and interactions based on large datasets like ours (i.e. with a large number of annotated variables and observations) is not a trivial task, neither for the model fitter nor for the reader, as it produces an overwhelming number of data tables for analysis. In what follows we therefore take the liberty to keep the models we report fairly simple. This means in particular that we will (i) restrict attention to those interaction effects previously reported as significant in the literature on the datasets subject to study here (specifically Grafmiller and Szmrecsanyi, 2018; Heller, 2018; Heller et al., 2017b; Röthlisberger, 2018b; Röthlisberger et al., 2017), (ii) eschew centering and standardizing constraints for the sake of enhancing interpretability (even though this may incur some collinearity), and (iii) refrain from substantial model trimming. In short, the models we report in this section draw inspiration, in terms of model design, from more sophisticated regression models published elsewhere while abstaining from their complexity whenever possible. We provide more information about the corpus-derived datasets on which our analysis is based in Chapter 4.

5.2

A Comparative Analysis of the Genitive Alternation

We begin by conducting a comparative analysis of the probabilistic grammar of the genitive alternation, as in (1) (reprinted as (48) below). (48)

The genitive alternation in English. a. Two other journalists who wrote a book criticising [the president] possessor ’s [brother]possessum were ordered to pay £6.3 million in fines (GloWbE AU B vexnews.com) (the s-genitive) b. Can you imagine a couple of years after WW2 the allies permitting [the brother]possessum of [the president]possessor bankrupting the central bank through embezzlement and getting away with it? (GloWbE GB G guardian.co.uk) (the of -genitive)

The genitive alternation is discussed in detail in Section 2.5.1 and in Chapter 4. To briefly recap the definition of the variable context, included as interchangeable in the dataset are all ’s and of -constructions with two NP constituents that did not fall into one of the following categories: appositive genitives, classifying genitives, double genitives, idiomatic/fixed genitives, partitive genitives, and genitives with indefinite possessums. The dataset further does not include of -genitives that do not contain a definite possessum. The dataset spans N = 13,798 tokens.

5.2 A Comparative Analysis of the Genitive Alternation

85

Table 5.1 Variant rates of genitive constructions across varieties of English. Upper half of the table: Inner Circle varieties of English; lower half of the table: Outer Circle varieties of English. of -genitive

s-genitive

total

BrE CanE IrE NZE

1175 (72.9%) 838 (61.6% ) 923 (70.3%) 1295 (69.4%)

437 (27.1%) 522 (38.4%) 390 (29.7%) 572 (30.6%)

1612 1360 1313 1867

HKE IndE JamE PhlE SgE

1039 (68.0%) 1381 (75.2%) 1009 (77.2%) 1258 (71.9%) 804 (65.6%)

490 (32.0%) 455 (24.8%) 298 (22.8%) 491 (28.1%) 421 (34.4%)

1529 1836 1307 1749 1225

total

9722

4076

13798

Variety

5.2.1

Variant Rates

Table 5.1 shows genitive variant rates as a function of regional variety of English. No clear-cut patterns, regional or otherwise, are identifiable, except that the of -genitive is consistently the more frequent variant, with a range of between 62 and 77 percent. The varieties attesting the highest s-genitive rates are CanE, SgE, and HKE; those that attest the lowest s-genitive rates are JamE, IndE, and BrE. In all, the s-genitive is moderately but significantly (chi-square = 15.1, d.f. = 1, p < 0.001) more widely used in Inner Circle varieties (mean rate: 31.2 percent) than in Outer Circle varieties (mean rate: 28.2 percent). This small differential may result from the well-known interlanguage universal that language learners avoid inflectional marking and prefer analyticity instead (see e.g. Klein and Perdue, 1997, 311), giving rise to the more analytic of -genitive.

5.2.2

Ranking the Importance of Constraints

The following section explores the probabilistic conditioning of genitive variation. The genitive dataset was annotated for a range of mostly languageinternal and some language-external predictors. To recap (see Chapter 4), we investigate the following constraints: PORANIMACYBIN: Possessor animacy (animate versus inanimate) PORLENGTH: Possessor length in words PUMLENGTH: Possessum length in words PORNPEXPRTYPEBIN: Possessor NP expression type (common noun versus other) • PORFINALSIBILANCY: Final sibilancy in possessor (present versus absent) • • • •

86

Alternation-by-Alternation Analysis

Figure 5.1 Conditional Random Forest permutation variable importance ranking for the pooled genitive dataset. C = 0.94.

• PREVIOUSCHOICE: of versus s versus none • SEMANTICRELATIONBIN: Semantic relation (prototypical versus nonprototypical) • PORHEADFREQ: Possessor head frequency • VARIETY_OF_E: BrE versus CanE versus HKE versus IndE versus IrE versus JamE versus NZE versus PhlE versus SgE • CIRCLE: Inner Circle (BrE, CanE, IrE, NZE) versus Outer Circle (HKE, IndE, JamE, PhlE, SgE) • GENRECOARSE: printed versus blog versus dialogue versus general monologue versus non-printed We begin by submitting these constraints to CRF analysis to assess variable importance. Figure 5.1 shows the variable importance ranking, pooling data from all varieties (but including VARIETY_OF_E and CIRCLE as constraints on variation). We observe that by far the most important constraint is whether or not the possessor is an animate entity or not. This finding is certainly not surprising, given the literature (see e.g. Rosenbach, 2014). Possessum and possessor weight, which jointly operationalize the principle of end-weight (see e.g. Behaghel, 1909) and/or Easy First (MacDonald, 2013), are the second and third-ranked constraints. The fourth-ranked constraint is Possessor NP expression type (PORNPEXPRTYPEBIN). Next we find a group

5.2 A Comparative Analysis of the Genitive Alternation

87

Figure 5.2 Conditional Random Forest permutation variable importance ranking of constraints on the genitive alternation by variety of English.

of five constraints (PORFINALSIBILANCY, PREVIOUSCHOICE, PORHEADFREQ, GENRECOARSE, VARIETY_OF_E) that do matter for variant choice but not that much. SEMANTICRELATIONBIN (are we dealing with a prototypical genitive relation or not?) and CIRCLE (are we dealing with an Inner Circle or Outer Circle variety?), finally, have fairly limited explanatory power. To obtain a clearer view on how variable importance differs across varieties of English, Figure 5.2 reports rankings from nine separate CRF models, one for each variety of English under study. The four Inner Circle varieties (BrE, CanE, IrE, NZE) are a fairly homogeneous group: Possessor animacy (PORANIMACYBIN) is the most important constraint throughout, which along with PORLENGTH, PUMLENGTH, and PORNPEXPRTYPEBIN comprises the set of the four most important constraints for this group. The Outer Circle

88

Alternation-by-Alternation Analysis

varieties are more heterogeneous. As in the Inner Circle varieties, PORANIMACY B IN is the highest-ranked constraint in SgE, while P UM L ENGTH is the highest-ranked constraint in HKE, JamE, and PhlE, but PORNPEXPRTYPEBIN in IndE. Finally, genre distinctions (GENRECOARSE) play some role in most Outer Circle varieties but not at all in JamE. We conclude from the comparison in Figure 5.2 that the probabilistic grammars regulating genitive choice do differ to some extent as a function of regional variety. 5.2.3

Effect Directions, Effect sizes, and Interaction Effects

We now turn to binary logistic regression analysis to investigate effect directions, effect sizes, and – crucially – interaction effects between languageinternal constraints and regional variety. We do this for the sake of gauging the scope of probabilistic grammar differences between varieties of English: what is the extent to which language internal constraints have different effect sizes and/or directions in different varieties of English? For expository reasons, the model that we will report here is a simplified version of the more complex model presented in Heller (2018, chapter 5; see also Heller et al., 2017b). That being said, we stress that the qualitative conclusions are essentially the same. For simplicity, the language-internal constraints considered are the five most important predictors according to the overall CRF ranking in Figure 5.1: PORANIMACYBIN, PORLENGTH, PUMLENGTH, PORNPEXPRTYPEBIN, and PORFINALSIBILANCY. In addition, we include VARIETY_OF_E as a languageexternal factor (contrast coding, default variety: BrE), which is allowed to interact with PORANIMACYBIN and PORFINALSIBILANCY (drawing inspiration from the more sophisticated model reported in Heller (2018, 120), which demonstrates the existence of these interactions). The model reported here also contains two random effects (intercept adjustments): FileID_pruned (to cover speaker/writer idiosyncracies), and PumHeadLemma_pruned (to control for lexical effects). The model is reported in Table 5.2. The language-internal constraints all have the theoretically expected directions as main effects: the positive coefficient associated with the presence of an animate possessor indicates that animate possessors increase the odds for the s-genitive, compared to inanimate possessors. The negative coefficient associated with PORLENGTH means that increasing length of the possessor decreases the odds for the s-genitive, while increasing length of the possessum increases the odds for the s-genitive – in conjunction, this is the principle of end weight in action. Common noun possessors disfavor the s-genitive, vis-à-vis other types of possessors; and the presence of a final sibilant in the possessor phrase likewise disfavors the sgenitive. Again, all of these effect directions are predicted by the literature (see e.g. Rosenbach, 2014). As to VARIETY_OF_E, this language-external factor has the following significant main effect(s): compared to BrE (the default

0.31 0.34 0.30 0.31

Significance codes: “***” – 0.001; “**” – 0.01; “*” – 0.05; “.” – 0.1

−0.74 −1.35 −0.53 −0.75

Interactions between VARIETY_OF_E and PORFINALSIBILANCY PORFINALSIBILANCY=present:VARIETY_OF_E=CanE PORFINALSIBILANCY=present:VARIETY_OF_E=HKE PORFINALSIBILANCY=present:VARIETY_OF_E=IndE PORFINALSIBILANCY=present:VARIETY_OF_E=SgE

0.23 0.21

0.14 0.13 0.13 0.13 0.14

0.45 0.74 0.51 0.33 0.57 −0.71 −1.06

0.20

0.06

0.02

0.16 0.03

0.16

Std. Error

−0.95

−0.95

Interactions between VARIETY_OF_E and PORANIMACYBIN PorAnimacyBin=inanimate:VARIETY_OF_E=HKE PorAnimacyBin=inanimate:VARIETY_OF_E=PhlE

PORFINALSIBILANCY (default: final sibilant absent) final sibilant present VARIETY_OF_E (default: BrE) CanE HKE NZE PhlE SgE

PORNPEXPRTYPEBIN (default: other) common noun

0.58

2.06 −0.67

PORANIMACYBIN (default: animate) inanimate PORLENGTH

PUMLENGTH

−0.99

(Intercept)

Coefficient

−2.37 −3.99 −1.76 −2.42

−3.06 −5.04

3.31 5.83 3.93 2.48 4.06

−4.80

−16.60

25.72

12.87 −24.91

−6.17

z value

Table 5.2 The genitive alternation: fixed effect coefficients in mixed-effects logistic regression analysis. Predicted odds are for the s-genitive. Percent correctly predicted: 83.7 percent (baseline: 70.5 percent). C = 0.90. κ = 22.9. Only significant coefficients are shown.

* *** . *

** ***

*** *** *** * ***

***

***

***

*** ***

***

p

90

Alternation-by-Alternation Analysis

variety in the model), CanE, HKE, NZE, PhlE, and SgE all significantly favor the s-genitive. This pattern is consistent with the variant rates shown in Table 5.1. Leaving the main effects aside, we focus on the interaction effects, which indicate probabilistic grammar differences from a comparative perspective. Heller (2018, 120) reports interaction effects between variety type and both possessor animacy and final sibilancy, which is why we included corresponding interaction effects in the model reported in Table 5.2. For PORANIMACYBIN, according to Table 5.2 the main effect has a coefficient of 2.06. This is the effect that the presence of an animate possessor has on dative choice in BrE, the default variety in the model. Log odds of 2.06 can be converted into an odds ratio of e2.06 = 7.85. Therefore, in BrE if the possessor is animate instead of nonanimate, the odds for the s-genitive increase 7.85-fold. For PORFINALSIBILANCY , the main effect (in BrE) is such that the presence of a final sibilant in the possessor decreases the odds for the s-genitive by a factor of e−0.95 = 0.39 (i.e. by 61 percent). Both of these main effects are in line with the literature. The interaction effects in the lower part of Table 5.2 then indicate how the effect of PORANIMACYBIN and PORFINALSIBILANCY differs as a function of regional variety. For PORANIMACYBIN, two varieties show significant differences to BrE: HKE and PhlE. In HKE, the coefficient associated with the interaction effect is -0.71. The negative sign indicates that the s-genitivefavoring effect of animate possessors is weaker in HKE than in BrE. Specifically, in HKE an animate possessor increases the odds for an s-genitive by a factor of only e2.06−0.71 = 3.86. The factor in PhlE is e2.06−1.06 = 2.72, also significantly lower than in BrE. As to PORFINALSIBILANCY, the negative coefficients (on top of the negative main effect) associated with the interaction terms indicate that in CanE, HKE, and SgE, the s-genitive-disfavoring effect of a final sibilant in the possessor is significantly stronger than in BrE. The aforementioned probabilistic grammar differences relating to possessor animacy and final sibilancy are visually depicted in the partial effects plots in Figures 5.3 and 5.4. The plots summarize information from the regression model by systematically altering the values of the constraints under scrutiny while holding the values of all other predictors in the model constant at their default levels (see Fox, 2003). In other words, the plots illustrate probabilistic grammar differences by showing how users of different varieties of English react differently – as measured by variant selection probabilities – to what we may casually call stimuli here, such as animate possessors or possessors ending in a final sibilant. Figure 5.3 plots the strength of the possessor animacy effect: for each variety, the larger the distance between the white dots (representing the probability of selecting the s-genitive when the possessor is animate and all other predictors are at their reference level) and the black dots (representing the

5.2 A Comparative Analysis of the Genitive Alternation

91

Figure 5.3 Partial effects plot of the interaction of VARIETY_OF_E with PORANIMACYBIN in the genitive regression model. Predicted probabilities (vertical axis, expressed in percent) are for the sgenitive. Vertical distance between dots is proportional to effect size. The plotted probabilities obtain when all other categorical constraints are set at their default levels, or at 0 in the case of numerical constraints. British English (shaded) serves as the reference variety in the underlying regression model. Varieties to the left are Inner Circle varieties, varieties to the right are Outer Circle varieties.

probability of selecting the s-genitive when the possessor is inanimate and all other predictors are at their reference level), the stronger the possessor animacy effect. In short, the effect size is reflected in the vertical distance between the dots. Observe that in comparison to BrE, the effect of possessor animacy is weaker in HKE and PhlE; these are the two varieties which the regression model in Table 5.2 has singled out as having significantly different effect sizes compared two BrE. Beyond these two varieties, we see a general tendency in Figure 5.3 that in Outer Circle varieties – with the exception of SgE, which is more similar to the Inner Circle varieties in the sample – the animacy effect tends to be smaller than in Inner Circle varieties (on this point see also Heller et al., 2017b, 18–19).

92

Alternation-by-Alternation Analysis

Figure 5.4 Partial effects plot of the interaction of VARIETY_OF_E with PORFINALSIBILANCY in the genitive alternation. Predicted probabilities (vertical axis, expressed in percent) are for the sgenitive. Vertical distance between dots is proportional to effect size. The plotted probabilities obtain when all other categorical constraints are set at their default levels, or at 0 in the case of numerical constraints. British English (shaded) serves as the reference variety in the underlying regression model. Varieties to the left are Inner Circle varieties; varieties to the right are Outer Circle varieties.

Moving on to the effect that final sibilancy has on genitive choice, Figure 5.4 shows how the difference between the “final sibilant present” condition (visually represented by white dots) and “final sibilant absent” condition visually represented by black dots) is particularly large for CanE, HKE, and SgE, those varieties for which the regression model in Table 5.2 detects significant differences from BrE. 5.3

A Comparative Analysis of the Dative Alternation

We move on to a comparative analysis of the dative alternation, as in (2) (reprinted as (49) below).

5.3 A Comparative Analysis of the Dative Alternation

93

Table 5.3 Variant rates of dative constructions across varieties of English. Upper half of the table: Inner Circle varieties of English; lower half of the table: Outer Circle varieties of English. Variety

ditransitive dative

prepositional dative

total

BrE CanE IrE NZE

932 (72.9%) 968 (71.3% ) 924 (72.8%) 1056 (71.1%)

386 (29.3%) 389 (28.7%) 346 (27.2%) 429 (28.9%)

1318 1357 1270 1485

HKE IndE JamE PhlE SgE

1129 (63.9%) 912 (58.8%) 977 (71.1%) 995 (66.1%) 1094(70.9%)

639 (36.1%) 638 (41.2%) 398 (28.9%) 510 (33.9%) 449 (29.1%)

1768 1550 1375 1505 1543

total

8987

4184

13171

(49) The dative alternation in English. a. A victim will be asked to giveverb [the police]recipient [a statement] theme explaining what has happened. (GloWbE CA G slsedmonton.com) (the ditransitive dative) b. Neither of them gaveverb [a statement]theme to [the police]recipient . (GloWbE JM G jamaicaobserver.com) (the prepositional dative) We refer the reader to Section 2.5.2 for an in-depth discussion of the dative alternation. To summarize the definition of the variable context (see Chapter 4 for details): dative constructions were extracted from the corpus material using a list of dative verbs adapted from previous literature. Constructions that are not interchangeable were then weeded out manually, based essentially on the guidelines used in Bresnan et al. (2007); for example, beneficiary constructions (as in we get them typed photocopies) were discarded. The dataset spans N = 13, 171 tokens. 5.3.1

Variant Rates

The pattern that emerges from Table 5.3 is that variant rates are curiously similar in the four Inner Circle varieties of English in the upper half of the table: Inner Circle varieties show a distribution of roughly 70 percent : 30 percent in favor of the ditransitive construction. Variant rates are far less homogeneous in Outer Circle varieties (lower half of the table), though the generalization is that with a mean distribution of 66 percent : 34 percent in favor the ditransitive dative construction, the ditransitive variant is used significantly (chi-square

94

Alternation-by-Alternation Analysis

= 44.2, d.f. = 1, p < 0.001) less often in Outer Circle varieties (in particular HKE, IndE, and PhlE) than in Inner Circle varieties. Note here that the prepositional dative variant includes the explicit prepositional marker to, and thus we may speculate that the relative popularity of the prepositional dative construction in Outer Circle varieties may have to do with the well-known tendency toward analyticity in creoles, second languages, and interlanguages (see e.g. Seuren and Wekker, 1986; Wekker, 1996a,b).

5.3.2

Ranking the Importance of Constraints

Which constraints are important when choosing dative variants? The languageinternal and language-external predictors for which the dataset was annotated are discussed in detail in Chapter 4. The following list is a summary of the constraints that we will investigate: • LOGWEIGHTRATIO: Log weight ratio between recipient and theme (weight measured in number of orthographically transcribed characters) • RECPRON: Recipient pronominality (pronominal versus nonpronominal) • THEMEBINCOMPLEXITY: Theme complexity (complex versus simple) • THEMEHEADFREQ: Theme head frequency • THEMEPRON: Theme pronominality (pronominal versus nonpronominal) • THEMEDEFINITENESS: Theme definiteness (definite versus indefinite) • RECGIVENNESS: Recipient givenness (given versus new) • RECHEADFREQ: Recipient head frequency • VARIETY_OF_E: BrE versus CanE versus HKE versus IndE vs. IrE versus JamE versus NZE versus PhlE versus SgE • CIRCLE: Inner Circle (BrE, CanE, IrE, NZE) versus Outer Circle (HKE, IndE, JamE, PhlE, SgE) • GENRECOARSE: printed versus blog versus dialogue versus general monologue versus non-printed We begin by submitting these constraints to CRF analysis. Figure 5.5 shows the variable importance ranking, pooling data from all varieties (but including VARIETY_OF_E and CIRCLE as language-external factors). The weight ratio between the recipient and the theme (LOGWEIGHTRATIO) – an operationalization of the principle of end weight (Behaghel, 1909) – turns out as the by far most important constraint, surprisingly powerful in fact against the backdrop of similar analyses in the dative alternation literature (see e.g. Szmrecsanyi et al., 2017). Recipient pronominality (RECPRON) is the second most important constraint. Next we find a series of constraints (THEMEDEFINITENESS, THEMEHEADFREQ, THEMEPRON, THEMEBINCOMPLEXITY,RECGIVENNESS, RECHEADFREQ) which do make a fairly smallish contribution to predicting dative outcomes. The language-external

5.3 A Comparative Analysis of the Dative Alternation

95

Figure 5.5 Conditional Random Forest permutation variable importance ranking for the pooled dative dataset. C = 0.97.

constraints (VARIETY_OF_E, GENRECOARSE, CIRCLE) are per se fairly insignificant, according to CRF analysis. How does variable importance differ across varieties of English? Figure 5.6 provides some answers. Once again, the four Inner Circle varieties (BrE, CanE, IrE, NZE) are fairly homogeneous: LOGWEIGHTRATIO is consistently ranked as the most important constraint, and RECPRON as the second most important constraint. Genre differences (GENRECOARSE) do not play an important role in Inner Circle varieties. More often than not LOGWEIGHTRATIO is also the highest-ranked constraint in the Outer Circle varieties under study, with the exception of IndE where RECPRON is substantially more important than LOGWEIGHTRATIO. Another difference among the Outer Circle varieties concerns GENRECOARSE: as in the Inner Circle varieties, genre differences are negligible in JamE, PhlE, and SgE, though they turn out to be a bit more important in HKE and IndE. 5.3.3

Effect Directions, Effect Sizes, and Interaction Effects

We move on to binary logistic regression analysis to investigate effect directions, effect sizes, and interaction effects between language-internal constraints and regional variety. The model that we report here is a simplified approximation of the more sophisticated model presented in Röthlisberger (2018b,

96

Alternation-by-Alternation Analysis

Figure 5.6 Conditional Random Forest permutation variable importance ranking of constraints on the dative alternation by variety of English.

chapter 5; see also Röthlisberger et al., 2017). The simplified model is summarized in Table 5.4.1 In its fixed-effect structure, the model includes the five most important constraints according to the overall CRF ranking in Figure 5.5: LOGWEIGHT R ATIO , R EC P RON , T HEME D EFINITENESS , T HEME H EAD F REQ , and T HE ME P RON . Also included is the language-external factor VARIETY _ OF _E, as 1 Normally we would be concerned about the relatively high condition number κ indicating

“potentially harmful” (Baayen, 2008, 200) collinearity – remember that for the sake of maximizing readability and interpretability we did not take anti-collinearity measures such as centring/standardizing predictors etc. But because the model in Table 5.4 agrees qualitatively with the models reported in Röthlisberger (2018b, chapter 5) and Röthlisberger et al. (2017), we argue that the collinearity is not actually harmful and in the case at hand so feel justified in proceeding to interpret the model nonetheless.

5.3 A Comparative Analysis of the Dative Alternation

97

Table 5.4 The dative alternation: fixed effect coefficients in mixed-effects logistic regression analysis. Predicted odds are for the prepositional dative construction. Percent correctly predicted: 88.8 percent (baseline: 68.2 percent). C = 0.95. κ = 42.1. Only significant coefficients are shown. Coefficient

Std. Error

z value

p

(Intercept)

1.26

0.36

3.44

***

LOG W EIGHT R ATIO

1.49

0.13

11.71

***

−1.75

0.23

−7.57

***

0.66

0.07

8.99

***

particles > datives

0 = 0.62 0 = 0.75

particles > datives > genitives genitives > datives > particles

0 = 0.79

genitives > particles > datives

0 = 0.73

genitives > datives > particles

dative alternation is least stable (because it has the lowest mean coefficient); the particle placement alternation takes the middle road. Inspecting similarity coefficients per line of evidence, we note that similarity coefficients across lines of evidence are more often than not similarly sized except in the genitive alternation, which yields a comparatively low coefficient for the second line of evidence (effect strength). The value in the bottom-right corner of Table 6.5 is the CORE GRAMMAR SCORE 0: it is the mean similarity coefficient across lines of evidence and across all alternations subject to study. 0 thus abstracts away from particular alternations and lines of evidence: The higher 0, the more similar the varieties. The dataset studied here (three grammatical alternations × nine varieties of English;) yields a core grammar score of 0 = 0.76. Relying on customary schemes for interpreting (correlation) coefficients (e.g. De Vaus, 2002, 272), we thus see “very strong” similarities between the varieties under study. The coefficients in Table 6.5 are based on an analysis of all available data – written and spoken, including both Inner Circle and Outer Circle varieties. But what would happen to the similarity coefficients and core grammar scores that we calculated above if we restricted attention to particular subsets of the data? To address this question, Table 6.6 reports core grammar scores 0 for a number of sub-datasets, along with hierarchies of stability as far as individual alternations are concerned (the idea being that the higher similarity coefficients are on average, the stabler the alternation in question). The largest core grammar score is obtained when attention is restricted to Inner Circle varieties (0 = 0.79), indicating that these varieties are particularly homogeneous and similar to each other. Outer Circle varieties are substantially less homogeneous, with a core grammar score of 0 = 0.73. As to the difference that medium makes, written varieties are clearly more homogeneous

6.5 Evaluating Coherence

123

(0 = 0.75) than spoken varieties (0 = 0.62). (Note that these and the other findings will be interpreted in more detail in Section 6.7.) Turning now to differences between alternations (consider the rightmost column in Table 6.6), we have seen before that when we investigate all available data, the hierarchy of stability is genitives > particles > datives (meaning that the way language users choose between genitive variants is most similar across varieties, while the conditioning of dative choices is least similar). The genitive alternation turns out to be most stable also when we restrict attention to the various sub-datasets in Table 6.6, with the exception of the spoken subdataset, where the genitive alternation is actually the least stable one. This is primarily due to very low similarity coefficients for the second and third line of evidence in spoken materials, meaning that effect strengths and the rankings of constraints on genitive variation are rather dissimilar across varieties.

6.5

Evaluating Coherence

Lectal coherence and cohesion is currently a hot topic in the variationist (socio)linguistics community (see e.g. a recent double special issue of the journal Lingua on Coherence, covariation and bricolage (Guy and Hinskens, 2016), and the papers in Beaman and Guy, 2022). The point is that linguists customarily define varieties and lects by certain features, but how unified and coherent are varieties and lects really? Do linguistic features systematically define varieties and lects, and relationships between those lects and varieties? Guy (2013), for example, investigates whether speakers consistently use stigmatized or prestige variants, but finds that it is not easy to demonstrate correlations in the behavior of variables, even if they are generally thought to vary along the same social dimension. This is contrary to expectations: the null hypothesis – informed by the “orderly heterogeneity” postulated in Weinreich et al. (1968) – is that “the orderly variables that define the community should collectively behave in parallel: variants (or rates of use of variants) that index a given style, status, or a social characteristic should co-occur” (Guy and Hinskens, 2016). In short, the study of lectal (non)coherence has potentially important implications for sociolinguistic theorizing. Against this backdrop, this section will show how VADIS can contribute to the study of lectal coherence. The point is that VADIS can be tweaked to yield DISTANCE-BASED COHERENCE (henceforth: DBC) measurements. The idea behind DBC is that coherence between sub-datasets can be operationalized as the extent to which these sub-datasets yield similar distance relations between lects – and following the VADIS philosophy, these distance relations are a function of the extent to which the conditioning of variation is dissimilar. In this section, we investigate three different aspects of DBC (see also Szmrecsanyi, 2022):

124

Distances, Similarities, and Coherence

(i) DBCalternation is defined as being proportional to the extent to which the probabilistic conditioning of different grammatical alternations predicts similar linguistic distances between varieties. (ii) DBCmedium is defined as being proportional to the extent to which measurements in spoken language materials and in written language materials predict similar linguistic distances between varieties. (iii) DBCevidence is defined as being proportional to the extent to which different lines of evidence (see (51)) predict similar linguistic distances between varieties. To make this more concrete, these definitions translate into the following research questions: (i) DBCalternation : If two varieties (say, British English and Canadian English) turn out to be similar when we investigate a particular alternation (e.g. the dative alternation), will those varieties also turn out to be similar when we probe another alternation (e.g. the particle placement alternation)? (ii) DBCmedium : If two varieties turn out to be similar when we restrict attention to grammatical choice-making in spoken production, will those varieties also turn out to be similar when we restrict attention to choicemaking in written language production? (iii) DBCevidence : If two varieties turn out to be similar when we investigate a particular line of evidence (e.g. constraint significance), will those varieties also turn out to be similar when we probe another line of evidence (e.g. constraint ranking)? It is obvious, then, that the research questions about DBCalternation and DBCmedium are of theoretical significance, while the research question about DBCevidence is rather methodological in nature. To calculate the various DBC measures, we take the pairwise distance values calculated in steps 3, 4, and 6 of the VADIS pipeline (see Section 6.3) and arrange these values in so-called distance matrices. Distance matrices are the customary input in classical dialectometry (Séguy, 1971; Goebl, 1982; Nerbonne et al., 1999; Szmrecsanyi, 2013a) and function essentially like distance tables in road atlases, which indicate geographic distances between locations. Let us illustrate: for each alternation in our study and for each of the three lines of evidence, we create one distance matrix. As we are exploring n = 9 varieties of English, this yields n × (n − 1)/2 = 9 × 8/2 = 36 unique variety pairings per alternation and line of evidence. Figure 6.1 exemplifies by displaying the distance matrix for the third line of evidence (constraint ranking) in the particle placement alternation. All distances are scaled between 0 (no distance) and 1 (maximum distance). Consider

6.5 Evaluating Coherence

125

Figure 6.1 Variation-Based Distance and Similarity Modeling (VADIS) distance matrix for the third line of evidence in the particle placement alternation (all data included, eight constraints considered). Scores range between 0 (maximum similarity) and 1 (maximum distance).

now, for example, the pairing between BrE and NZE, which is associated with a comparatively small distance value of 0.095. Thus, BrE and NZE are very similar in terms of the constraint ranking in the particle placement alternation. By contrast, the distance between BrE and IndE is 0.548, which is considerably larger. Note that distance matrices as in Figure 6.1 may also be fused at various levels. Initially, we obtain three different distance matrices per alternation (one per line of evidence). For one thing, line-of-evidence-specific distance matrices may be fused at the level of individual alternations, thus arriving at line-merged but alternation-specific distance matrices, which leaves us with one distance matrix per alternation. We may then take a further aggregation step for the sake of raising the analysis of distance relationships to an even higher level of generalization. This we can accomplish by fusing the three alternation-specificdistance matrices into a single compromise distance matrix merged across all lines and alternations, or 0-matrix for short.10 We can compare distance matrices at various level of aggregation by quantifying the overlap between the matrices using the Mantel test (Levshina, 2015, 348–349).11 The Mantel test is a permutation-based technique commonly used in ecology and designed to determine the correlation between matrices. The test yields correlation coefficients that range between −1 (total reversal) and +1 (total overlap). Mantel correlation coefficients can be interpreted as direct 10 We use the fuse() function in R package analogue to fuse distance matrices (see https://cran

.r-project.org/web/packages/analogue/analogue.pdf). All input matrices are weighted equally. This could in principle be changed, but we see no compelling reason to weigh up or down particular lines of evidence. 11 We use the mantel() function in R package vegan to calculate Mantel coefficients (see https: //cran.r-project.org/web/packages/vegan/vegan.pdf).

126

Distances, Similarities, and Coherence

Table 6.7 DBCalternation measurements: Mantel correlation coefficients between fused distance matrices (combining all lines of evidence and based on all available data). overlap genitive alternation/dative alternation overlap genitive alternation/particle alternation overlap dative alternation/particle alternation

r = 0.05 ( p = 0.41) r = 0.52 ( p = 0.01) r = 0.11 ( p = 0.31)

measures of distance-based coherence (DBC): the higher the coefficients, the higher the DBC. 6.5.1

Coherence Across Alternations: DBCalternation

To recapitulate, in this section we ask the following question: if, according to alternation A, two varieties are linguistically similar in terms of how people choose between different ways of saying the same thing, will the two varieties also be similar when the analysis is based on alternations B or C? To the extent that there is a correlation, we will see significant DBCalternation coefficients. Table 6.7 reports the resulting three correlation coefficients. In short, we do find significant and substantial overlap between the genitive alternation and the particle placement alternation, while the dative alternation does not overlap with either one of the other alternations. 6.5.2

Coherence Across the Spoken–Written Dichotomy: DBCmedium

The corpora we investigate to study grammatical variability cover both spoken language (e.g. face-to-face conversation, unscripted speeches) and written language (e.g. press news reports, web-based materials). Given this set-up, the question that we are going to ask now is the following: If two varieties turn out to be close when we restrict attention to grammatical choice-making in spoken production, will those varieties also turn out to be close when we restrict attention to choice-making in written language production? To address this question, we split up the datasets into spoken sub-datasets (covering spoken production in the ICE corpora) and written sub-datasets (covering written materials in ICE and GloWbE). Subsequently, we conduct separate VADIS analyses, which generate two different 0–matrices (one for spoken production and the other one for written production). To quantify DBCmedium , we once again turn to the Mantel test for the sake of measuring the correlation between the two distance matrices: r = 0.56 ( p = 0.02). According to customary thresholds in interpreting correlation coefficients in the social sciences, we thus see “substantial to very strong” (De Vaus, 2002, 272) overlap between the distance matrices. In other words, there is a good deal of coherence between spoken and written language production.

6.5 Evaluating Coherence

127

Table 6.8 Mantel correlation coefficients between line-of-evidence-specific distance matrices. Genitive alternation

Dative alternation

Particle alternation

overlap 1st line/2nd line r = 0.41 ( p = 0.03) r = 0.12 ( p = 0.34) r = 0.36 ( p = 0.05) overlap 1st line/3rd line r = 0.07 ( p = 0.36) r = −0.01 ( p = 0.50) r = 0.25 ( p = 0.13) overlap 2nd line/3rd line r = 0.47 ( p = 0.03) r = −0.15 ( p = 0.77) r = 0.68 ( p = 0.00)

6.5.3

Coherence Across Lines of Evidence: DBCevidence

Recall that the VADIS method draws on concepts and ideas originally developed in comparative sociolinguistics (e.g. Tagliamonte, 2001; Tagliamonte, 2012, 162–173; Tagliamonte et al., 2016a). One of these ideas is that variational similarity between lects may be evaluated by considering three lines of evidence, as we saw before: (i) Are the same constraints significant across varieties? (ii) Do the constraints have the same strength across varieties? (iii) Is the constraint hierarchy similar? DBCevidence quantifies the extent to which the three lines of evidence really measure the same thing or not: If two varieties turn out to be close if we restrict attention to a particular line of evidence (e.g. constraint significance), will those varieties also turn out to be close when we probe another line of evidence (e.g. constraint strength)? To evaluate DBCevidence , let us now consider three separate distance matrices per alternation, one for each line of evidence (rather than creating compromise distance matrices per alternation which merge line-specific distance matrices, as in the foregoing discussions). We next calculate Mantel correlation coefficient between these line-specific distance matrices. The results are shown in Table 6.8. We note, first, that the dative alternation is the odd one out in that none of the lines overlap with each other in this alternation. Second, the genitive alternation and the particle placement alternation are similar in that they both show moderate but significant overlap between the first line of evidence (constraint significance) and the second line of evidence (constraint strength), as well as substantial overlap between the second line of evidence and the third line of evidence (constraint ranking). We do not see significant overlap anywhere between the first line of evidence and the third line of evidence. In short, DBCevidence scores suggest that there is some – but not perfect – overlap (or redundancy) in the three lines of evidence. Again, a more thoroughgoing interpretation of these findings will be offered in Section 6.7.

128

Distances, Similarities, and Coherence

6.6

Mapping Out (Dis)similarity Relationships

In this section, we move on to visually depicting distances between the varieties subject to study, with the aim of mapping out (dis)similarity relationships between the underlying probabilistic grammars. We begin by noting that distance matrices of the type that we investigated in the previous section (see e.g. Figure 6.1) are informative but somewhat hard to process via eyeballing. Luckily, there are a number of techniques to visualize distance matrices. One of these is Multidimensional Scaling (MDS) (see e.g. Kruskal and Wish, 1978), which reduces a higher-dimensional distance matrix to a lower-dimensional representation. MDS is a dimension-reduction technique originally developed in psychometrics that translates distances between objects – in this case, linguistic distances between varieties of English – in high-dimensional space (note that in the distance matrix in Figure 6.1, each variety is characterised by distances to eight other varieties) into a lowerdimensional representation that can be visually depicted in two-dimensional plots. In such plots, distances between the lects are optimized to correspond as closely as possible to the linguistic distances in the original high-dimensional distance matrix. In other words, proximity in an MDS plot indicates linguistic similarity. For example, Figure 6.2 is an MDS representation of the distance matrix shown in Figure 6.1. Note that the axes (and the scaling that these depict) of MDS plots are typically not strongly interpreted, which is another way of saying that the absolute position of the data points in the plots does not matter much. What matters is proximity between data points: again, proximity in the plot is proportional to linguistic similarity, while distance in the plot is proportional to linguistic difference. Note now that BrE and NZE are close in the plot, while BrE and IndE are fairly distant – which is in line with the numerical values in Figure 6.2: BrE and NZE have a distance of only 0.095, while BrE and IndE are considerably more distant (0.548). Recall that per alternation, we are initially dealing with three separate distance matrices (one per line of evidence). Let us abstract away now from individual lines of evidence by fusing the three line-specific distance matrices, thus arriving at three line-merged but alternation-specific distance matrices. These are the same distance matrices that were analyzed in Table 6.7. Figure 6.3 displays the corresponding MDS plots. Inspection of the plots reveal differences between alternations: in line with our earlier finding that the dative alternation does not overlap much with the other alternation, we note that the dative alternation plot shows some different patterns than the other plots. For instance, only in the dative alternation plot does HKE end up very close to BrE, or CanE very distant from IrE. By contrast we see – again consistent with

6.6 Mapping Out (Dis)similarity Relationships

129

Figure 6.2 Multidimensional scaling representation of third line distances for the particle placement alternation (see Figure 6.1). Distances between data points in plot is proportional to probabilistic grammar distances between varieties.

our coherence analysis in Section 6.5.1 – more overlap between the genitive alternation and the particle placement alternation plots. Specifically, in both plots we see reasonably tight clusters of the Inner Circle varieties BrE, IrE, CanE, and NZE. This cluster is absent from the dative alternation plot. With all that being said, there are also patterns that are common to all three plots: for example, across all three alternations, IndE and PhlE are distant from the other varieties. As we explained in Section 6.5, the analysis of probabilistic (dis)similarities between varieties may be raised to a higher level of abstraction by fusing the three alternation-specific-distance matrices into a single compromise distance matrix merged across all lines and alternations, or 0-matrix for short. This 0-matrix indicates cross-constructional probabilistic distances, as it were, between the varieties under study. An MDS visualization of this matrix is shown in Figure 6.4. We note first of all that the three Outer Circle varieties JamE, IndE, and PhlE can be considered outliers, thanks to their peripheral position in the plot. Enclosed in the box we find the four Inner Circle varieties BrE, IrE, NZE, and CanE, a pattern similar to that on display in the genitive alternation and particle placement alternation plots in Figure 6.3. This comparatively tight cluster of Inner Circle varieties also echoes our earlier finding based on the analysis of core grammar coefficients (see Table 6.6) that Inner Circle varieties form a tighter (i.e. internally more homogeneous) cluster than Outer Circle varieties. Also located in the box is SgE, a variety that we generally consider in this book an Outer Circle variety a priori. However, at the risk of indulging in a just-so story, we take the liberty to note that according

130

Distances, Similarities, and Coherence (a)

(b)

(c)

Figure 6.3 Multidimensional scaling representation of compromise distances per alternation: a) genitive alternation; b) dative alternation; c) particle placement alternation. Distances between data points in plots is proportional to probabilistic grammar distances between varieties. Boxes depict Inner Circle clusters.

6.6 Mapping Out (Dis)similarity Relationships

131

Figure 6.4 Multidimensional scaling representation of the 0matrix (a single compromise distance matrix merged across all lines and alternations). Distances between data points in plot is proportional to probabilistic grammar distances between varieties. Box depicts Inner Circle cluster.

to some analysts (see e.g. Leimgruber, 2013, 122) SgE is an Outer Circle variety actually in the process of becoming an Inner Circle variety.12 In any event, what Figure 6.4 shows is a fairly robust pattern of Inner versus Outer Circle varieties. We now turn to two supplementary techniques designed to analyze distance matrices, for the sake of exploring our 0-matrix from a couple more angles. We begin with Hierarchical Agglomerative Cluster Analysis (Aldenderfer and Blashfield, 1984; Jain et al., 1999). The technique can be used to group a number of objects (in this study, varieties of English) into a smaller number of discrete clusters. While there are many different clustering algorithms to accomplish this task, we use here Ward’s Minimum Variance method (Ward, 1963), an algorithm that tends to create small and even-sized clusters and which is popular both in corpus linguistics (Gries and Wulff, 2005) and in dialectometry (e.g. Goebl, 2008).13 Cluster analysis yields so-called dendrograms, which essentially work like family trees and additionally depict cophenetic distances between the clustered objects – the longer the lines in the dendrograms, the more dissimilar varieties are. The dendrogram for our 12 We do acknowledge that this implies a teleological view of the evolution of these phenomena:

as an Outer Circle variety is increasingly made up of greater numbers of L1 speakers, its probabilistic grammars should gradually converge with those of other Inner Circle varieties. This is an argument that does place a lot of explanatory weight on L1 acquisition processes at the level of individual speakers. 13 We use the hclust() function in R (method=“ward.D2”).

132

Distances, Similarities, and Coherence

Figure 6.5 Dendrogram: clustering the 0-matrix (a single compromise distance matrix merged across all lines and alternations). Hierarchical agglomerative cluster algorithm: WARD.

0-matrix is shown in Figure 6.5. Observe first that HKE and IndE form a cluster that is separate from the other varieties, thus confirming cluster-analytically their peripheral location in the MDS plot in Figure 6.4. The other varieties are located in the right branch of the dendrogram. The two European varieties of English in our sample, BrE and IrE, form a subcluster, and so do NZE and SgE. PhlE and JamE are also part of the right branch. The split according to cluster analysis between Inner Circle and Outer Circle varieties is not as clear, then, as the pattern we have seen in the MDS plot in Figure 6.4 – these are two different analysis techniques, after all – but we do note that all Inner Circle varieties are located in the right branch of the dendrogram. We also add that supplementary inspection of silhouette widths (see Levshina, 2015, 312) indicates that the 0-matrix lacks substantial cluster structure. Next we would like to discuss a NeighborNet diagram visually depicting the 0-matrix. NeighborNet diagrams were originally developed in bioinformatics to represent uncertainty in phylogenies and reticulate effects such as genetic recombination (see e.g. Huson and Bryant, 2006). They are now also increasingly utilized in all those fields in linguistics that are in the business of analyzing distances between linguistic objects, such as cross-linguistic typology (e.g. Wälchli, 2014), dialectology (e.g. McMahon et al., 2007), and

6.6 Mapping Out (Dis)similarity Relationships

133

variationist linguistics (e.g. Grafmiller and Szmrecsanyi, 2018). The NeighborNet algorithm is related to hierarchical agglomerative cluster analysis methods. But unlike classical cluster analysis, NeighborNet diagrams can depict conflicting signals in the underlying distance matrices. This means in practice that when the distance matrix – or a segment of the matrix – can be adequately represented as a hierarchical tree (again, as in Figure 6.5), the neighbor-net diagram (or the relevant segment in the diagram) will be tree-shaped. But when the distances provide a conflicting signal, so-called reticulations (which indicate the coexistence of several possible tree structures) will be drawn. The resulting boxy shapes are in biology often interpreted as being indicative of horizontal gene transfer, and in linguistics as suggesting language contact. Szmrecsanyi and Wolk (2011) sum up the technicalities as follows: As with hierarchical agglomerative cluster analysis, the distance matrix is searched for the pair of points with the shortest distance. Instead of immediately fusing these points to a single cluster, however, they are just marked, and this procedure is repeated until the same point is marked twice. Then, these points are replaced with two clusters, each representing the doubly marked point in relation to one of its marked neighbors. This process is repeated until only three clusters are left. The fusion sequence can subsequently be used to generate a network-like diagram. This procedure has some beneficial properties. First, the result will not be needlessly complex. For cases where a segment of the data can be adequately represented as a hierarchical tree, the corresponding segment of the network will be tree-shaped. Second, the method will always produce graphs that are planar, i.e. without crossing lines, which aids visual interpretation. (Szmrecsanyi and Wolk, 2011, 575)

In what follows we use the NeighborNet algorithm without insisting on a strictly phylogenetic interpretation. Instead, we use the method essentially as an advanced visualization technique to highlight general trends in a complex dataset and to explore the similarities between the varieties under study. Figure 6.6 shows a NeighborNet diagram depicting the 0-matrix.14 In the diagram, each line represents a way of splitting the total set of varieties into exactly two groups. The longer a given line or set of lines, the greater the difference between the groups. For example, IndE and HKE (in the lower-right hand part of Figure 6.6) are fairly obviously set apart from the other varieties, a split that we also saw in the dendrogram in Figure 6.5. Also as in the dendrogram in Figure 6.5, BrE ends up close to IrE, and NZE close to SgE. But in addition the NeighborNet diagram reveals that CanE also wants to be close, via reticulations, to PhlE – an Outer Circle variety that, for historical reasons, has “an American rather than a BrE touch” (Schneider, 2011, 155). BrE, on the other hand, is shown to have similarities not only to IrE but also to SgE. The general pattern, then, on display in Figure 6.6 is that all Inner Circle varieties plus SgE are located in the left part of the diagram while most Outer Circle varieties are 14 We use the neighborNet() function in R package phangorn to calculate the NeighborNet (see

https://cran.r-project.org/web/packages/phangorn/phangorn.pdf). The output is similar to that of the well-known SplitsTree package (Huson, 1998).

134

Distances, Similarities, and Coherence CanE PhIE NZE

JamE SgE

BrE

HKE IrE

IndE

Figure 6.6 Visualizing aggregate similarities: NeighborNet diagram depicting the 0-matrix (a single compromise distance matrix merged across all lines and alternations). Internode distances (branch lengths) are proportional to cophenetic linguistic distances.

located in the right part of the diagram. Reticulations we find especially in the upper left-hand part of the diagram. As we saw, reticulations are basically representations of uncertainty in the data, that is, the algorithm’s inability to make robust splits given the evidence. The absence of clear splits, in turn, indicates that phylogeny (in biology) or family tree models (in linguistics) cannot tell the whole story. This is why reticulations are interpreted as being suggestive of horizontal gene transfer (in biology), and of language or dialect contact (in linguistics). We would thus conclude from Figure 6.6 that contact may play a role particularly when it comes to the relationship(s) between BrE, SgE, NZE, CanE, PhlE, and JamE.15 15 All that said, we hasten to add that NeighborNet diagrams do not provide conclusive evidence

for language/dialect contact – the reticulations are merely suggestive.

6.7 Discussion

6.7

135

Discussion

This chapter has utilized Variation-Based Distance and Similarity Modeling (VADIS) to quantify the distance and similarity between varieties of English as a function of the (non)correspondence of the ways in which language users choose between different ways of saying the same thing – concretely, how language users choose between genitive, dative, and particle placement variants. As we have seen, VADIS is inspired by concepts and ideas developed in comparative sociolinguistics, by work in experience-based/usagebased probabilistic grammar, and by techniques widely used in dialectometry and quantitative typology. The idea, in short, is that we can gauge the extent and structure of inter-speaker variation by assessing the systematicity of intra-speaker variation. Informed by methodologies developed in comparative sociolinguistics, VADIS relies on three lines of evidence: (i) Are the same constraints significant across varieties? (ii) What is the extent to which constraints have similar effect strengths? (iii) What is the extent to which the ranking of constraints is similar? Key findings uncovered in this chapter may be summarized as follows. As to quantifying similarities and dissimilarities, Section 6.4 calculated VADIS similarity coefficients to assess the extent to which language varieties are similar in terms of the probabilistic grammars that regulate any number of variables and alternations. Analysis shows that the nine World Englishes under study here are overall remarkably similar to each other in terms of variation patterns: on a scale from 0 (total dissimilarity) to 1 (total similarity), core grammar scores range between 0 = 0.6 and 0 = 0.8, which means that there is overall strong overlap with regard to the probabilistic grammars regulating variation. The interpretation is that we are dealing with a rather solid probabilistic “common core” (Quirk et al., 1985, 33) of the grammar of English. With that being said, we also saw that all grammatical alternations are not equal, and therefore the evidence presented in this chapter converges with what we saw in Chapter 5. In particular, VADIS suggests that the genitive alternation tends to be more stable across varieties than the other alternations. Again, we interpret this as evidence that the alternations under study are differentially sensitive to “probabilistic indigenization” (Szmrecsanyi et al., 2016a, 133; see Chapter 5.5.4 for discussion), with more abstract alternations – such as the genitive alternation – being less hospitable to probabilistic indigenization. Experimentation with subsets of the data subsequently provided two key insights: First, spoken language production tends to be more heterogeneous (read: regionally unstable) than written language production, because similarity coefficients are lower when attention is restricted to spoken materials. This is a bit surprising given a widely held suspicion that the production of spoken language is subject to universal (and thus potentially homogenizing)

136

Distances, Similarities, and Coherence

processing and production constraints and biases (e.g. Hawkins, 1994; MacDonald, 2013) in a way that the production of written language is probably not. We also know, however, that while especially vernacular speech is “the style in which the minimum attention is given to the monitoring of speech” (Labov, 1972, 208), written language is more “governed by prescription” (D’Arcy and Tagliamonte, 2015, 255), a fact that may level out regional differences. Second, with regard to Kachru’s (Kachru, 1992, 1985) distinction between Inner Circle (native L1) and Outer Circle (indigenized L2) varieties of English, the evidence suggests that Inner Circle varieties form a tighter typological cluster (i.e. similarity coefficients are higher) than the Outer Circle varieties, where similarity coefficients are lower. The comparative heterogeneity of Outer Circle varieties is likely due to regionally diverse substrate and contact influences, which plausibly play a more important role in the Outer Circle than in the Inner Circle. In Section 6.5 we took distance-based coherence (DBC) measurements, building on the notion that coherence between sub-datasets can be operationalized as the extent to which these sub-datasets yield similar distance relations between varieties. As to coherence between alternations (DBCalternation ), we saw that the distance matrices derived from the genitive and particle placement alternations do overlap substantially, but the dative alternation distance matrix does not overlap significantly with any of the other distance matrices. In other words, the dative alternation is an outlier. The deeper theoretical question here is whether grammar (or the variable parts of grammar) is essentially a collection of independent and/or independently conditioned alternations, or whether alternations actually ‘agree,’ as it were, about lectal or varietal differences. Our analysis suggest that we are dealing with a mixed picture, in that we see both coherence and noncoherence. With regard to non coherence, it is unexpected that (and unclear why) the dative alternation does not pattern with the other alternations – all three alternations under study are, after all, permutation alternations that are constrained by similar factors (constituent length, animacy, and so on). Further work is needed to investigate why the dative alternation is different from the other alternations. What is clear is that this finding calls into question the idea that grammar is the orderly aggregation of binary alternations. As to coherence across the spoken-written distinction (DBCmedium ), the data show that there is substantial overlap between distance matrices calculated on the basis of spoken materials only and distance matrices calculated on the basis of written materials only. This qualifies to some extent our earlier finding that spoken language production tends to be more heterogeneous (that is, regionally unstable) than written language production: yes, inter-variety distances in the spoken medium may be more pronounced, but they still correlate significantly with the (smaller) distances we find in the written medium. Thus, as a rule, if two varieties turn out to be similar when we restrict attention to spoken production, those varieties will also tend to be similar when we restrict attention to choice-making in written language

6.7 Discussion

137

production. This sort of coherence is maybe not entirely unexpected to those who have always believed that spoken and written language production go underlyingly by the same rules. Finally, as to coherence across lines of evidence (DBCmedium ), analysis shows that there tends to be overlap between the first line of evidence (constraint significance) and the second line of evidence (effect size), as well as between the second line of evidence and the third line of evidence (constraint ranking). This is true for the genitive alternation and the particle placement alternation; the distance matrices generated on the basis of data from the dative alternation do not overlap at all (underscoring the outlier status of the dative alternation established earlier). All this emphasizes the wisdom of designing the comparative sociolinguistics method such that it relies on three lines of evidence that show partial overlap but do not measure exactly the same thing. The prime take-home message from mapping out (dis)similarity relationships using Multidimensional Scaling (MDS) plots, cluster analysis dendrograms, and NeighborNet diagrams (Section 6.6) is that more often than not, we see a more or less robust split between Inner Circle and Outer Circle varieties – a split that is well-known in the dialect-typological literature on World English (see e.g. Röthlisberger and Szmrecsanyi, 2019) and that we already encountered in Chapter 5.5.3. The reasoning is once again that Outer Circle varieties, as second-language varieties, are subject to substrate and contact influences in a way that Inner Circle varieties are not. In addition, SLA universals (in the spirit of e.g. Klein and Perdue, 1997) shape Outer Circle grammars, but not Inner Circle grammars. We hasten to add that the Inner Circle/Outer Circle split is virtually absent from the dative alternation distance matrix (see Figure 6.3-b) – again emphasizing the outlier status of the dative alternation, for reasons that are presently unknown. That said, according to VADIS measurements the Inner Circle/Outer Circle split is fairly obvious in the genitive alternation and the particle placement alternation. We conclude this chapter with some methodological remarks. We believe VADIS is a promising approach for scaling up comparative variationist methods, yet a few questions remain regarding its validity and reliability. For one, it is difficult to evaluate the validity of the method because we rarely have ways of independently measuring grammatical similarities in probabilistic conditioning among varieties. Correlations with other external factors – region, nativeness, Dynamic Model phases, etc. – are at best just correlations and at worst invitations to confirmation bias. But before we can start interpreting our data, we need to know that the method can adequately represent genuine differences among variable grammars. A related problem is that we lack a baseline of meaningful (dis)similarity against which we can evaluate our results in any given case. In the spirit of other guides on meaningful effect sizes (e.g. Cohen, 1992), we would like to have some idea what a range of small, medium, or large similarity coefficients or coherence scores is likely to be. Finally, it is vital to bear in mind that the statistics obtained from the underlying models

138

Distances, Similarities, and Coherence

MDS dimension 2

0.2 0.1 0.0 Variety

−0.1

A B

−0.2

C D

−0.3

E

−0.4

−0.2

0.0

MDS dimension 1

0.2

0.4

Figure 6.7 Multidimensional scaling representation of the fused distances from all three VADIS lines for seventy-five simulated datasets representing five hypothetical dialect varieties. Points represent hypothetical “speakers” of five “dialects.” Distances between points in the plot are proportional to probabilistic grammar distances between datasets. Note that when simulating the data, varieties C and D were intentionally designed to be much more similar to one another than to the other three varieties.

(e.g. coefficient estimates in regression models), always come with some degree of uncertainty, and this should in principle be taken into account somehow. Ideally we would like to have a way of incorporating the variance in the underlying models into the VADIS output. With these concerns in mind, we conducted a simulation study in which we artificially created realistic datasets for a hypothetical grammatical variable of the kind common in comparative sociolinguistic studies, and applied the VADIS method to them. Specifically, we created five hypothetical “dialects” each with a distinct variable grammar, and for each “dialect” we simulated 2000 observations for fifteen hypothetical “speakers,” who varied only in their baseline frequency of use of the hypothetical outcome variants. These simulated datasets were specifically designed to vary in the degree of similarity among them, and thus allowed us to test the validity and reliability of the method in a hypothetical scenario. The complete details of this study are in a supplementary document available at https://osf.io/5hvtw. In our simulated data we find that the VADIS method is capable of accurately capturing genuine grammatical differences, where differences are most effectively represented through the visualizations such as MDS plots (Figure 6.7) or cluster diagrams (Figure 6.8). Thus while the method does not provide tests

6.7 Discussion Cn Ch Ce Cd Ci Cl Ck Ca Cb Co Cc Cf Cj Cg Cm Dj Dc Dn Dl Di De Dk Do Dm Dh Dg Df Da Dd Db El Ee Em Eh Ef Ek Ej Eg Eo Ei Ed Ea En Eb Ec Ao Ag Aa Ae Ac Aj Am Ak Ah Ai An Ad Ab Af Al Bl Be Bg Bb Bi Bm Bj Ba Bn Bd Bc Bh Bf Bo Bk

Figure 6.8 Hierarchical clustering of simulated datasets based on VADIS Line 2. Leaf nodes represent individual “speakers” (lowercase letters) across five distinct “dialects” (uppercase letters), for example, “Bk” represents speaker “k” in dialect “B,” and so on. Note that dialects are clearly separated from one another as we designed them to be.

139

140

Distances, Similarities, and Coherence

of statistical significance, it can provide a useful way of exploring grammatical similarity among varieties in a more holistic fashion. We also estimate that Similarity Coefficients (based on the fused lines) greater than 0.85 represent practically indistinguishable variety grammars, and we can also make a tentative case for scores of 0.5 as representing a very low degree of similarity. These values are based on average measures of within-dialect similarity, where individual speakers in a given dialect were simulated to have identical constraints yet differing baselines. Thus we have a conservative estimate of the reasonable upper bound of similarity that we are likely to see in real data. On the downside, results from the different lines of evidence were sometimes inconclusive (especially for line one) or even contradictory, and further evaluation of how best to combine the lines is something we are continuing to investigate. Thus the case for the reliability of DBCevidence measures remains open. Lastly, we find that the inherent uncertainty in the underlying predictive models can in principle be represented in the visualizations via, for example, MDS “variety clouds” and/or the distributions of similarity scores (see the supplementary document for illustration). Our simulation study demonstrates the potential value of the VADIS method as a unique and powerful tool for interpretation in comparative variationist analysis. The method offers a reliable approach for the exploration of grammatical similarity among different varieties’ linguistic variables. We also show that the method can easily scale up to compare dozens of varieties (dialects, individuals, etc.) at a time, although further testing and application in real world contexts is necessary to more fully understand the strengths and limitations of the method.

7

Experimental Corroboration

In this chapter we discuss the results of the experimental component of the study, which is designed to examine the degree to which the probabilistic contrasts uncovered in the corpus analysis are also discernible in language users’ introspective preferences. To our minds there is now an undeniable body of evidence that linguistic knowledge is profoundly shaped by aspects of experience which manifest themselves in complex and often quite subtle influences on linguistic behavior (Bybee, 2010; Divjak and Gries, 2012; Ellis et al., 2016; Goldberg, 2019; Pütz et al., 2014; among many others). Empirically, these influences are detected in frequency patterns and statistical correlations between linguistic and contextual features, and the increased interest in statistical analysis of naturally occurring language data (corpora) has fuelled a “quantitative turn” (Janda, 2017) in linguistics. Corpus-based variationist research is one of many research paradigms that have capitalized on this turn, and recent years have seen a dramatic rise in the use of ever more sophisticated techniques for analyzing and modelling variationist corpus data (Gries and Deshors, 2015; Barth and Kapatsinski, 2017; Szmrecsanyi et al., 2019; Tagliamonte and Baayen, 2012). At the same time, there is a growing awareness for the need for methodological pluralism (Klavan and Divjak, 2016), and the use of experimental approaches as a complement to corpusbased observational studies has also become increasingly popular (Bresnan and Ford, 2010; Divjak et al., 2016; Engel et al., 2022; Gilquin and Gries, 2009; Hoffmann, 2011; Horch, 2019). The use of corroborating methodologies is valuable as both an independent check on the (implications of) findings from individual studies, as well as a means of probing research questions and hypotheses that are ill-suited to corpus-based approaches. We enthusiastically embrace this pluralistic approach, and to that end conducted a series of experiments designed to corroborate and interrogate the scope of the patterns observed in our corpus models. Our experimental investigation is guided by three related research questions. RQ1 The first question asks: when participants are provided with the same contextual information as a corpus model, do they exhibit a similar manner of graded preferences for the use of a particular variant? 141

142

Experimental Corroboration

RQ2 Our second question asks: when changes in contextual features (e.g. accessibility, animacy, length) correlate with changes in the probability of a variant in the corpus model, do those contextual changes correlate in the same way with changes to participants’ ratings? In other words, if a change in constituent length decreases the corpus probability of a variant, does that change in length also decrease participants’ ratings for that same variant? RQ3 Finally, we are interested in cross-varietal patterns as well, and our third question asks: to what extent do cross-varietal differences in the probabilistic effects of internal factors in a corpus model correlate with cross-varietal differences in effects of those same factors on participants’ ratings? Our hypothesis is that language production and metalinguistic judgments are tapping into similar experience-based aspects of linguistic knowledge, therefore we should see strong correlations between ratings and corpusmodel-based probabilities, and furthermore, these correlations should covary with changes to the specific linguistic context as well as general dialectal differences. With these questions in mind, we present here a set of experiments across four varieties – British English (BrE), New Zealand English (NZE), Indian English (IndE), and Singapore English (SgE) – in which we set out to replicate some of the contrasts we observed in corpus analyses from these same varieties (e.g. Szmrecsanyi et al., 2016a; Grafmiller and Szmrecsanyi, 2018). We focus on one phenomenon, the particle placement alternation, and examine the degree to which one of its most dominant contextual constraints, the length of the direct object, might influence participant ratings in an offline judgment task. Results of our experiments provide fairly strong evidence that participant ratings do correlate – to an impressive degree in fact – with participants’ judgments overall (RQ1). We also find evidence that participants are sensitive to contextual cues (RQ2), specifically the length of the direct object in the particle placement alternation. As to our third research question, we find a noticeable cross-varietal difference in the overall preference for one variant over another in the same direction as in the corpus data (i.e. BrE and NZE speakers favor the split variant more so than IndE and SgE speakers), and we find slight evidence of differences in the effects of direct object length among the varieties, though again only when we compare the Inner Circle varieties (BrE and NZE) to the Outer Circle ones (IndE and SgE). While largely successful, our experiments also raise important methodological and theoretical questions. Despite some early successes in this line of research (e.g. Bresnan and Ford, 2010), our findings suggest that the relationship between corpus probabilities and judgment ratings is not straightforward (see also Divjak et al., 2016), and that we should be careful about extrapolating between observational and experimental approaches.

7.1 Methods

7.1

143

Methods

For our experimental design we followed the work pioneered by Bresnan (2007) who use the so-called “100-split task” to assess participants’ preferences for one or the other variant in a given alternation (see also Bresnan and Ford, 2010; Ford and Bresnan, 2013). In this design, participants are presented with observations of a given alternation sampled directly from corpus datasets, and asked to distribute 100 points between two competing variants. The test alternations are presented as part of their surrounding context in order to assess the influence of contextual factors and to increase the overall ecological validity of the stimulus items (compared to rating sentences in isolation). An example stimulus item involving the particle placement alternation is provided in (52). (52)

The programme had a biggish audience (in radio terms) because it followed the Today programme, and because people listened to it in their cars on the way to work. They either loved it or loathed it. I once had a fan letter from Neil Kinnock saying what a good way it was to start Monday morning and asking me how I got away with it. On the other hand, I got a letter from a regular BBC correspondent who said he always [turned the radio off / turned off the radio] immediately if it was my turn on the programme, but he would like to take issue with something I had said last week. turned the radio off turned off the radio

Participants would then be asked to distribute points for each of the two variants according to their preference in this context. Using genuine corpus examples as stimulus items allows us to directly compare the degree to which participants’ preferences for the respective variants aligns with the probabilities derived from the corpus data in our statistical models. For example, the corpus-based model assigns a 68 pecent probability of the split variant turned the radio off in (52). The question then becomes, how will the participants rate this same token – will they assign roughly the same range of points (60–70) to the turned the radio off variant? Significant positive correlations between participants’ ratings and the probabilities assigned by the corpus model provides evidence that similar forces are guiding both the offline meta-linguistic judgments and online language production. 7.1.1

Participants

Due to the challenges of conducting surveys of speakers from different regions around the globe, participants were recruited using several different methods. British participants were recruited via the Prolific online recruitment

144

Experimental Corroboration

platform (https://prolific.co), but we were unable to recruit sufficient participants from the other three regions with this platform. We therefore used the recruitment services offered by Qualtrics to recruit participants from Singapore and New Zealand, and Indian English speaking participants were recruited by work of mouth through colleagues in contact with large populations of IndE speakers. For each country, we initially recruited 100 participants at random, with the expectation that an unknown percentage would be filtered out based on certain criteria. To ensure that participants were representative of their respective varieties, we included several posttest demographic questions about where participants grew up and lived most of their lives, whether English was their first language, and whether they had taken a linguistics course before. For India and Singapore participants, we also asked additional questions about their use and proficiency in English (53). (53)

a. How would you rate your level of English? (0–100) b. How confident do you feel using English? (0–100) c. How often do you use English in your daily life? (“Every day,” “at least once a week,” “a few times a month,” “hardly ever”) d. How many years have you been using English?

For all four countries, we excluded any participants who reported growing up in a country other than the target one, and anyone who reported having taken a linguistics class. For BrE and NZE participants we also excluded anyone claiming not to be a native speaker. For IndE and SgE participants, we included only those who claimed to use English every day, and who rated their confidence in using English above fifty. We also excluded participants who either were too quick or too slow in their response as well as those who failed the comprehension checks (see below), in order to ensure participants were properly attending to the task. In total, 260 participants were included in the final analysis (BrE = 60, NZE = 81, IndE = 55, SgE = 64).1

7.1.2

Materials

Thirty stimulus items were created, each consisting of an edited excerpt from the BrE component of the ICE corpus dataset. For the presentation of the items, we inserted the relevant choices into the text, where they were highlighted for the participant. An example item is shown in (54). The order of the split and continuous variants was randomized across items. 1 Results presented here are based on data from all 260 participants. We ran an additional analysis

using a randomly downsampled dataset of only fifty participants from each variety and found virtually no difference in the results.

7.1 Methods

(54)

145

Much time and effort was spent on board structures – experiments were tried with worker-directors. They failed because it was almost impossible to define their role, particularly as it was clear that they represented no one. They were not elected by their fellow workers – nor even appointed by their trade unions. So the steel industry developed, as had many of the earlier nationalized industries, into a prime example of conflict between management and labour. There was little evidence of management and workers trying in harmony to create an enterprise to serve the nation. Nationalization had merely [taken the ownership and ultimate control away / taken away the ownership and ultimate control] from private investors, and had passed it on to the politicians.

We sampled items from across the range of corpus model probabilities derived from a model of BrE ICE data, as reported by Grafmiller and Szmrecsanyi (2018). Specifically, six items were selected from each of five bins corresponding to the probability quintiles of the distribution of model predictions. We selected items with varying direct object lengths within each bin as much as possible, though the distribution of tokens within the corpus data limited this to some extent, as tokens at the extreme ends of the distribution do not vary much with respect to direct object length. Likewise, it was not possible to control for the specific particle verb with this method, as not all verbs were present in every bin. However, the corpus model did include controls for the effect of individual lexical items (e.g. through random effects), and we include by-item controls in our analysis of the experimental data below. More precise control and construction of test items would help address potential concerns about possible lexical effects, but with artificially constructed items we lose the direct comparability with the observed corpus data. Figure 7.1 shows the fitted corpus model probabilities for the thirty items used in our experiments. Items were presented in pseudo-random order, and manually adjusted to reduce the appearance of patterns. Where necessary, items were adapted to fit a British, New Zealand, Indian or Singaporean context by changing or removing names or lexical items that might not be familiar to speakers of the respective national varieties. After editing, all experimental items were independently reviewed and assessed for authenticity by native speakers of English from Great Britain, New Zealand, India, and Singapore. In addition to the thirty test items, we created fifteen filler items, which presented lexical or grammatical choices comparable to the particle placement alternation. For example, filler variables included the genitive alternation, the dative alternation, complementizer omission, and variation among relativizers (that vs. which). The full lists of test and filler items is available at the OSF repository at https://osf.io/5hvtw.

146

Experimental Corroboration

Corpus model probability

1.00

0.75

0.50

0.25

0.00 A11 B11 B2 A2 A1 B1 A3 A4 B3 B12 B13 A12 B4 A13 B14 B5 A14 A5 A7 B15 B10 A6 A15 B6 A9 A10 B8 A8 B7 B9

Item

Figure 7.1 Experimental items versus corpus model predicted probabilities.

7.1.3

Procedure

The surveys were run online using the Qualtrics software. Participants were shown a welcome page with instructions on how the survey would work (Figure 7.2). They were told that we were interested in how people choose between different ways of saying or writing the same thing. They were told that they would see a number of excerpts from different texts about various topics, and that each text includes sets of choices that the writer had to make about how to say certain things. Participants were instructed that their task was to read each passage carefully and rate how natural each of the options sounded. The survey was delivered via Qualtrics’ online interface. Rather than assign points directly to the items, participants were presented with a slider button with the two variants on either end, and asked to move the slider towards the variant that they felt sounded more natural in the context provided. Participants were shown an example illustrating the task as part of the initial instructions (Figure 7.2). As an additional check, we included five comprehension questions about a few texts’ content in order to test whether participants were reading the texts. Participants who answered more than two comprehension questions incorrectly

7.1 Methods

147

Figure 7.2 Welcome page and instructions for the ratings experiments.

were excluded. At the end of the survey, participants were asked to provide personal information regarding their gender, age, education level, and degree of familiarity and comfort with English. In order to keep the survey length to a manageable duration, we created two blocks of fifteen test questions together with the fifteen filler items. For the

148

Experimental Corroboration

BrE

NZE

100 75

Preference rating for split variant

50 25 0

−7.5

−5.0

−2.5

0.0

2.5

SgE

−7.5

−5.0

−2.5

0.0

2.5

−5.0

−2.5

0.0

2.5

IndE

100 75 50 25 0

−7.5

−5.0

−2.5

0.0

2.5

−7.5

Corpus model predictions for split variant (logit scale)

Figure 7.3 Experimental ratings versus corpus model predicted log odds, with LOESS smooths (BrE r = 0.54; NZE r = 0.48; IndE r = 0.47; SgE r = 0.42). Ratings > 50 indicate greater preference for the split variant (pick the book up), ratings < 50 indicate greater preference for the continuous variant (pick up the book). Dots show ratings for each individual trial.

first block, three test items were randomly selected from each probability bin, and the remaining items were used in the second block. Each participant saw only one block, thus each participant saw only half of the test items. Pilot tests estimated the task to take approximately twenty minutes, and participants who took less than ten minutes or longer than one hour were excluded. 7.2

Results

7.2.1

Corpus Predictions versus Ratings in Aggregate

Before moving to the statistical analysis, we first visually explore the patterns in our participants’ responses. Looking at the relationship between the corpus model predictions and experimental ratings in total (Figure 7.3), we find strong

7.2 Results

149

Variety Predicted rating for split variant

100

75

73.5

70

50

58

58

SgE

IndE

25

0 BrE

NZE

Figure 7.4 Experimental ratings across varieties, with median rating. Ratings > 50 indicate greater preference for the split variant (pick the book up), ratings < 50 indicate greater preference for the continuous variant (pick up the book).

positive correlations among participants from all four varieties: BrE r = 0.54; NZE r = 0.48; SgE r = 0.41; IndE r = 0.47. Setting aside the slight dips between x = −7 and x = −3 due to missing data, the trends are strikingly linear, although we do find a considerable number of cases in which participants assigned the maximum or minimum rating, or even express no preference at all (i.e. rating = 50). A complete list of corpus model predictions and experimental ratings for every participant can be found in the online supplement. Averaging across all participants and items, we find that the median ratings for the split variant among BrE and NZE participants was roughly 12–15 points higher than among the SgE and IndE participants (Figure 7.4). This is in line with the findings from the ICE data, which showed higher proportions of split particle verbs in the Inner Circle varieties. As for the individual stimulus items, we find a similar degree of consistency among all four varieties. Figure 7.5 plots the average rating across participants for each individual item against the corpus model predictions, and the means and standard deviations for each stimulus item across the four varieties are provided in Table 7.1. Items at the extreme upper and lower ends of the corpus probabilities appear to be rated similarly for participants across all four varieties. For example, for items A6 and B10 (55), the participants and corpus model largely agree that the split variant is considerably better/more likely than the continuous variant, while for items A11 and B11 (56), the participants and corpus model agree that the split variant is considerably worse/less likely than the alternative.

150

Experimental Corroboration

BrE

NZE

100

B10 B8 A6 A10B9 A9 B3 A15 B15 B7 A5 A4 A8 B1 B5 B6 B12 A7

75

Mean preference rating for split variant

50 B11

25 0

A1

A2 B2

A11

−7.5

−5.0

A11

0.0

2.5

SgE

−7.5

−5.0

−2.5

75 50

A1

B11 A11

−7.5

2.5

B10 B8 A15 A6 B9 B4 A8 A4 A5 A9 B7 A10

B10 A15 B8 A6 B9 B12 B5 B6 A9B7

25

0.0

IndE

100

0

B11

A3

−2.5

B8 A6 B7 A9 A10 B10 A15 B9 A5 A4 A8 B15 A7 B3 A1 B6 A13 B1 A2 A3 B2 B4

B4

A2 B2

−5.0

A4 B3 B1

A10 A8 A13

A1

A7 A11

A3

−2.5

B11

0.0

2.5

−7.5

B2

−5.0

B3

B6

B1 A2 A3

−2.5

A7

0.0

2.5

Corpus model predictions for split variant (logit scale)

Figure 7.5 Experimental ratings versus corpus model predictions, averaged by item and variety. Ratings > 50 indicate greater preference for the split variant (pick the book up), ratings < 50 indicate greater preference for the continuous variant (pick up the book). Dots show ratings for each individual trial.

(55) High split variant rating. a. When you’re slogging away every day it’s good to get (back) something (back). [A6] b. Walk tall; keep (up) your head (up) and your mind focused on your surroundings. [B10] (56) Low split variant rating. a. Dr Fuller knew he had carried (out) a caning that was fully deserved (out) and he had done his duty. [A11] b. Though Mr Leigh-Pemberton did not rule (out) a role for such an institution (out) . . . [B11] For a more complete picture of the distribution of ratings across individual items, see the supplementary materials available online.

7.2 Results

151

Table 7.1 Average ratings (with standard deviations) by item across participants. Ratings > 50 reflect preference for the split variant; ratings < 50 reflect preference for the continuous variant. BrE

7.2.2

NZE

SgE

IndE

Item

Mean

SD

Mean

SD

Mean

SD

Mean

SD

B10 B8 A6 A10 B4

94.7 94.0 87.9 87.8 82.5

14.4 13.2 18.9 16.1 21.5

88.7 92.7 90.0 80.5 89.8

23.9 16.9 22.1 32.7 15.7

81.2 84.2 77.6 62.1 60.7

24.0 23.5 29.9 41.4 37.9

88.4 86.6 87.6 55.1 76.4

23.8 24.5 22.4 36.2 33.3

A9 B9 A15 B7 A8

82.4 77.3 76.1 68.1 67.2

18.7 15.5 22.0 23.9 25.2

89.9 80.0 79.4 81.2 66.1

21.6 24.7 31.0 30.6 37.5

72.4 71.9 82.9 61.6 56.2

37.4 34.0 22.3 39.6 37.7

69.1 79.6 77.0 55.5 71.4

32.7 25.8 27.6 40.3 33.3

B3 B15 A5 A12 B14

66.5 65.2 58.5 57.8 56.9

27.2 29.1 35.1 30.7 35.1

47.9 64.6 67.4 58.7 64.5

37.4 35.4 39.2 40.2 39.0

43.2 56.2 59.2 61.6 53.2

35.6 39.5 42.3 39.6 39.9

35.4 48.5 69.6 55.4 51.7

34.6 36.0 28.8 32.7 39.4

A13 B6 A4 A14 B13

56.5 55.3 54.0 53.2 48.0

29.5 30.8 28.3 29.5 28.9

39.1 43.5 58.6 62.3 60.3

37.9 35.5 37.0 38.3 37.7

44.1 64.3 57.2 46.8 45.4

41.4 33.6 40.2 39.5 40.6

43.7 44.2 60.4 46.4 42.9

33.6 32.8 35.4 37.5 36.1

B5 B1 A7 B12 A1

46.2 42.0 41.6 39.0 32.5

29.2 28.1 34.9 28.9 27.4

54.5 31.0 45.8 57.9 33.9

39.5 35.9 43.2 38.3 38.2

61.2 34.1 38.2 67.1 33.2

36.0 39.0 33.6 38.6 34.2

47.4 32.6 26.9 49.1 35.6

32.2 35.8 26.2 39.0 36.0

A3 B11 A2 A11 B2

24.9 21.4 20.4 15.1 7.6

23.8 21.3 22.7 21.3 15.1

19.2 18.1 16.7 12.3 5.5

31.4 29.7 31.2 22.7 9.9

12.4 21.5 10.6 12.3 8.6

23.7 29.4 18.5 21.8 11.8

11.2 10.2 10.7 13.3 5.6

15.5 16.3 16.1 17.1 9.6

Statistical Analysis

Ratings data were analyzed using a linear mixed-effects model.2 The initial model included fixed effects of DIRECT OBJECT LENGTH, CORPUS MODEL 2 Additive models were also fit with by-variety smooth terms for DIRECT OBJECT LENGTH and

CORPUS MODEL PREDICTION, however the estimated degrees of freedom were between 1 and 2 for all these effects. Therefore we opted for simpler linear models for the final analyisis.

152

Experimental Corroboration

PREDICTION, as well as participant GENDER (male vs. female) and EDUCATION LEVEL (no university vs. university vs. postgraduate). Corpus model predictions were included as a control for possible effects of contextual factors other than direct object length, and for our corpus predictions we refit the corpus model with DIRECT OBJECT LENGTH removed, and recalculated predictions for our stimulus items with this second model. Thus the corpus predictions included in the analysis here did not incorporate information about the length of the direct object. The initial model also included interaction effects of each of these factors with VARIETY (BrE, NZE, SgE, IndE). For the VARIETY factor we used a custom coding to allow for three comparisons, which are illustrated in (57). (57) VARIETY comparisons in the model. a. Level 1: BrE and NZE versus SgE and IndE (Inner vs. Outer Circle) b. Level 2: BrE versus NZE c. Level 3: SgE versus IndE For the random effects, we began with a (near) maximal structure justified by the design, which included by-item and by-participant intercepts, as well as by-participant slopes for the effects of DIRECT OBJECT LENGTH and CORPUS P REDICTION , and a by-item slope for the effect of VARIETY . Models with by-item slopes for VARIETY failed to converge, and so this effect was removed from the model. Difference between models with and without the byparticipant slope for DIRECT OBJECT LENGTH was not significant based on a likelihood ratio test, suggesting that the effect of length did not vary much across participants. However we kept this effect as a check since DIRECT OBJECT LENGTH was an effect of particular interest. Finally, for the fixed effects structure, the interactions of VARIETY with the social factors GENDER and E DUCATION showed very little evidence of an effect and were also dropped from the final model. Results of the final model are shown in Table 7.2. Turning first to the fixed effect terms, we find a significant effect of VARIETY , though only when we compare the Inner and Outer Circle varieties (Level 1). In our model, participants in the Inner Circle rated the split variants significantly higher on average than participants in the Outer Circle (β = 5.78, t = 4.23, p < 0.001). Within Circle comparisons (BrE vs. NZE and SgE vs. IndE) did not reach significance. Thus BrE participants did not rate the split variant higher on average than NZE participants, and likewise IndE participants did not rate the split variant higher on average than SgE participants. Figure 7.6 shows the estimated ratings for the four participant groups, where higher scores represent greater preference for the split variant. This pattern is in line with the results from the corpus analysis, in which the split variant was significantly more likely overall in the Inner Circle data compared to

7.2 Results

153

Table 7.2 Summary statistics, fixed effects coefficients, and random effects standard deviations for linear mixed-model fitting participant rating as function of DIRECT OBJECT LENGTH, CORPUS MODEL PREDICTION, and participant VARIETY, GENDER, and EDUCATION LEVEL. Estimate

Std. Error

t value

p-value

3.125 5.776 −0.808 1.832

2.787 1.366 1.859 1.921

1.121 4.230 −0.434 0.954

0.270 < .001 0.664 0.341

−13.012 33.271 0.925 −0.710 −0.405

5.230 5.242 1.307 1.651 1.739

−2.488 6.347 0.707 −0.430 −0.233

0.019 < .001 0.480 0.667 0.816

Inner vs. Outer:DirObjLength BrE vs. NZE:DirObjLength SgE vs. IndE:DirObjLength

−6.012 −3.022 7.334

2.225 3.058 3.207

−2.702 −0.988 2.287

0.007 0.324 0.023

Inner vs. Outer:CorpusPred BrE vs. NZE:CorpusPred SgE vs. IndE:CorpusPred

1.424 −3.954 −0.308

2.103 2.847 3.071

0.677 −1.389 −0.100

0.499 0.166 0.920

Fixed effects (Intercept) Inner vs. Outer BrE vs. NZE SgE vs. IndE DirObjLength CorpusPred Gender = Male Education = Postgrad Education = No Uni.

Random effects item: (Intercept) id: (Intercept) id: by-DirObjWordLength id: by-CorpusPred Residual

Std.Dev. 14 6.52 7.39 5 30.56

Model summary Marginal R2 = 0.23, Conditional R2 = 0.397, LogLik = −18991.39, AIC = 38022.79

N = 3900

Outer Circle data. Neither of the two external factors we included, GENDER and EDUCATION level, showed any meaningful effect in our model. As for the internal predictors of interest, we find a significant main effect of CORPUS PREDICTION (β = 33.27, t = 6.35, p < 0.001), which shows a strong positive correlation with participant ratings, as we would expect. We also find a significant negative effect of DIRECT OBJECT LENGTH (β = −13.01, t = −2.49, p = 0.019), which again, fits our predictions. Participants’ preferences are aligned well with the corpus model predictions, and furthermore their ratings are sensitive to the length of the direct object, when taking the correlation with corpus model predictions – which are based on numerous other

154

Experimental Corroboration

Predicted rating (centered at 50)

Variety 10 5 0 −5 BrE

NZE

SgE

IndE

Figure 7.6 Partial effects plots of interaction of VARIETY on participant ratings (with CORPUS PREDICTION = 0, DIRECT OBJECT LENGTH = 0.4, GENDER = “female,” and EDUCATION = “university”). Positive ratings indicate greater preference for the split variant (pick the book up), negative ratings indicate greater preference for the continuous variant (pick up the book).

contextual factors – into account. Regarding the interaction of CORPUS PREDICTION with VARIETY , the statistical tests and partial effects plots (Figure 7.7) show no evidence of cross-varietal differences in the effect of CORPUS PREDICTION. On average, we find solid evidence that the correlations between the corpus model predictions and judgment ratings are fairly stable across the four varieties examined here. We do however, find evidence of cross-varietal differences in the effect of DIRECT OBJECT LENGTH, both with respect to difference between the Inner and Outer Circles (β = −6.01, t = −2.70, p = 0.007), and to the contrast between IndE and SgE (β = −7.33, t = 2.287, p = 0.023). From the partial effects plot (Figure 7.8) we can see that the effect of DIRECT OBJECT LENGTH is slightly stronger among BrE and NZE participants, as indicated by the steeper slope in the lines. Not surprisingly, the difference is largest with very short direct objects (the left side of the plot), and diminishes somewhat as the direct object gets longer. Such a pattern is typical of length effects in word order phenomena, which tend to be asymptotic. That is, beyond a certain length in the direct object, the probability of the split variant effectively falls to 0. We would expect such a pattern under the theory that length effects of this kind are largely driven by limitations in processing capacity (e.g. Gibson, 2000; Hawkins, 2004; Temperley, 2007; MacDonald, 2013). What is particularly puzzling in our data is the pattern of weight effects in the Singapore participants’ ratings. We find a significant interaction between DIRECT OBJECT LENGTH and variety (IndE vs. SgE), yet no difference

7.2 Results

155

NZE BrE SgE IndE

20 0 −20

Predicted rating (centered at 50)

−40

−1.0

−0.5

0.0

BrE 25

0

0

−25

−25 −1.0

−0.5

0.0

0.5

−50

SgE 25

0

0

−25

−25 −1.0

−1.0

−0.5

0.0

0.5

−0.5

0.0

0.5

IndE

25

−50

1.0

NZE

25

−50

0.5

−0.5

0.0

0.5

−50

−1.0

Corpus predicted log odds (z−score standardized)

Figure 7.7 Partial effects plots of interaction of VARIETY and CORPUS PREDICTION on participant ratings (with DIRECT OBJECT LENGTH = 0.4, GENDER = “female,” and EDUCATION = “university”). Top plot shown to allow for easier comparison across varieties. Positive ratings indicate greater preference for the split variant (pick the book up), negative ratings indicate greater preference for the continuous variant (pick up the book).

156

Experimental Corroboration 20 10

SgE 0

Predicted rating (centered at 50)

−10

−0.5

0.0

0.5

BrE 30 20 10 0 −10 −20 −30

−0.5

−0.5

1.0

NZE

0.0

0.5

1.0

30 20 10 0 −10 −20 −30

SgE 30 20 10 0 −10 −20 −30

NZE BrE IndE

−0.5

0.0

0.5

1.0

0.0

0.5

1.0

IndE

0.0

0.5

1.0

30 20 10 0 −10 −20 −30

−0.5

Number of words in direct object (z−score standardized)

Figure 7.8 Partial effects plots of interaction of VARIETY and DIRECT OBJECT LENGTH on participant ratings (with CORPUS PREDICTION = −0.19, GENDER = “female,” and EDUCATION = “university”). Top plot shown to allow for easier comparison across varieties. Positive ratings indicate greater preference for the split variant (pick the book up), negative ratings indicate greater preference for the continuous variant (pick up the book).

7.3 Discussion

157

between these varieties in their average ratings overall (β = 1.83, t = 0.95, p = 0.341). In other words, we do not find that the Singapore participants simply disfavored the split variant substantially more than the other groups, rather it seems that they were particularly insensitive to the difference in DIRECT OBJECT LENGTH, as indicated by the flatter line in Figure 7.8. This is illustrated in items such as B05 and B12, which have the longest direct objects in our stimulus set (58). (58)

a. Item B05 (Direct object = 37 letters): Figures from the Treasury last week suggested the policy was not raising the expected amount of revenue and was threatening to [drive leading business people and entrepreneurs away / drive away leading business people and entrepreneurs] from Britain. b. Item B12 (Direct object = 30 letters): We calculated that BSC spent something in excess of £200m a year just [moving around raw materials and finished products / moving raw materials and finished products around].

All things being equal, the continuous variant should be preferred with these items, yet our SgE participants rated the split variant considerably higher on average (61.2 and 67.1, respectively) than participants from the other countries (see Table 7.1). Curiously, we do find many examples in our SgE corpus data of split particle verbs with long direct objects, as in (59), but not with a noticeably higher frequency than in our data from other varieties.3 (59)

7.3

a. Even picked up enough energy to keep calling a female classmate up. b. On Saturday afternoon, Brendan Rodgers will resume his efforts to turn perennial sleeping giant Liverpool around . . . c. He was too busy with his business commitments, and devoted much of his time to taking his newly wed wife out in his Jaguar. Discussion

The experimental component to our study was designed to examine the degree to which the probabilistic constraints uncovered in prior corpus-based research are also discernible in language users’ introspective preferences. We focused on the particle placement alternation, which has been shown to exhibit differential effects of certain internal constraints across some varieties, and in particular across the Inner versus Outer Circle divide (Szmrecsanyi et al., 3 Informal conversations with British and American English speakers suggest that many such

examples also sound completely unmarked to speakers of Inner Circle varieties.

158

Experimental Corroboration

2016a; Grafmiller and Szmrecsanyi, 2018). We set out with three main research questions. RQ1 asked: Do participants exhibit a graded preferences for the use of a particular variant, similar to what we find in corpus models? RQ2 asked: Does a change in direct object length correlate with participants’ ratings for the split variant in the same manner as it does in a corpus model (i.e. longer direct objects result in lower ratings/probabilities)? RQ3 asked: Do cross-varietal differences in the probabilistic effects of direct object length in a corpus model correlate with cross-varietal differences in effects of those same factors on participants’ ratings? We answer each of these questions in the following sections. 7.3.1

Participants’ Ratings Align with Corpus Model Predictions

Averaging across participants, we find strong support for the hypothesis that when making metalinguistic judgments, participants are tapping into aspects of their linguistic knowledge that align with the quantitative distributions of alternation variants and corresponding contextual features. That is to say, we have good reason to believe that our corpus-based statistical models are capable of modeling probabilistic linguistic knowledge, even if they may not themselves represent realistic models of genuine human cognition (Divjak et al., 2016; Klavan and Divjak, 2016). Nonetheless, we do find some intriguing tendencies in our data that warrant further exploration. One avenue of investigation lies in the degree to which our participants’ choices agree with those of the model (see e.g. Klavan and Divjak, 2016). We can treat participants’ ratings above fifty as a “prediction” for the split variant, and ratings below fifty as a “prediction” for the continuous variant, and compare those to the predicted responses from our model. Averaging across varieties, we find the highest amount of agreement with the corpus model among British participants at 69.2 percent, and the lowest amount of agreement among those from Singapore, at 64.5 percent (Table 7.3). These scores are reasonably good, but there is certainly space where the model and participants disagree. If we look more closely at the individual items, we find that by and large, the greatest source of disagreement seems to be in those items that the corpus model gets wrong (Table 7.4). This is particularly striking in items B4 and A5, where 70–80 percent of participants rate the split variant as better, yet the model incorrectly predicts the continuous variant. These are shown in (60) in the form observed in the corpus. (60)

a. Item B04: If there is no suitable wall-port in your office, I can [wire one up / wire up one], but you may find it simpler to use WiFi.

7.3 Discussion

159

Table 7.3 Agreement scores reflecting the percentage of experimental items in which participants’ preferred variant matched the variant predicted by corpus model (100 percent indicates complete agreement between participants and corpus model). Variety British E New Zealand E Indian E Singapore E

% agree 69.2 67.4 65.4 67.2

b. Item A05: New in investment at Milton Keynes, Croydon and Manchester [helped profits up / helped up profits] from £30.3 million to £38.5 million. For these items the corpus model predicted the continuous variant with a small to moderate degree of certainty, yet a large majority of our participants agree that the corpus model got it wrong. These cases suggest that while the corpus model may be overall quite accurate, there are instances in which the model is lacking some important information, which our human raters are presumably keying into. At the same time, there are a fair number of items for which corpus and human agreement is negligible, that is, between only 40–65 percent. For many of these items the corpus model tends to make the correct prediction, while the human participants show only very weak preferences. This is comparable to similar studies comparing human and machine ratings, where it was found that participants in a forced choice classification task tended to be on average less accurate than corpus-based models (Klavan and Divjak, 2016; Klavan, 2020). Another direction of inquiry worth pursuing lies in the individual differences across participants. We observe that despite being given the opportunity to give a scalar, gradient measure of their preference, many participants simply moved the slider to either end of the scale, and/or left it at or close to the center position. Thus many participants seemed to treat the task as an “unforced” choice task between the two poles, where they also had the option to leave the slider at the neutral position. Similar alternation studies comparing a gradient rating task to a forced choice task suggest that both yield comparable results (Divjak et al., 2016), yet the possibility of “no choice” in our study presents an additional wrinkle not present in other forced choice experiments. It is worth noting here that participants can always assign a 50-50 rating in a 100-split task, yet

160

Experimental Corroboration

Table 7.4 Agreement scores reflecting the percentage to which participants’ preferred variant matched the variant predicted by corpus model (100 percent indicates complete agreement between participants and corpus model). Observed variant and corpus model correctness included for comparison. Item Mean rating Corpus model probability % agree Observed

Corpus model correct?

B2 A11 B8 A2 B11

6.9 13.2 89.1 15.1 18.1

0.034 0.001 0.921 0.025 0.002

99.2 94.0 92.1 91.8 89.7

Continuous Continuous Split Continuous Continuous

Yes Yes Yes Yes Yes

A3 B10 A6 A15 A9

17.6 87.8 86.5 78.8 80.4

0.202 0.756 0.614 0.831 0.634

89.6 88.9 88.8 84.3 84.3

Continuous Split Split Split Split

Yes Yes Yes Yes Yes

B9 A1 A10 B1 B7

76.9 33.8 73.4 34.7 66.8

0.934 0.058 0.864 0.192 0.888

81.7 76.1 75.4 71.4 66.7

Split Continuous Split Continuous Split

Yes Yes Yes Yes Yes

A8 A7 B15 A13 B3

65.4 39.4 58.7 45.2 47.9

0.920 0.407 0.695 0.367 0.287

65.7 64.2 61.9 56.7 56.3

Split Continuous Split Split Continuous

Yes Yes Yes No Yes

B13 A14 B5 B6 B12

49.3 53.8 53.0 52.4 54.4

0.370 0.542 0.217 0.868 0.119

54.8 52.2 50.0 49.2 46.0

Continuous Split Split Split Continuous

Yes Yes No Yes Yes

A12 B14 A4 A5 B4

58.4 56.7 57.6 64.1 76.7

0.497 0.374 0.486 0.428 0.311

44.8 42.9 42.5 34.3 20.6

Split Split Continuous Split Split

No No Yes No No

other studies do not report a high incidence of such ratings (Bresnan and Ford, 2010; Klavan and Divjak, 2016). It is possible that asking participants to use a slider bar rather than write down actual numbers made the “neutral” option more attractive, but we cannot say at this point. We note too that this is the first such study involving native and non-native speakers, and we know relatively little about how L2 speakers might approach this task vis-à-vis L1 speakers.

7.3 Discussion

161

L2 speakers may feel less confident or certain in their choices, and therefore may be more likely to leave the slider in the most “neutral” position. This too is a dimension in need of further research. Finally, the lack of a consistent choice for many items is itself useful information, as it may well be the case that participants simply do not have strong feelings one way or the other for some of these specific items. In such cases, the decision to leave the slider at the center may be a perfectly reasonable choice. Of particular interest to us would be those cases in which our participants express ambivalence, yet the corpus model predicts with confidence. For example, we have items B6 and B16 (61), which the corpus model correctly predicts with a probability of 0.87 and .88 (1 – 0.12) respectively, yet which our participants rate at 52 and 54 on average. (61)

a. Item B06: Threats were made about [taking the funds away / taking away the funds] from schools in this position. b. Item B12: We calculated that BSC spent something in excess of £200m a year just [moving around raw materials and finished products / moving raw materials and finished products around].

What the source of this discrepancy could be remains an area of further research. Regardless of the task though, there is growing acknowledgement that individual variation in grammatical knowledge and processing is sensitive to differences in experience with language (Dabrowska, ˛ 2015; Verhagen et al., 2018), and perhaps even more so among L2 speakers (Dabrowska, ˛ 2012). Moreover, the degree to which individual differences (e.g. in working memory, exposure, education, etc.) might measurably manifest themselves in online versus offline language behaviors is also not well understood (James et al., 2018). Such questions and issues extend well beyond the scope of this book, but they offer potentially quite fruitful opportunities for further research in this domain.

7.3.2

Sensitivity to the Length of the Direct Object Varies Across Varieties

As predicted, there was a noticeable negative correlation between the length of the direct object and the average ratings from all four varieties. This accords with the findings from our corpus investigation in Chapter 5, which found a strong influence of this factor in all four varieties. Again this is in line with the standard view of end-weight effects in English, which maintains that such effects are largely the result of the drive to minimize the cognitive burden

162

Experimental Corroboration Spoken monologue

Spoken dialogue Variety CanE BrE IndE SgE

0.6

0.4

Predicted probability of joined variant

0.2

0.0

Written informal

Written formal

0.6

0.4

0.2

0.0 2

4

6

2

Length of direct object in words

4

6

Figure 7.9 Predicted probabilities by DIROBJLENGTH (in words), register and VARIETY obtained from random forest model predicting the split variant (adapted from Szmrecsanyi et al., 2016a, figure 3).

incurred in planning, production and comprehension (Gibson, 2000; Hawkins, 2004; Temperley, 2008; MacDonald, 2013). It is also worth noting that the experimental results presented here are similar to those derived from a random forest analysis of a subset of the same ICE corpus data used in the present study (Figure 7.9; Szmrecsanyi et al., 2016a). Under a processing model of end-weight, floor effects such as these may be interpreted as a reflection of an upper threshold on users’ ability to manage long distance dependencies (see also Cappelle, 2009). As Szmrecsanyi et al. (2016a, 122) observe: We assume that speakers of all varieties are equally sensitive, on average, to the costs of accessing and processing longer constituents, and have roughly equivalent short-term memory capacities. Thus, it is not surprising to find what appears to be a consistent floor effect of end-weight across varieties . . . We predict therefore that to the extent regional differences in processing-driven factors exist, we should see them emerge only in those contexts where the processing load is relatively minimal.

7.3 Discussion

163

The results of our experimental rating task largely support this prediction. Our model reports a significant interaction of DIRECT OBJECT LENGTH with VARIETY, particularly in the comparison of Inner versus Outer Circle varieties. The former show significantly stronger effects of DIRECT OBJECT LENGTH, similar to what we find in the corpus data (Chapter 5, Figure 5.13). But this difference appears to be due mainly to the outlier SgE, in which we find the most attenuated effect of DIRECT OBJECT LENGTH (note that the IndE vs. SgE comparison is also significant. Curiously, however, it is in the instances of extremely long direct objects where SgE differs most from the other varieties (see Figure 7.8). Again, we cannot explain this pattern and we leave it as a topic for further research. 7.3.3

Cross-Varietal Differences in Length Effects Are Tricky to Replicate Across Tasks

At this point, one question we might naturally consider is why we might find differences across tasks. The corpus model and human ratings point to (slightly) different conclusions – the corpus model suggests that length effects are largely consistent across varieties, the experimental ratings say otherwise. At this point we are still unsure how to interpret these kinds of unexpected discrepancies, but for the time being we take these results as given, even if they lead to diverging conclusions. That is to say, until we have evidence to the contrary, we have no reason to think that either the corpus or experimental results are not showing us something “real.” At first blush, it may seem odd to suggest that that both methods could be accurately capturing true differences in (the grammars of) the populations we’re examining, but this is a possibility. In this regard, it is worth reiterating that the two methods rely on potentially very different kinds of behavioural data. Producing spoken or written language in a naturalistic setting, and making introspective judgments about specific choices are not the same tasks. Corpora comprise observational data of naturalistic language production, which is undoubtedly shaped by various psycholinguistic factors. Many such factors may or may not be measurable in a given instance. To what extent these same factors are operative in other kinds of linguistic behaviors (e.g. rating two or more prespecified options in an offline judgment task) remains unclear. Therefore, we cannot rule out the possibility that subtle differences in the two kinds of tasks could be behind our results here. The argument that we are not comparing like to like is not a new one, but as others have noted (Klavan and Divjak, 2016), the line between production and comprehension, both in speech and in reading/writing, is not as sharp as sometimes assumed (Tooley and Bock, 2014; Breen, 2014; Pickering and Gambi, 2018). Prediction is a key aspect of comprehension, and recent evidence suggests that prediction in comprehension is partly achieved through

164

Experimental Corroboration

covert imitation (or simulation) of language production (Martin et al., 2018; Pickering and Gambi, 2018). Under this view, participants rating (or choosing between) grammatical alternatives must be drawing on their implicit grammatical knowledge in a way that is sensitive to processes involved in producing those same structures, as has been argued for many other cognitive processes (Barsalou, 2009). Indeed, a particpants’ sense of what is “most natural” in a given context surely must be drawing upon something, and it seems entirely possible that this involves some internal, or even explicit, recitation of the target items. Finally, experimental alternation studies using online comprehension and production tasks find considerable overlap with results from rating tasks (Bresnan and Ford, 2010; Ford and Bresnan, 2013), offering solid empirical evidence for the validity of comparing these different kinds of data. Another intriguing, if speculative, possibility is that the effects of certain internal processing-based factors may in fact be more readily detected in (offline) comprehension tasks than in rapid online speech production. The reason is that tasks like the one used here do not force participants to commit to their first choice, thus allowing them more time to weigh the alternatives. In speech, once a choice is made (i.e. a variant is produced), it cannot be unmade. In (unspeeded) rating tasks participants have time to evaluate the items, and to a certain degree may be able to revise their responses. Given time to consider the “goodness” of two variants, where again part of that consideration process may involve implicit simulation of producing those variants, it is possible that users are more likely to arrive at the choice that more optimally satisfies various processing constraints. Notably this also applies to some extent to writing, where unlike speech, postchoice editing or revision is possible. Thus the emergence of certain processing-driven phenomena such as end-weight effects are not only to be expected in writing (e.g. Temperley, 2007) but it may well be that their influence is even stronger and/or more reliable than in speech. Lastly, we could be seeing evidence of a change in progress within our varieties. The GloWbE corpus was collected in the early 2010s, and the ICE corpora for these four varieties were compiled in the late 1990s and early 2000s. Thus there is a sizeable gap in time between when our corpus data were produced and when our ratings were collected. Whether the discrepancies we find are due to recent changes in one or more of these varieties is hard for us to say, but given the subtlety of the effects we find, it seems entirely plausible. Recent work has documented evidence of changes in progress with respect to particle placement in the Inner Circle (Haddican et al., 2021; Röthlisberger and Tagliamonte, 2020), but these changes largely apply to the base rates of individual variants. These studies find no evidence for changes in the associations of internal constraints over time, including the length of the direct object, so we are hesitant to attribute our results, at least not entirely, to language change. However, without comparable observational data from contemporary users or rating data from an earlier time, we cannot fully exclude change as a

7.3 Discussion

165

possible explanation. Nonetheless, the present study is valuable in that it points to an exciting direction for further research. One advantage of experimental methods is that they are relatively quick and easy to replicate, in contrast to compiling new large-scale corpora parallel to ICE or GloWbE. Experimentation makes investigation of changes in progress much more manageable, for instance allowing researchers to take an apparent time approach by sampling different ages of participants at a given time (we did not ask about participant ages in this study) or to use real time sampling at successive points in time. Such work is highly desirable and would be vital for situating our findings and others within a broader social context (Röthlisberger and Tagliamonte, 2020).

8

Where Are We Now, and Where to Next?

In the preceding chapters we have presented a large-scale investigation of the probabilistic constraints governing grammatical variation across nine international varieties of English. Our investigation comprised a series of studies examining three morphosyntactic alternations from distinct yet complementary vantage points. These studies were designed with a single overarching objective in mind: to examine the stability and plasticity of probabilistic grammar(s) across (standard) varieties of English around the globe. Our work was guided by a number of interrelated questions. On the descriptive plane, we seek to explore the degree of stability in the (sets of) constraints on these alternations – for a given alternation, how consistent are the probabilistic effects of the variable grammar’s constraints across varieties? Orthogonal to this are the questions of whether or to what extent some alternations may vary more than others with respect to their probabilistic conditioning, and of whether/to what extent some (types of) constraints may be more variable than others. Theoretically, we also hope to shed some light on questions of how and where to draw boundaries between distinct probabilistic variable grammars, which has been a recurrent theme in comparative sociolinguistics (Tagliamonte, 2001, 2013; Guy, 2015; Tamminga, 2019; Beaman and Guy, 2022) as well as dialectology and dialectometry (Nerbonne et al., 1999; Trudgill, 2011; Röthlisberger and Szmrecsanyi, 2019). We are also interested in why we find the patterns that we do. From a psycholinguistic perspective, we expect constraints that are (assumed to be) driven by deeper cognitive processes and principles, for example, accessibility biases, end-weight effects and structural persistence (Hawkins, 2004; MacDonald, 2013), to be cross-lectally robust even across L1 and L2 varieties (Kaan, 2014; Kaan and Chun, 2018). And to the extent that a given alternation is driven predominantly by such factors, we expect it to be relatively stable. In our experimental study (Chapter 7) we enquire about the extent to which the patterns we observe in corpus data can be replicated in other kinds of behaviors such as metalinguistic judgments. If it is true that the patterns we find in the corpus data accurately reflect the probabilistic grammatical knowledge within a given variety, we expect to find comparable patterns in participants’ preferences for the same data in experimental tasks. The corroboration of findings across 166

8.1 Empirical Summary

167

behavioral tasks thus lends greater validity to inferences about the grammars of these phenomena (Hoffmann, 2011; Ford and Bresnan, 2013; Divjak et al., 2016; Klavan and Divjak, 2016). At the same time, we are curious about the sociolinguistic and typological dynamics of these alternations, specifically the degree to which the crossvarietal patterns we find align with our current understanding of typological variation among English varieties. The present study is primarily an exploratory one in this regard, as it is not yet apparent to us how to devise direct tests of theories as complex as the Dynamic Model, at this level of abstraction (Gries et al., 2018; Hundt, 2020). Still, we believe it is worth reflecting some on the extent to which we can explain (at least partly) why we observe the patterns we do by appealing to existing models of World Englishes and dialect typology. More important, we are curious whether (post-hoc) exploration of these patterns, building upon psycholinguistic and socio-historical factors, can help us to develop testable predictions for future work. We view these questions about the global scope of morphosyntactic alternations through the lens of what we have called probabilistic indigenization, which we characterize as the process in which numerous internal and external linguistic forces shape the development of region-specific variable grammars. The investigation of probabilistic indigenization thus sits at the intersection of cognitive and sociolinguistic perspectives on variation (see also Geeraerts et al., 2010; Kristiansen and Geeraerts, 2013; Pütz et al., 2014). Finally, to address many of the questions above we had to overcome methodological challenges presented by the statistical analysis of multiple variables across multiple varieties. In essence, we had to address the challenge of how to synthesize these multiple dimensions of variability within a unified representation of meaningful cross-lectal patterns. A major aim of the project then was to develop a method for exploring variability among probabilistic grammars that could be applied to medium- and large-scale studies of varieties and alternation phenomena. The VADIS method (Chapter 6) builds upon established methods in comparative sociolinguistics while expanding our analytical toolkit to incorporate methods common to dialectology. 8.1

Empirical Summary

To accomplish these goals, we investigated three well-known grammatical alternations, the genitive, dative, and particle placement alternations, in nine varieties of English. As we discuss in Chapter 2, the groundwork for these investigations has been laid over decades of study of all three alternations within the (corpus-based) variationist tradition. The constraints governing these alternations are well-known and have been shown to vary to some degree regionally (Hinrichs and Szmrecsanyi, 2007; Bresnan and Hay, 2008; Haddican et al., 2021), yet there is very scant evidence that they serve as

168

Where Are We Now, and Where to Next?

indicators along any other social dimensions, such as age, sex, ethnicity, or class (Kendall et al., 2011; Jankowski and Tagliamonte, 2014). Moreover, these alternations have been relatively little studied in more than a handful of select varieties. Another appealing aspect of these phenomena is that they share a number of constraints between them (e.g. animacy, end-weight, information status), which allows us to better compare alternations to one another. The selection of conditioning factors in our studies was therefore motivated by what has been found in prior work on these alternations, as well as by psycholinguistic theories of language production (e.g. Bock and Warren, 1985; Hawkins, 2004; MacDonald, 2013). For the corpus analyses, we extracted tokens from components of the ICE and GloWbE corpora corresponding to the nine varieties investigated. Tokens were automatically retrieved from the corpora, and then subjected to further semi-automatic filtering to remove noninterchangeable tokens. We identified the envelope of variation for each alternation, defined as any context in which either variant would be acceptable and truth-conditionally equivalent. Only tokens occurring within the envelope of variation, that is, tokens that could in principle have been used in either constructional variant, were included in the study. The final dataset for the three alternations comprised approximately 38k tokens in all (Ngenitives = 13,798; Ndatives = 13,241; Nparticleverbs = 11,340). After data collection and filtering, tokens for each alternation were annotated for the various internal linguistic constraints known or hypothesized to influence the choice of variant in that alternation. The additional language external factor of text type (genre) was included as the corpora contain data from a number of different spoken and written registers. Procedures for data extraction and annotation are extensively documented in Chapter 4 and in supplementary materials available in our OSF repository. Analysis of the corpus data proceeded in two phases. In the first phase (Chapter 5), we analyzed each alternation individually, focusing on the overall rate of use, the relative importance of the different constraints to predicting variation, and the size and direction of those constraints’ effects. In terms of the variant rates across varieties, we found that in both the dative and particle placement alternations there is a noticeable tendency for Inner Circle varieties to pattern alike, with the ditransitive dative (Table 5.3) and split particle order (Table 5.5) being somewhat more frequent in BrE, CanE, NZE and IreE than in the other Outer Circle varieties (though the latter group was also internally more variable). Baseline rates of s-genitive usage were also slightly higher in Inner Circle varieties, although this is not as marked as in the other alternations (Table 5.1). Constraint importance was measured via importance scores based on alternation-specific conditional random forest models. When it comes to the constraint rankings, we found that of the three alternations, particle placement exhibits the most cross-lectal variation. In both the genitive and dative

8.1 Empirical Summary

169

alternations (Figures 5.2 and 5.6), we find the same few constraints (e.g. possessor animacy and constituent length, and relative length and recipient pronominality, respectively) tend to dominate in Inner Circle varieties, while these factors are more variable across the Outer Circle varieties. In the particle placement alternation, only the factor surprisal is consistently strong across all varieties (although direct object length is also fairly reliable). The language external factor of genre also appears to be unusually variable between the Inner and Outer Circles in this alternation (Figure 5.9). This is perhaps not surprising given that the split variant, and particle verbs more generally, tend to be a feature of informal language in these varieties. Genre seems to play relatively little role in predicting the other two alternations however. Regarding the impact of variety as a predictor in itself (measured as a nine-way factor or as a binary Inner versus Outer Circle factor), we found that this is relatively important only in the particle placement alternation, suggesting that the baseline rates in use of the genitive and dative variants are relatively consistent across varieties, once we take other internal linguistic factors into account. To better assess the direction and size constraint effects, we subjected the data to mixed-effects logistic regression analysis, in which we included by-variety interaction terms with our other constraints. Across all three alternations we found a high degree of stability in the directions of predictor effects. For instance, throughout our datasets animate possessors favor the sgenitive variant, pronominal recipients favor the ditransitive dative, and the more predictable a particle is given the verb the more likely the joined particle placement variant. Similar patterns obtain for other predictors of at least moderate importance. The probabilistic grammars regulating variant choice are thus qualitatively very similar across varieties. Again, this is entirely to be expected in that many of these constraints are hypothesized to be reflections of deeper principles of language processing. But while the effect directions are highly stable, quantitative differences in effect sizes were reported by our models. The chief findings from the regression analyses in Chapter 5 are repeated here: • The genitive alternation: – Animate possessors are more strongly associated with the s-genitive in BrE, compared to HKE and PhlE. – The presence of a final sibilant is less strongly associated with the of genitive in BrE, compared to CanE, HKE, and SgE. • The dative alternation: – Comparatively long recipients are more strongly associated with the prepositional dative in BrE compared to IndE. – Pronominal recipients are less strongly associated with the prepositional dative construction in BrE compared to IndE.

170

Where Are We Now, and Where to Next?

• The particle placement alternation: – The presence of a directional PP is more strongly associated with the split variant in Inner Circle varieties compared to Outer Circle varieties. – Compositional semantic meaning is more strongly associated with the split variant in Inner Circle varieties compared to Outer Circle varieties. As we noted, the constraints involved in these interactions run the gamut of semantic, syntactic, lexical, and phonological structural domains, and include both categorical constraints (e.g. pronominality vs. nonpronominality) and continuous constraints (e.g. weight ratio). In our data, it seems that everything is potentially regionally flexible. In the second phase of the corpus analysis, we provided a more bird’seye view of variation through the application of the Variation-based Distance and Similarity Modeling (VADIS) method (Szmrecsanyi et al., 2019). This method builds upon prior work in comparative sociolinguistics by triangulating three “lines of evidence” to evaluate the degree of relatedness among varieties (Poplack and Tagliamonte, 2001; Tagliamonte, 2013; Tagliamonte et al., 2016a). In a nutshell, VADIS offers a way of quantifying the similarity among variable grammars of two or more varieties by asking three questions: (i) Are the same constraints significant across varieties? (ii) Do the constraints have the same strength and direction across varieties? (iii) Is the relative importance of the constraints the same across varieties? We detailed the full VADIS analytical pipeline in Chapter 6. The end result of the method is a set of measures representing distances between varieties for each line of evidence applied to each alternation, which can be evaluated from various angles. For example, we showed how SIMILARITY COEFFICIENTS can be used to assess the degree of cross-varietal similarity for a given alternation. These similarity coefficients for individual lines or alternations can be further combined to assess an overall CORE GRAMMAR SCORE. In our case we found a high degree of similarity among our varieties with respect to these three alternations, particularly among the Inner Circle varieties. In terms of alternations, we observed a cline of decreasing cross-lectal stability from the genitive (most stable) to particle placement to finally the least stable dative alternation (see Table 6.6). This is roughly, though not perfectly, in line with our prediction that the more lexically specified an alternation is – how strongly associated the alternation is with concrete representations containing specific lexical items (Tamaredo et al., 2019, 17) – the more variability we should see in its cross-varietal probabilistic indigenization effects. Inspired by work on lectal coherence and cohesion (Guy and Hinskens, 2016), we also showed how the VADIS approach can be used to measure what we term DISTANCE-BASED COHERENCE (DBC). DBC scores reflect the degree to which the conditioning of variants in one alternation, medium, or line

8.2 Typological Universals

171

of evidence predicts cross-lectal similarities in other alternations, mediums, or lines. We found a high degree of coherence across spoken and written data (DBCmedium ), and some degree of coherence among alternations. The genitive and particle placement alternations tended to correlate with one another with regard to their cross-lectal probabilistic conditioning, that is, similarity in genitive conditioning predicted similarity in particle placement conditioning, but the dative alternation proved to be a rather stubborn outlier. Coherence among lines of evidence also turned out to be something of a mixed bag: the first and second lines, and the second and third lines tended to cohere for the genitive and particle placement alternations, but again not for the dative. Finally, we showed how VADIS similarities can be represented and explored visually via methods such as MDS and clustering. In the case of the genitive and particle placement alternation these diagrams regularly suggested a split among Inner and Outer Circle varieties, though yet again, no discernible patterning was found for the dative alternation. In the final phase of the study, we conducted a ratings experiment to corroborate our prior findings from the corpus data, and to explore one of the more intriguing prior findings, namely that of cross-varietal differences in the effect of direct object length on particle placement (Szmrecsanyi et al., 2016a; Grafmiller and Szmrecsanyi, 2018). We presented participants from four varieties (BrE, NZE, SgE, and IndE) a modified version of Bresnan et al.’s (2010) 100-split task, in which they saw excerpts from our corpora containing interchangeable particle verbs and were asked to rate their preference for the split versus joint variants in the provided context. Our experimental results suggest a small but significant difference in the effect of direct object length between the Inner and Outer circle varieties, and within the Outer Circle between the IndE and SgE participants. This pattern fits with what previous work had found (Szmrecsanyi et al., 2016a), where Inner Circle varieties showed stronger effects of direct object length than Outer Circle ones. Additionally, in our analysis of the corpus data in Chapter 5, we found some evidence of an interaction between variety and direct object length, though this only emerged when we discretized length as short versus long. Thus, we have some tentative evidence that cross-varietal differences in effect strength corroborate one another. 8.2

Typological Universals

As a point of departure for this section, we turn to the typological generalizations that emerge from our findings. At the most basic level, we might expect to see the kinds of robust typological patterns found for surface-level features (e.g. Szmrecsanyi and Kortmann, 2009; Kortmann and Lunkenheimer, 2012) mirrored in our distribution of probabilistic grammar-based measures. One conclusion we can confidently make is that in comparison to more concrete aspects of linguistic structure, where much of the cross-lectal diversification

172

Where Are We Now, and Where to Next?

is hypothesized to prevail (Schneider, 2007, 46), the probabilistic grammars of these alternations are remarkably stable across these varieties. But can we conclude that these alternations, individually or together, constitute genuine “angloversals” (Szmrecsanyi and Kortmann, 2009)? In other words, should the constraints governing the genitive, dative and particle placement alternation count as part of the “common core” of Standard English? On the one hand, the high degree of similarity measured by our VADIS coefficients would seem to say yes, especially for the genitive alternation. But these coefficients are based on a combination of measures, which raises the question of which specific aspects of our analyses, if any, we might point to concretely as descriptive angloversals. Recall for instance that the strength of the effect of possessor animacy, arguably the dominant factor in the genitive alternation (Rosenbach, 2014), varied significantly across some varieties, as did its relative importance. It seems we cannot say that the strength of the animacy effect on the genitive alternation, in absolute or relative terms, is universally shared across English lects. Similar problems apply to the other constraints and other alternations. In all, we find that no one constraint shows perfect cross-lectal consistency, and perhaps it is unreasonable to expect this to be so. At best it seems we can point to the direction of (some) effects as possible universals. For example, we can say that animate possessors favor the s-genitive to a considerable degree in all varieties we cover. Thus we might posit a range of angloversals that are probabilistic rather than categorical in nature. That is, when we focus on typological universals at the level of alternations, we are talking about universal patterns in the direction of statistical preferences or associations, whose exact strengths many vary, rather than the simple presence or absence of a feature. But this is not to say that the measures of absolute and relative strength, as assessed by the regression coefficients and random forest rankings, are not useful. To the extent that these do align across varieties we would argue that such alignment represents evidence of particularly high stability, and by implication, the likelihood of a robust sociohistorical contiguity and/or functional overlap between varieties. The VADIS method allows us to probe such patterns in depth, and in so doing constitutes a much-desired extension of the dialectometric tradition to linguistic structure beyond the surface level (e.g. Nerbonne et al., 1999; Kortmann and Lunkenheimer, 2012; Szmrecsanyi, 2013a). Such “probabilistic angloversals” are only descriptive generalizations, yet like other typological generalizations they cry out for explanation (Schneider, 2007, 88–90). The simplest explanation – that it is all merely random – cannot be fully discounted, but it is worth reflecting upon more compelling potential explanations. Why should we find greater stability and coherence among Inner Circle varieties compared to Outer Circle ones? The view from World Englishes theorizing is that varieties at more advanced stages of nativization tend to be more differentiated from Modern British English (Mukherjee and Gries, 2009).

8.3 Forces Driving Probabilistic Indigenization

173

The nativization process is defined linguistically by the emergence of locally distinct grammatical structures, which are the cumulative result of processes involving language contact, acquisition, sociolinguistic stratification, and general dialect drift (Schneider, 2007, 82). The Inner Circle varieties included here are situated firmly in Schneider’s Phase 5, while the Outer Circle ones all fall somewhere further down the Dynamic Model cline (see Chapter 3). Linguistically, the advanced phases manifest themselves in an increasing complexity and diversity of region-specific vocabulary, collocations, and morphosyntactic frames. In our case, however, it is the less advanced varieties that tend to be most dissimilar from one another and the Inner Circle varieties, which are themselves strikingly homogeneous. Thus, while the Dynamic Model would seem to predict that parent and child varieties should drift farther apart as a new variety proceeds along the cycle, this does not seem to be the case at the level of the abstract probabilistic grammar (see also Deshors and Gries, 2016; Röthlisberger et al., 2017; Hundt et al., 2021). Finally, we stress that the comparisons in our study are synchronic, and we do not assume that the source variety (BrE) or other Inner Circle varieties have simply remained unchanged since the early days of colonialization, while the Outer Circle alone has diversified (cf. Gries et al., 2018, 261–262). Even for BrE, there is ample evidence that none of our alternations has remained historically inert (e.g. Wolk et al., 2013; Rodríguez-Puente, 2016; Haddican et al., 2021). Similarities between our synchronic datasets should therefore not be taken to reflect internal stability within a given variety. We cannot speak to how much or how little the variable grammars in a given variety have changed over the past few centuries. Nonetheless, we believe the aggregate patterns across all our varieties taken together are hard to explain without some appeal to the kinds of synchronic and diachronic factors thought to shape other kinds of typological regularities. We consider what some of these factors might be in the next section. 8.3

Forces Driving Probabilistic Indigenization

Following Grafmiller and Szmrecsanyi (2018, 406–407), we suggest two explanatory perspectives for the prevalence of probabilistic grammar universals: shared histories and shared humanity. These two perspectives proceed naturally from usage-based theories of language structure, and embody the strands of sociohistorical and cognitive-psycholinguistic forces that we see as driving probabilistic indigenization. The shared history perspective appeals to the common ancestry of our varieties. Because these regional lects have largely developed from the same ancestral source, they have inherited many of the same syntactic forms, and very likely the same underlying grammatical systems. Along this dimension then we should expect to see primarily the effects of direct parent-to-child

174

Where Are We Now, and Where to Next?

transmission (Labov, 2007), or the lack thereof, on the development of new probabilistic grammars. That postcolonial Inner Circle varieties would tend to exhibit the highest degree of similarity with British English would seem to lend support to this explanation, as for example, Canadian and New Zealand English evolved from settlement colonies in which native speakers were directly transplanted from the home country. This is further bolstered by studies of L1 acquisition of grammatical alternations, which have consistently found considerable overlap between adults’ and children’s probabilistic grammars (Diessel and Tomasello, 2005; Gries, 2011; de Marneffe et al., 2012; Stephens, 2015). Shared history is also manifest in patterns of horizontal diffusion (Labov, 2007) between certain varieties, particularly those that have enjoyed relatively close contact with and exposure to one another. Diffusion is correlated with the length and depth of continued contact with Great Britain as these countries have developed postindependence, hence the similarities among British and Irish English for instance. Theoretically, the role of diffusion is particularly pertinent to Mair’s (2013) World Systems model where global influence is argued to propagate outward from the sociolinguistically central “hub” varieties (i.e. Standard British and American English). The prediction would be that greater cultural contact with, exposure to, and sociolinguistic orientation toward these hub varieties should correlate with greater similarity in our VADIS scores. Additionally, identity construction and orientation, which take center stage in the Dynamic Model, are also relevant to questions of diffusion, as close contact between communities/varieties does not necessarily lead to greater convergence. The often deep and complex sociolinguistic dynamics in a given society thus complicate our ability to make concrete predictions (Hundt, 2020), and as we noted before, the question of how to operationalize these complex social dynamics in a study such as ours is not straightforward. Shared humanity on the other hand, refers to the influence of theoretically universal biases in production and comprehension that influence language structure. Though differing in many respects, cognitive and usage-based linguistics theories all converge on the same basic idea that many of the typological patterns we observe across languages and phenomena are the cumulative result of the cognitive architecture that guides language processing (e.g. Gibson, 2000; Hawkins, 2004; Ferreira and Bock, 2006; Bybee, 2010; MacDonald, 2013; Goldberg, 2019). All language users are endowed with this architecture and are subject to the same psycholinguistic processes, therefore we expect similar general tendencies to predominate as structures that satisfy users’ biases become more frequent. For example, it has long been argued that English users’ preference for more efficiently processable dependencies leads them to place shorter units before longer ones in production (e.g. Gibson, 2000; Temperley, 2007; Futrell et al., 2015). MacDonald (2013) proposes two key principles of language production and comprehension that are especially relevant to our study: Easy First and Plan Reuse. Easy First postulates that

8.3 Forces Driving Probabilistic Indigenization

175

users will tend to select elements in production planning that are “easier” (e.g. more frequent or less complex) to retrieve from memory. The cross-lectal and cross-alternation stability in the effects of many of our constraints can plausibly be attributed to this principle (Behaghel, 1909; Bock and Warren, 1985; Rosenbach, 2005). Plan Reuse, on the other hand, represents a potentially countervailing force in that it emphasizes users’ tendency to reuse structures that they have used or experienced before (Szmrecsanyi, 2006; Gries and Kootstra, 2017; Schmid, 2017). In a given instance these forces may work with or against one another, such as when an Easy First order is not a frequently used one, and over time their joint action gives rise to the regular statistical patterns we discern in our datasets. From the usage-based view, these patterns are the accumulated effects of many, many instances of language use among many, many individuals, and form the statistical regularities that are the primary input for language users and learners. At the same time, users are continually innovating and thus the emergence of novel forms and functions will lead to variability in the quantitative associations between specific variants and features of the linguistic context. These cognitive biases and learned lexical/functional associations are themselves probabilistic, which leads to the development of the constrained variation in probabilistic indigenization that we observe in our study. Because language users of all ages are implicitly tracking (changes to) these statistical associations among words and contextual cues (Thothathiri and Rattinger, 2016; Perek and Goldberg, 2017), we fully expect that some cross-lectal variability should emerge among probabilistic grammars in just about anywhere we look. The push and pull of these various forces leads us to expect that variability in probabilistic grammars should be most conservative with respect to the sets of constraints and the direction of their effects, and most fluid with respect to their size and relative importance. This is exactly what has been found in numerous corpus-based variationist studies (e.g. Hinrichs and Szmrecsanyi, 2007; Bresnan and Hay, 2008; Bernaisch et al., 2014; Grafmiller, 2014; Deshors and Gries, 2016; Gries et al., 2018; Hundt et al., 2021). Returning to the question of universals, we propose then that similar effect strengths may be seen as examples of (probabilistic) “varioversals,” which are defined as features shared by varieties with similar histories and modes of acquisition (Szmrecsanyi and Kortmann, 2009, 33). It follows from our account that L1 transmission should be a major determinant of probabilistic indigenization, at least at this level of grammatical structure. Thus, we predict we should see strong similarities between our BrE, CanE and NZE varieties and other ENL/Inner Circle varieties, such as American or Australian English. And these similarities should be apparent not just in the direction of constraint effects, but in their strength as well. Crucially however, this should not be taken to imply that these varieties cannot diverge, possibly substantially, from one another or within themselves. Again, we do not expect perfect stasis,

176

Where Are We Now, and Where to Next?

rather it is a matter of degree. On the whole, we expect varieties with similar pathways of acquisition and transmission (e.g. those in the Inner Circle) to be more similar to each other than to other Outer/Expanding Circle varieties. However, this account should not be taken to imply that we therefore expect Outer Circle varieties to align more with Inner Circle ones as they shift from predominantly L2 to L1 acquisition of English. We elaborate on this last point next. 8.4

L1 Transfer and L2 Learning

So far we have considered the factors behind the stability of probabilistic grammars in our Inner Circle varieties. We now turn to the Outer Circle varieties and consider whether we can offer some explanation for the lack of stability we find among these by again appealing to common aspects of their histories and speakers. Simplifying somewhat, the Outer Circle varieties we examined here share a common evolutionary pathway – English in all these regions is, or was at some prior stage, predominantly acquired and used as a second language. It therefore seems plausible that the patterns we find in our data could be in part the result of influence from substrate languages in these regions and/or general biases in L2 acquisition (Williams, 1987). From a probabilistic grammar perspective, we are most concerned here with “indirect transfer” (Silva-Corvalán, 1994), in which the probability of a construction in English, such as the prepositional dative with a nominal recipient in IndE, is increased due to the prevalence of a similar construction in the L1, for example, light verb compounds in Hindi (see also MacWhinney, 1997; Heine and Kuteva, 2005). Identifying substrate effects at the level of probabilistic grammar poses a substantial challenge, especially when working with corpus data, yet any claims about universal tendencies among L2 varieties must ultimately confront the issue of L1 influence (Sharma, 2009). In this regard, there is a small but growing body of work pointing to evidence of just this kind of influence on English varieties (e.g. Sharma, 2009; Meriläinen et al., 2017; Hundt et al., 2021). As for our specific alternations, we are aware of but a few tantalizing studies that may shed some light on this question. For example, Heller (2018, 179–185) found that in HKE and JamE speakers used significantly more of genitives with animate possessors (62), contrary to the prevailing pattern in Inner Circle varieties. (62)

a. And as for your argument right you you mean argument about the style of uh D H Lawrence b. Oh so the mother of your lodger in Bristol in Liverpool c. It represents the good wishes of the people

8.4 L1 Transfer and L2 Learning

177

Heller notes that in these regions, the basilectal substrate languages almost exclusively use constructions with a possessor-possessum (s-genitive like) order to express possession, and he suggests that use of the contravening of genitive order may function as a salient marker of acrolectal English. The high prestige of (acrolectal) English is especially prominent in postcolonial societies, where acquisition of English has long been associated with high socioeconomic status. Unfortunately, while this suggests a (very) tentative account of our findings in HKE, it is hard to square with the differences we observe in PhlE and other varieties. In a more direct test of L1 transfer in the English genitive, Rosenbach (2017) showed that Afrikaans L2 English speakers used the s-genitive significantly more with inanimate possessors compared to BrE and White South African English L1 speakers. She attributed this to the fact that the comparable sgenitive-like construction in Afrikaans, the se genitive as in Harry se hart (“harry’s heart”), freely occurs with animate and inanimate possessors, unlike the English s-genitive. Rosenbach (2017, 17) highlights the potential impact of this finding: If effect strengths of constraints can carry over from one language to another one, this might have implications for shift- or ESL-varieties of English, where substrate languages are in close contact with English.

Without a doubt this is an area in need of further research. In our analyses of the dative alternation, IndE is the one variety that stood apart from the rest. There is a well-documented tendency for IndE to use the prepositional dative construction more than other varieties, and this has been attributed to the fact that give is commonly used as a light verb in IndE, as in 8.4. This use of give has been extended to contexts and uses not found in BrE or other Inner Circle varieties (Mukherjee and Hoffmann, 2006; Mukherjee and Gries, 2009; De Cuypere and Verbeke, 2013; Schilk et al., 2013). (63)

a. . . . explanatory hypothesis which should give explanation to [explain] the problem b. They are not giving development to [developing] their career that is my point (Mukherjee and Hoffmann, 2006, 155)

De Cuypere and Verbeke (2013, 180–182) note that these IndE forms mirror common compound forms in Hindi, suggesting that they can be seen as a kind of calque. The high degree of bi- and multilingualism among IndE users makes this direct calque explanation quite plausible. They further speculate that “the preference in IndE for a [prepositional dative] construction may be related to explicit case marking in Hindi or other Southern Indian languages” (De Cuypere and Verbeke, 2013, 182), however, strong evidence for such a relationship is lacking.

178

Where Are We Now, and Where to Next?

Figure 8.1 Partial effects plot of the interaction of VARIETY_OF_E with RECPRON in the dative alternation model. Probabilities (vertical axis) are for the prepositional dative construction. Vertical distance between dots is proportional to effect size. The plotted probabilities obtain when all other categorical constraints are set at their default levels, or at 0 in the case of numerical constraints. British English (shaded) serves as the reference variety in the underlying regression model. Varieties to the left are Inner Circle varieties; varieties to the right are Outer Circle varieties.

Still, we suggest that even these admittedly tentative connections may help us understand our findings for IndE. Recall that one of the largest differences between BrE and IndE is in the association of nonprominal recipients with the prepositional dative. In Figure 8.1, reproduced from Figure 5.7 (Chapter 5), we see that for IndE the probability of the prepositional dative is considerably higher (∼ 0.86) in IndE compared to BrE (∼ 0.67) when the recipient is a noun. Given the general preponderance of light verb “calques” in IndE, the frequency of the prepositional dative construction – and constructions superficially similar to it – is therefore much higher in this variety than others. This higher frequency could likely result in greater entrenchment of the prepositional dative construction (cf. Plan Reuse), such that in many cases in which a BrE user might prefer/use the ditranstive variant, for IndE users the ditranstive may be statistically preempted (Goldberg, 2019, 74–94) by the more common

8.4 L1 Transfer and L2 Learning

179

prepositional dative. We note also that this relates to the decreased effect of comparative length in IndE as well, since the strong bias against the prepositional dative construction with nominal recipients leaves less room for length to play a role in the alternation. Because most dative variation in IndE occurs only when the recipient is a pronoun, and hence very short compared to the theme, there is simply less opportunity for variability in length to function as a cue for this alternation. This can be seen from the substantially reduced importance of comparative length in the IndE random forest variable importance scores. In a nutshell, what we may be seeing in the probabilistic indigenization of the IndE dative alternation is the downstream effects of structural nativization at the lexis-syntax interface, which in this case is/was driven largely by influence from local substrate languages. Lastly, Röthlisberger (2018b, 190–191) also argues for the possibility of similar substrate effects in JamE, though we find little evidence of differences between BrE and JamE in our analysis. Finally, as to the particle placement alternation, we are not aware of any work demonstrating direct transfer of probabilistic cues across languages. We suspect that the strongest influence of substrate languages on particle placement may lie in the difficulty imbued by an L1 on the general acquisition of particle verbs. Although particle verbs are quite common and productive in (L1) English, they are notoriously challenging for L2 learners, especially those whose first language lacks phrasal verbs (Liao and Fukuya, 2004; Gardner and Davies, 2007; Gilquin, 2011; Tiv et al., 2019. Learners typically avoid using particle verbs when they can (Liao and Fukuya, 2004; Siyanova and Schmitt, 2007), and when learners do use them, they tend to overuse the continuous variant in contexts where native speakers might not use it (Paquot et al., 2019; Wulff and Gries, 2019). These tendencies persist even though L2 users are capable of learning the statistical associations between particle verb variants and contextual features, although these associations still may not match the rates of those for L1 users (Gilquin, 2015b,a). This results in a relatively impoverished particle verb input for users of even nativized L2 varieties such as those in the Outer Circle. At the same time, there is evidence that L2 users innovate with particle verbs in ways that are unusual to L1 users. For instance, Gilquin (2015a, 106) and Zipp and Bernaisch (2012, 198) found frequent use of the redundant particle up with verbs whose meaning already expressed upward movement or telicity (64). (64)

a. fifty percent of the class girls will be married by the time MA completes up b. The women participation in labour-force, more than doubled up between 1960–90.

It is worth noting that many of these forms are intransitive, and thus do not vary with respect to particle placement. While such innovation is found in all

180

Where Are We Now, and Where to Next?

varieties, the increased prevalence of these kinds of forms in Outer Circle varieties likely serves to further entrench the joined variant in users grammatical representations. Ultimately, the successful acquisition of a fully productive syntactic alternation requires a sufficient degree of type diversity among competing variants (Hoffmann, 2014; Perek and Goldberg, 2015, 2017). Insufficient variability leads to the failure to recognize the interchangeability between items, and hence the decreased likelihood that users will abstract over specific uses to arrive at a more schematic represenation (Cappelle, 2006; Gilquin, 2015b). In effect, when a given verb is observed in one variant, learners are more willing to infer that the same verb can also be used in the other variant if they have witnessed many other verbs alternating between the two – what Goldberg (2019, 62–72) refers to as “coverage” (see also Hoffmann, 2014; Gilquin, 2015b). Within the Outer Circle varieties then, the joint effects of general particle verb avoidance and the tendency toward regularization of the continuous variant have likely led to a reduction in the diversity of fully interchangeable particle verbs in these varieties. As Grafmiller and Szmrecsanyi (2018, 405) note: The net effect of this process is a high degree of variational asymmetry among individual verbs where one variant, in this case the split variant, becomes strongly associated with a much narrower range of verbs, leading to an overall dispreference for the split variant in the developing variety. Changes in the uses and associations of specific [verbs] in specific variants will likely lead to changes in the probabilistic associations among those variants and the higher-level constraints operative in speakers’ [probabilistic grammar].

Such changes will naturally be highly variable across varieties, and it remains to be seen whether we can identify any consistent trends in their impact on probabilistic indigenization. As we discussed above, it appears that change at the level of abstract probabilistic conditioning is relatively conservative in contexts of first language transmission, though even here we find variation among the Inner Circle varieties. Again, nothing is ever truly static. But it also seems that probabilistic indigenization is particularly sensitive to disruptions in the transmission process, such as when a variety passes through a kind of second language acquisition “filtering” stage, as happened in the Outer and Expanding Circles. The last thing to say here is that relatively little is known about the nature and history of ESL/EFL acquisition of the dative and genitive alternations in the Outer Circle, and this remains a major area for future research. 8.5

Implications for Variationist Sociolinguistics

A central goal of comparative variationist linguistics is determining to what extent two or more varieties of a language are grammatically “similar” (Tagliamonte, 2012, 2013). Comparative variationist research investigates to what extent language users’ grammatical choices depend on specific

8.5 Implications for Variationist Sociolinguistics

181

linguistic, social, and/or contextual constraints, and more importantly, to what extent those constraints influence choices differently across varieties. With this in mind, we now turn to consider our findings in the context of two prominent topics in variationist sociolinguistic theory: lectal coherence and covariation, and the nature of factors conditioning of variation at the individual and community level. 8.5.1

Coherence and Covariation in Morphosyntax: Looking Beyond the Vernacular?

Previously we noted the growing interest in the topic of lectal coherence and cohesion (see Section 6.5), where coherence is defined as the systematic covariation among linguistic variables (Guy and Hinskens, 2016, 1). The driving question in this domain is whether (sets of) linguistic features define varieties and lects, and relationships between them, in systematic ways. Put another way, can we use covariation in (socio)linguistic variables to define and delineate patterns among lects, similar to the way sets of surface-level features are used to draw isoglosses in traditional dialectology? The expectation is that the “orderly heterogeneity” (Weinreich et al., 1968) of (socio)linguistic variation should give rise to parallel (i.e. coherent) patterning of features within a speech community. This vision of orderly diversity implies that speech communities are sociolinguistically coherent, in the following sense: the orderly variables that define the community should collectively behave in parallel: variants (or rates of use of variants) that index a given style, status, or a social characteristic should co-occur. (Guy and Hinskens, 2016, 2)

Recent work in this domain has painted a rather complicated picture, with some finding relatively weak coherence even among similar social variables (e.g. Guy, 2013; Waters and Tagliamonte, 2017), while others (e.g. Oushiro, 2016) do find evidence of coherence among structurally unrelated variables (for a recent review see Tamminga, 2019). The question is then: do our alternations exhibit such parallel behavior across regions or registers, and if so why? If not, why not? In their introduction to a special issue in Lingua on cohesion and covariation, Guy and Hinskens (2016, 4) pose a number of important questions facing theorists working on the topic. The three most relevant to our study are: (i) Which features correlate and which do not? To what extent, and in what ways, do the characteristic variables associated with a dialect or a speech community covary? [G&H question 1] (ii) When correlated usages are encountered, are they better understood as indicators of the social coherence of a variety, or as (perhaps inevitable) consequences of structural factors or historical factors? [G&H question 6] (iii) Are some kinds of language varieties (e.g. local dialects) more coherent than others (e.g. speech styles)? If so, why? [G&H question 7]

182

Where Are We Now, and Where to Next?

Regarding the first question, we find that only two of our alternations, the genitive and particle placement alternation, tend to covary to any meaningful degree. Again the dative alternation is correlated with neither. We did not expect this given our lexical specificity hypothesis, which predicts that grammatical alternations that are strongly associated with (i.e. occur very frequently with) specific lexical items should be the most likely to exhibit cross-varietal differences in their constraint effects. Furthermore, our results do not accord with Tamaredo et al. (2019), who did find the expected genitive > dative > particle placement cline of covariation. Based on this, we expected to see the greatest coherence among genitives and datives, but that is not what we found. Why this should be is difficult to determine. The literature on inter-speaker covariation, and by implication intra-lectal covariation, suggests that correlations are most likely to occur among structurally related variables (Tamminga, 2019, 121), and the three alternations we’ve studied here are not structurally related in the traditional sense. They are three distinct constructions and therefore the lack of covariation is perhaps unsurprising. Nonetheless, all three variables are strongly influenced by similar conditioning factors, which we expect to guide any changes to these variables in similar directions. Still, there is room for variation to emerge, as we find in our data. Further investigation of covariation, ideally at the level of individual users and/or specific registers, is needed. In Sections 8.3 and 8.4 above we addressed the topic of the second question at length, arguing that the major typological divisions we find are largely the consequences of common historical patterns and structural effects of L1 and L2 acquisition. But Guy and Hinskens’ (2016) point about the indexing of social characteristics is worth further thought, as the social salience of variables may play a crucial role in driving specific patterns of covariation (Guy, 2013; Tamminga, 2019; Haddican et al., 2022). Aside from possible registerspecific stylistic differences, there is little evidence that the genitive, dative, or particle placement alternations index any broad social characteristics we commonly find with other sociolinguistic variables (Kendall et al., 2011; Jankowski and Tagliamonte, 2014; Haddican et al., 2021). Our alternations have simply not reached the level of social saliency to even be considered sociolinguistic indicators in any standard varieties. It remains to be seen whether any of these alternations will ever cross the threshold of social significance. Given the deep history of these variables, it seems unlikely – we would expect it to have happened somewhere by now – but the possibility exists. Finally, we find that coherence is greatest when we compare spoken versus written modes. This is not surprising, as structural differences in registers work on all three variables in similar ways. For example, in formal writing inanimate referents tend to be more frequent and personal pronouns less so, lexical items tend to be more diverse, and constituents tend to be longer and more complex (Biber and Conrad, 2009). At the same time, registers impose constraints on

8.5 Implications for Variationist Sociolinguistics

183

structure in other ways, based on different aspects of their communicative purposes, such as the drive for greater economy of expression in certain written registers. We expect the probabilistic factors to interact with the communicative functions of different registers in similar ways, and we expect these factors to vary between registers in somewhat parallel fashion across varieties. Again, since these structural, rather than social, constraints are the dominant factors governing our alternations, the high degree of register-internal coherence makes sense. So what can we conclude? It seems for morphosyntactic variables lacking social significance, lectal coherence is relatively weak along social dimensions, e.g. region, ethnicity, class (although some degree of coherence is still present). As Kendall et al. (2011) observe, not all linguistic variables are sociolinguistic variables, and language variation is subject to different forces at different levels of organization. The three alternations examined here do seem to exhibit the (formal) stylistic variation typical of sociolinguistic markers (Labov, 1972), yet they exhibit a lack of social awareness typical of indicators. Ultimately though, they show no evidence of the social grouping that defines sociolinguistic variables in general. To the extent that we do find coherence among such alternations, we therefore expect it to be strongest within lects delineated more by their communicative function (Biber and Conrad, 2009) than by any other social characteristics. That is to say, we predict coherence among morphosyntactic alternations such as these should be greatest within registers, at least from the perspective of large-scale regional varieties. If we are on the right track here, then what is needed within the larger variationist sociolinguistic field is a greater attention to and appreciation of register and register variation (see Szmrecsanyi, 2019). This is especially important for the study of World Englishes, in which the complex dynamics of diglossia and languagebased social stratification loom large. In such contexts register variation can provide fertile soil out of which social variation may grow, as register-specific variants can be recruited for sociosemiotic purposes, such as when a more informal/basilectal variant is used to signal a “casual” or “authentic” style, or when formal/acrolectal variants come to index higher education and social status (e.g. Campbell-Kibler, 2007; Eckert, 2008). Only by looking beyond the vernacular to the full scope of language variation can we hope to untangle the many theoretical and methodological questions raised by Guy and Hinskens (2016) and others (e.g. Tamminga, 2019; Beaman and Guy, 2022). 8.5.2

Conditioning of Variable Grammars: How Stable Are Constraints Really?

Returning to the results of our individual alternation analyses (Chapter 5), recall that we found that almost any constraint could and did vary across varieties. Many of these constraints involve the kinds of features associated with

184

Where Are We Now, and Where to Next?

the elements of linguistic representation that are commonly viewed in sociolinguistics as internal linguistic factors, what Tamminga et al. (2016a) term “i-conditioning.” I-conditioning is summarized as “the effects that elements of linguistic representation in the environment surrounding and containing an instance of a variable can have on that variable’s realization” (Tamminga et al., 2016a, 304). Such factors in our study would include the presence of a final sibilant on a possessor or the semantic relation expressed by a genitive construction, the definiteness or pronominality of the theme and recipient in the dative alternation, and the presence of a following directional PP in the particle placement alternation. Effects of these factors are distinguished from so-called “s-conditioning,” which involves the association of variables with various social-stylistic dimensions (Tamminga et al., 2016a, 304). Aside from register, s-conditioning plays only a very minor role in the three alternations we examine. Because i-conditioning (and s-conditioning) is tied to aspects of linguistic representations that must be learned from the environment, its effects can be language and variety-specific. Therefore we would predict that to the degree that morphosyntactic alternations vary across lects, they should vary primarily with respect to their i-conditioning factors. This is indeed what we find in our individual regression models: the strongest interaction effects tend to be those involving i-conditioning factors. Curiously though, interactions in our regression models also implicate effects of psycholinguistic factors governing language production and comprehension, such as end-weight effects (e.g. Wasow and Arnold, 2003; Hawkins, 2004; Temperley, 2007), the tendency to produce more accessible elements first (e.g. Bock and Warren, 1985; Prat-Sala and Branigan, 2000; Branigan et al., 2008; MacDonald, 2013), and the tendency to reuse similar structures (e.g. Ferreira and Bock, 2006; Bock et al., 2007; MacDonald, 2013; Branigan and Pickering, 2017). Tamminga et al. (2016a) designate such effects as (cognitive) “p-conditioning,” which are argued to be “architecturally distinct” from i-conditioning. The variability of such factors in our study is intriguing because it is generally assumed that we should not see much cross-lectal variation in this regard, given that p-conditioning is deeply rooted in the human cognitive architecture. We expect to find p-conditioning across all similar phenomena in all varieties, exerting a constant or at least predictably-distributed effect on all individuals (although interactions with i- and s-conditioning could complicate this simple picture in practice). Such an expectation does not hold for i-conditioning. (Tamminga et al., 2016a, 324)

Our results are at least qualitatively consistent with this kind of account of language production, where variation is governed in part by language users’ implicit drive to minimize processing load during production planning. But the scope of quantitative variability in these effects is still not well-understood. Pconditioning is thus closely connected to the notion of probabilistic universals

8.5 Implications for Variationist Sociolinguistics

185

discussed above. From a probabilistic grammar view on dialect typology, we expect that to the extent we can identify genuine angloversals, they will be most robust at the level of p-conditioning, and this should hold of individuals as well as communities. On the other hand, i- and s-conditioning effects are much more likely to vary and diversify as varieties evolve. At the macro-varietal level of standard language at least, the degree of diversity among varieties will surely be proportional to the extent of sociohistorical closeness among varieties as well as shared patterns of acquisition. Thus we would argue that the strongest structural evidence for the existence of varioversals among probabilistic grammars lies in the shared patterning among i- and s-conditioning factors. However, the notion of “universal” p-conditioning effects must not be not be taken to mean completely invariant (Tamminga et al., 2016a, 325). Aspects of grammatical structure and learned statistical patterns interact with production planning in complex ways, which can manifest as variability in the effects of p-conditioning factors. For instance, strong biases for particular variants in certain contexts – learned from exposure – in a given variety may lead to a measurable reduction in the importance of p-conditioning effects, as in the case of the dative alternation in IndE (see Section 8.4). Context can further play a potential role in shaping perceived variability as p-conditioning effects (e.g. priming or end-weight) are modulated situationally in response to shifts in register, medium, audience, etc. There is also the question of how to determine what counts as an i- versus p-conditioning effect. Tamminga et al. (2016a) view i-conditioning and p-conditioning as distinct domains of linguistic behavior, with only the former constituting genuine aspects of grammar. The dividing line is the ability of a given factor to condition categorical alternations, in addition to probabilistic (variable) ones. That is, when an “alternation is conditioned in ways that are not attested in categorical alternations, the conditioning is p-conditioning (or s-conditioning), and not i-conditioning” (Tamminga et al., 2016a, 322). While conceptually appealing, this distinction is hard to draw in practice. Effects of frequency, constituent length, and structural priming are compelling candidates for p-conditioning, but for other factors the picture is much less clear. A good illustration of the fuzziness around this distinction is the case of animacy effects in grammatical alternations. As we have shown, the effect of animacy on the English genitive alternation is neither categorical nor crosslectally invariant. However, there is ample evidence from numerous languages for categorical effects of animacy on case-marking, word order, or other grammatical phenomena (e.g. de Swart et al., 2008; Malchukov, 2008). This would imply then that we should view the variable animacy effects we find in English as cases of i-conditioning. Complicating this view is the fact that probabilistic animacy effects reported across many phenomena over and over again seem to point in the same direction: animate referents tend to be accessed and integrated into structures earlier

186

Where Are We Now, and Where to Next?

in production planning than inanimate referents (e.g. Hundt, 2004; Kempen and Harbusch, 2004; Branigan et al., 2008; Bresnan and Hay, 2008; Tanaka et al., 2011; Verhoeven, 2014; Hundt et al., 2021). We find this too in our study, where the size of the effect varied, but its overall direction remained constant. The ubiquity of this tendency across many languages and phenomena is therefore often taken to be a reflection of universal aspects of human psychology (Rosenbach, 2002; Branigan et al., 2008). It has even been argued that certain conceptual domains such as animacy are part of our evolved cognitive organization (Caramazza and Shelton, 1998; Mahon and Caramazza, 2009). Thus we could make a reasonable argument for treating animacy effects as examples of p-conditioning, whose cumulative impact has gradually become fixed as a categorical rule in some languages. Furthermore, processing driven (p-conditioning) effects can function categorically, as in the limits on center embedding (Tamminga et al., 2016a, 323), thus the existence of categorical patterns with respect to animacy is not itself evidence for treating probabilistic animacy effects as i-conditioning. Similar arguments could be made for constraints related to definiteness, expression type, and information structure. We believe examples like this raise many questions about the role of different kinds of conditioning in language variation and change. If we treat a given factor such as animacy as a case of i-conditioning, then it follows that this effect should be free to vary across languages and phenomena in potentially arbitrary ways. This arbitrariness implies that we could see even a direct reversal of a given effect. And yet we do not see this with animacy effects. It is possible that i-conditioning may interact with other factors to constrain pathways of variation, giving the appearance of universality, but evidence suggests that animacy effects function independently of other factors (e.g. Rosenbach, 2005). Animacy effects appear to be quite robust crosslinguistically, hence the suspicion that they are tied to something deeply rooted in human cognition. We are thus left to wonder, if a particular instance of i-conditioning is (was) ultimately derived from p-conditioning, what does that mean for the possible development of the variable? Can the (emergent) i-conditioned rules subsequently change in any direction? If so, why should the p-conditioning forces cease to affect the variable in the same ways they did before? If not, what does that imply for the separation of grammar and use (Tamminga et al., 2016a, 326–329)? To what extent might we view p-conditioning as a stabilizing force governing grammatical rules and grammatical variation more generally? These are difficult questions, to which we do not yet have satisfactory answers. To our mind the jury is still out regarding how best to view the internal linguistic system, that is, as a unified system or a bifurcated model dividing grammar from usage, though we tend to see the former as the more parsimonious view (see Grafmiller et al., 2018). What we can say at this point is that a great deal more work is needed to understand how to disentangle different potential causes of variation in practice.

8.6 The Road Ahead

8.6

187

The Road Ahead

We close our discussion with some further reflections on the various challenges facing large-scale projects of the kind we have presented here. At beginning of this book we noted the growing popularity of large-scale variationist studies of morphosyntactic alternations, and a primary motivation for our work has been to draw connections between subfields of linguistics that we believe would benefit from greater cross-communication. We sketched a picture of variation in probabilistic grammars across different native and nonnative varieties of English that incorporated ideas and perspectives from variationist sociolinguistics, World Englishes studies, and dialect typology. We further drew upon cognitive and psycholinguistic research as well as work on L1 and L2 acquisition to make sense of the patterns we found in our data and to provide some understanding of the possible causes driving the probabilistic indigenization of morphosyntactic alternations in new varieties. The empirical picture was derived using a combination of methodological techniques developed over the years by comparative sociolinguistics and dialectometry, two subfields that share related goals but until now have rarely crossed paths. Looking forward, we believe the VADIS method can provide a means for exploring a number of empirical and theoretical questions in ways that have heretofore been difficult. By incorporating techniques from variationist and dialectometric traditions, the method allows us to simultaneously evaluate the degree of internal lectal coherence among structurally distinct sociolinguistic variables, as well as explore the typological landscape of those variables across varieties. We found in our study that coherence was strongest among registers rather than regions, at least for the three grammatical alternations we looked at, yet we also found compelling evidence for a Inner versus Outer Circle divide for the genitive and particle placement alternations. Thus far, however, the VADIS method has not been applied to variables at other levels of language structure, for example, phonetics and vernacular usage, where most of the sociolinguistic “action” is (Labov, 1972; Preston and Niedzielski, 2010b; Tagliamonte, 2012). It remains to be seen what the method might reveal when extended to more dynamic and socially salient variables, and to other usage contexts. But the complementary quantitative aspects of the method can provide potential insight into the kinds of linguistic phenomena that can be help us define lects across social, geographical, and/or functional domains. Identifying and understanding the patterns the VADIS method may reveal is an exciting challenge for future work. Another challenge for variationist researchers involves probing the degree to which complex patterns of variation might persist across linguistic behaviors. Recent years have seen an increasing desire to validate or corroborate findings from corpus data with experimental studies (e.g. Bresnan and Ford, 2010; Ford and Bresnan, 2013; Klavan and Divjak, 2016; Engel et al., 2022). Assuming

188

Where Are We Now, and Where to Next?

that the patterns observed in production data reflect learned statistical associations stored as part of users’ grammatical knowledge, cross-lectal effects are predicted to replicate between different tasks. This largely seems to be the case, but subtle deviations nonetheless do emerge, as in our study (see Chapter 6.5.3) and others (Divjak et al., 2016). The deeper question remains, what can we make of differences we observe in studies that use very different methods to examine related phenomena. We say “related” here because producing spoken or written language in a naturalistic setting, and making introspective judgments about specific choices are not necessarily the same tasks. We agree that a pluralistic approach is superior to a single-minded corpus-based or experimental one, however, it is far from clear at this juncture how exactly we should compare results obtained from these very different kinds of linguistic data. Klavan and Divjak (2016, 378) for instance, stress the importance of using “behavioral data [i.e. judgments] to evaluate [a corpus] model,” yet producing an utterance is itself a particular kind of behavior, and arguably a far more natural one than rating subtly different expressions on some numerical scale in a lab or online survey. Corpora comprise observational language data produced in a naturalistic setting (or range of settings), and analyses based on such data thus primarily investigate aspects related to language production. Such data are invaluable for addressing any number of questions, including we would argue, questions about the factors that shape language users’ choices in a given context. Crucially, these are choices about language production, and such production is undoubtedly shaped by various psycholinguistic factors, many of which may or may not be measurable in a given instance. To what extent these same factors are operative in the same way and to the same degree in other kinds of linguistic behaviors, such as rating tasks, is less wellunderstood. Thus, while we take to heart concerns raised in recent work about the (over)reliance on corpus models in cognitive and variationist linguistics, we see no reason to privilege one source of data over another as automatically providing a more accurate window into grammatical structure. The preceding discussion touches on yet another hot topic in (variationist) sociolinguistics and cognitive linguistics, which is the growing interest in individual variation, and the relation between grammar at the level of the individual and the community (Tamminga et al., 2016a; Dabrowska, ˛ 2016b; Schmid et al., 2020). The nature of the data in our study precluded the investigation of individual patterns, as the corpora do not contain sufficient data to allow comparison of individual speakers/writers. But in principle any researcher with sufficient data could avail themselves of the methods employed here. Our methods are well-suited for the kinds of work represent by, for example, Tamminga (2019), and could be especially fruitful for the comparison of sociophonetic variables among speakers. For studies of grammatical variation, the challenge lies in the amount of individual-level data necessary for making inter-user comparisons. Most corpora, even very large ones, do not typically contain enough

8.6 The Road Ahead

189

instances of a given morphosyntactic alternation from individual users for us to apply the VADIS method reliably. Furthermore, grammatical features are also difficult to elicit naturally in sociolinguistic interviews. This is where experimental methods could be a tremendous boon, as recent work suggests that experimental elicitation of alternation variants can be quite successfully used as a proxy for corpus data (Ford and Bresnan, 2015). Such experimental methods could allow researchers to examine covariation and probabilistic conditioning among individuals in both production and comprehension. Whether these kinds of experiments can be adequately scaled up to provide sufficient data for the VADIS method is doubtful – one would need dozens, even hundreds of trials per participant – but absent the collection of very large and very narrowly focused corpora of individual usage, we see little alternative for studying inter-user variation in the kinds of grammatical phenomena investigated here. On the theoretical side, we also note that many of the explanations for indigenization and variation that we have proposed above are still rather speculative. This is perhaps inevitable in a study of this scope, where we cannot possibly delve into the complex histories of each variety and its (socio)linguistic details. We have tried to step back and draw connections at macro-varietal scale between our work and the tremendous body of research on these nine varieties of English (and many others) around the globe, yet we fully recognize the need for more empirical work on the finer specifics of the genitive, dative, and particle placement alternations in each variety. Such work would include not only more thorough investigation of the historical development of these varieties (if sufficient data exists, see e.g. Hundt and Szmrecsanyi, 2012; Gries et al., 2018) but also a deeper look into how these alternations are acquired and how their use develops among both L1 and L2 users (e.g. Gries, 2011; de Marneffe et al., 2012; Paquot et al., 2019). Beyond this, focused investigation of the lexical and semantic diversity of these alternations, within and across lects, could shed additional light on the particular pathways of nativization observed in these phenomena. The present study has assumed that the semantic scope of these alternations, from both onomasiological and semasiological perspectives, is consistent across varieties, yet this may be not the case. New variants expressing the “same” meaning may emerge in different varieties, and existing variants may be extended to novel functions for reasons related to L2 acquisition, language contact, or general drift, as we know has happened in some varieties (see Zipp and Bernaisch, 2012; Schilk et al., 2013). Such facts potentially rattle the uneasy consensus in variationist sociolinguistics regarding the semantic equivalence of grammatical variables (Lavandera, 1978; Sankoff, 1988), and complicate attempts to compare variables across varieties. The bright side, however, is that the field of World Englishes research, by virtue of its embrace of corpus-linguistic methods and long-standing focus on

190

Where Are We Now, and Where to Next?

the development of region-specific lexico-grammatical variation (e.g. Mukherjee and Hundt, 2011; Hundt and Gut, 2012; Collins, 2015), is expertly positioned to provide the necessary theoretical and empirical perspectives to deal with these issues. The discussion in this chapter, and indeed the entire book, has aimed to make connections between different perspectives clearer. As a final word on the generalizability of this kind of variationist research, we note that there is still much we do not know about the overall variability in these kinds of phenomena, especially when we take into account multiple varieties and registers. For numerous reasons, different studies rarely focus on the same corpora, registers, or even the same varieties, and it is therefore hard to evaluate the reliability of our findings. While researchers can confidently point to a number of robust results regarding certain factors’ influence on a growing number of phenomena, e.g. the influence of animacy on the genitive alternation, the strength of the cross-varietal differences (i.e. the interactions) observed in these factors are typically quite small and subtle. What we lack is a set of replication studies or meta-analyses of this kind of research on grammatical alternations. In the absence of suitable prior studies against which we might compare our findings for a given phenomenon, we are often left with more questions than answers. What this means is that CVL researchers cannot be sure just how robust some of the interaction effects may be in our corpus models or experimental studies. Understanding the nature of the factors conditioning variation will surely shed some light on this question, but achieving this will require the consistent, theoretically motivated operationalization of features in our models. Nor can we be certain of the degree to which our analyses of community-level patterns may reflect aspects of grammatical knowledge at the level of individuals without much more detailed user-level data. With all this in mind, we believe the road ahead leads to many exciting opportunities, but it is also fraught with many obstacles and potential garden paths. Undeterred, we proceed for now under the assumption that we are on firm ground, but we would do well to remember that we are only just learning to walk, and that we are bound to misstep from time to time.

References

Aldenderfer, Mark S., and Blashfield, Roger K. 1984. Cluster Analysis. Newbury Park, London: Sage Publications. Alejo-González, Rafael. 2010. Making Sense of Phrasal Verbs: A Cognitive Linguistic Account of L2 Learning. AILA Review, 23(Dec.), 50–71. Allen, Cynthia L. 2006. Case Syncretism and Word Order Change. Pages 201–223 of van Kemenade, Ans, and Los, Bettelou (eds.), The Handbook of the History of English. Chichester: Wiley-Blackwell. Altenberg, Bengt. 1982. The Genitive v. the of-Construction: A Study of Syntactic Variation in 17th Century English. Lund Studies in English. Lund: Gleerup. Anderwald, Lieselotte. 2003. Negation in Non-Standard British English: Gaps, Regularizations and Asymmetries. Abingdon: Routledge. Anderwald, Lieselotte. 2005. Negative Concord in British English Dialects. Pages 113– 138 of Iyeiri, Yoko (ed.), Aspects of English Negation. Amsterdam: Benjamins. Anderwald, Lieselotte. 2016. Language between Description and Prescription: Verbs and Verb Categories in Nineteenth-Century Grammars of English. Oxford Studies in the History of English. New York: Oxford University Press. Anderwald, Lieselotte, and Szmrecsanyi, Benedikt. 2009. Corpus Linguistics and Dialectology. Pages 1126–1140 of Lüdeling, Anke, and Kytö, Merja (eds.), Corpus Linguistics. An International Handbook, vol. 2. Berlin: Mouton de Gruyter. Angermeyer, Philipp S., and Singler, John Victor. 2003. The Case for Politeness: Pronoun Variation in Co-Ordinate NPs in Object Position in English. Language Variation and Change, 15(02), 171–209. Arnold, Jennifer E., Losongco, Anthony, Wasow, Thomas, and Ginstrom, Ryan. 2000. Heaviness vs. Newness: The Effects of Structural Complexity and Discourse Status on Constituent Ordering. Language, 76(1), 28–55. Baayen, R. Harald. 2008. Analyzing Linguistic Data: A Practical Introduction to Statistics Using R. Cambridge: Cambridge University Press. Baayen, R. Harald, Endresen, Anna, Janda, Laura A., Makarova, Anastasia, and Nesset, Tore. 2013. Making Choices in Russian: Pros and Cons of Statistical Methods for Rival Forms. Russian Linguistics, 37(3), 253–291. Baldwin, Timothy, and Villavicencio, Aline. 2002. Extracting the Unextractable: A Case Study on Verb-Particles, vol. 20. Association for Computational Linguistics. Bao, Zhiming, and Wee, Lionel. 1999. The Passive in Singapore English. World Englishes, 18(1), 1–11.

191

192

References

Barsalou, Lawrence W. 2009. Simulation, Situated Conceptualization, and Prediction. Philosophical Transactions of the Royal Society of London: Biological Sciences, 364, 1281–1289. Barth, Danielle, and Kapatsinski, Vsevolod. 2017. A Multimodel Inference Approach to Categorical Variant Choice: Construction, Priming and Frequency Effects on the Choice between Full and Contracted Forms of Am, Are and Is. Corpus Linguistics and Linguistic Theory, 13(2), 1–58. Bates, Douglas, Mächler, Martin, Bolker, Benjamin M., and Walker, Steven C. 2015. Fitting Linear Mixed Effect Models Using Lme4. Journal of Statistical Software, 67(1), 1–48. Bauer, Laurie, and Bauer, Winifred. 2002. Can We Watch Regional Dialects Developing in Colonial English? The Case of New Zealand. English World-Wide, 23(2), 169–193. Bauer, Laurie, and Warren, Paul. 2008. New Zealand English: Phonology. Pages 39–63 of Burridge, Kate, and Kortmann, Bernd (eds.), Varieties of English 3: The Pacific and Australasia. Berlin: Mouton de Gruyter. Beaman, Karen V., and Guy, Gregory R. 2022. The Coherence of Linguistic Communities: Orderly Heterogeneity and Social Meaning. New York: Routledge. Beaman, Karen V., Buchstaller, Isabelle, Fox, Susan, Walker, James A., and Cheshire, Jenny (eds.). 2021. Advancing Socio-Grammatical Variation and Change: In Honour of Jenny Cheshire. Routledge Studies in Sociolinguistics. Abingdon: Routledge. Behaghel, Otto. 1909. Beziehungen Zwischen Umfang Und Reihenfolge von Satzgliedern. Indogermanische Forschungen, 25, 110–142. Bernaisch, Tobias, Gries, Stefan Th., and Mukherjee, Joybrato. 2014. The Dative Alternation in South Asian English(Es): Modelling Predictors and Predicting Prototypes. English World-Wide, 35(1), 7–31. Bhatt, Rakesh M. 2004. Indian English: Syntax. Pages 1016–1030 of Mesthrie, Rajend (ed.), A Handbook of Varieties of English, vol. 2: Morphology and Syntax. Berlin: Mouton de Gruyter. Biber, Douglas, and Conrad, Susan. 2009. Register, Genre, and Style. Cambridge Textbooks in Linguistics. Cambridge: Cambridge University Press. Biber, Douglas, and Egbert, Jesse. 2018. Register Variation Online. Cambridge: Cambridge University Press. Biber, Douglas, Johansson, Stig, Leech, Geoffrey, Conrad, Susan, and Finegan, Edward. 1999. Longman Grammar of Spoken and Written English. Harlow: Longman. Biewer, Carolin. 2015. South Pacific Englishes: A Sociolinguistic and Morphosyntactic Profile of Fiji English, Samoan English and Cook Islands English. Varieties of English Around the World Amsterdam: Benjamins. Bock, J. Kathryn. 1986. Syntactic Persistence in Language Production. Cognitive Psychology, 18(3), 355–387. Bock, J. Kathryn, and Warren, Richard K. 1985. Conceptual Accessibility and Syntactic Structure in Sentence Formulation. Cognition, 21, 47–67. Bock, Karen, Dell, Gary S., Chang, Franklin, and Onishi, Kristine H. 2007. Persistent Structural Priming from Language Comprehension to Language Production. Cognition, 104(3), 437–458. Bod, Rens, Hay, Jennifer, and Jannedy, Stefanie (eds.). 2003. Probabilistic Linguistics. Cambridge, MA: MIT Press. Bolinger, Dwight. 1971. The Phrasal Verb in English. Cambridge, MA: Harvard University Press.

References

193

Box, George E. P. 1979. Robustness in the Strategy of Scientific Model Building. Pages 201–236 of Launer, Robert L., and Wilkinson, Graham N. (eds.), Robustness in Statistics. Academic Press. Brandner, Ellen. 2012. Syntactic Microvariation: Syntactic Microvariation. Language and Linguistics Compass, 6(2), 113–130. Branigan, Holly P., and Pickering, Martin J. 2017. Structural Priming and the Representation of Language. Behavioral and Brain Sciences, 40, 1–61. Branigan, Holly P., Pickering, Martin J., and Tanaka, Mikihiro. 2008. Contributions of Animacy to Grammatical Function Assignment and Word Order during Production. Lingua, 118(2), 172–189. Breen, Mara. 2014. Empirical Investigations of the Role of Implicit Prosody in Sentence Processing. Language and Linguistics Compass, 8(2), 37–50. Breiman, Leo. 2001. Random Forests. Machine Learning, 45, 5–32. Bresnan, Joan. 2007. Is Syntactic Knowledge Probabilistic? Experiments with the English Dative Alternation. Pages 75–96 of Featherston, Sam, and Sternefeld, Wolfgang (eds.), Roots: Linguistics in Search of Its Evidential Base. Berlin: Mouton de Gruyter. Bresnan, Joan, and Ford, Marilyn. 2010. Predicting Syntax: Processing Dative Constructions in American and Australian Varieties of English. Language, 86(1), 168–213. Bresnan, Joan, and Hay, Jennifer. 2008. Gradient Grammar: An Effect of Animacy on the Syntax of Give in New Zealand and American English. Lingua, 118(2), 245–259. Bresnan, Joan, Cueni, Anna, Nikitina, Tatiana, and Baayen, Harald. 2007. Predicting the Dative Alternation. Pages 69–94 of Boume, Gerlof, Krämer, Irene, and Zwarts, Joost (eds.), Cognitive Foundations of Interpretation. Amsterdam: Royal Netherlands Academy of Science. Britain, David. 2002. Diffusion, Levelling, Simplification and Reallocation in Past Tense BE in the English Fens. Journal of Sociolinguistics, 6(1), 16–43. Britain, David. 2010. Grammatical Variation in the Contemporary Spoken English of England. Pages 37–58 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. Abingdon: Routledge. Britain, David. 2018. Dialect Contact and New Dialect Formation. Chapter 8, pages 143–158 of The Handbook of Dialectology. Chichester: John Wiley & Sons, Ltd. Brook, Marisa, and Tagliamonte, Sali A. 2016. Why Does Canadian English Use Try to but British English Use Try and ? Let’s Try and/to Figure It Out. American Speech, 91(3), 301–326. Bruthiaux, Paul. 2003. Squaring the Circles: Issues in Modeling English Worldwide. International Journal of Applied Linguistics, 13(2), 159–178. Brutt-Griffler, Janina, and Samimy, Keiko K. 2001. Transcending the Nativeness Paradigm. World Englishes, 20(1), 99–106. Brysbaert, Marc, and New, Boris. 2009. Moving beyond Kuˇcera 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(4), 977–990. Buchstaller, Isabelle, and Corrigan, Karen. 2011. How to Make Intuitions Succeed: Testing Methods for Analysing Syntactic Microvariation. Pages 30–48 of Maguire, Warren, and McMahon, April (eds.), Analysing Variation in English. Cambridge: Cambridge University Press. Buschfeld, Sarah. 2014. English in Cyprus and Namibia. Pages 181–202 of Buschfeld, Sarah, Hoffmann, Thomas, Huber, Magnus, Kautzsch, Alexander, and Schneider,

194

References

Edgar W. (eds.), The Evolution of Englishes: The Dynamic Model and Beyond. Varieties of English around the World. Amsterdam: Benjamins. Buschfeld, Sarah, and Kautzsch, Alexander. 2017. Towards an Integrated Approach to Postcolonial and Non-Postcolonial Englishes. World Englishes, 36(1), 104–126. Buschfeld, Sarah, and Kautzsch, Alexander. 2019. Theoretical Models of English as a World Language. Pages 51–71 of Schreier, Daniel, Hundt, Marianne, and Schneider, Edgar W. (eds.), The Cambridge Handbook of World Englishes, first edn. Cambridge Handbooks in Language and Linguistics. Cambridge: Cambridge University Press. Buschfeld, Sarah, and Kautzsch, Alexander (eds.). 2020. Modelling World Englishes: A Joint Approach to Postcolonial and Non-Postcolonial Varieties. Edinburgh: Edinburgh University Press. Buschfeld, Sarah, Hoffmann, Thomas, Huber, Magnus, and Kautzsch, Alexander. 2014. The Evolution of Englishes The Dynamic Model and beyond. Introduction. Pages 1– 20 of The Evolution of Englishes: The Dynamic Model and Beyond. Varieties of English around the World. Amsterdam: Benjamins. Bybee, Joan. 2006. From Usage to Grammar: The Mind’s Response to Repetition. Language, 82(4), 711–733. Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: Cambridge University Press. Campbell, Aimee, and Tomasello, Michael. 2001. The Acquisition of English Dative Constructions. Applied Psycholinguistics, 22, 253–267. Campbell-Kibler, Kathryn. 2007. Accent, (Ing), and the Social Logic of Listener Perceptions. American Speech, 82(1), 32–64. Campbell-Kibler, Kathryn. 2010. Sociolinguistics and Perception. Language and Linguistics Compass, 4(6), 377–389. Cappelle, Bert. 2005. Particle Patterns in English: A Comprehensive Coverage. Ph.D. thesis, KU Leuven. Cappelle, Bert. 2006. Particle Placement and the Case for “Allostructions”. Constructions, Special Volume 1, 1–28. Cappelle, Bert. 2009. Contextual Cues for Particle Placement: Multiplicity, Motivation, Modeling. Pages 145–192 of Bergs, Alexander, and Diewald, Gabriele (eds.), Constructional Approaches to Language. Amsterdam: Benjamins. Caramazza, Alfonso, and Shelton, Jennifer R. 1998. Domain-Specific Knowledge Systems in the Brain: The Animate-Inanimate Distinction. Journal of Cognitive Neuroscience, 10(1), 1–34. Cedergren, Henrietta J., and Sankoff, David. 1974. Variable Rules: Performance as a Statistical Reflection of Competence. Language, 50(2), 333. Chambers, J. K. 2012. Homogeneity as a Sociolinguistic Motive in Canadian English. World Englishes, 31(4), 467–477. Chambers, Jack K. 2004. Dynamic Typology and Vernacular Universals. Pages 127– 145 of Kortmann, Bernd (ed.), Dialectology Meets Typology. Berlin: Mouton de Gruyter. Chambers, Jack K., and Trudgill, Peter. 1998. Dialectology. Second edn. Cambridge: Cambridge University Press. Chen, Ping. 1986. Discourse and Particle Movement in English. Studies in Language, 10(1), 79–95. Cheshire, Jenny. 2005. Syntactic Variation and beyond: Gender and Social Class Variation in the Use of Discourse-New Markers. Journal of Sociolinguistics, 9(4), 479–508.

References

195

Childs, Claire. 2017. Integrating Syntactic Theory and Variationist Analysis: The Structure of Negative Indefinites in Regional Dialects of British English. Glossa: a journal of general linguistics, 2(1), 106. Chomsky, Noam, and Halle, Morris. 1968. The Sound Pattern of English. Cambridge, MA: MIT Press. Christie, Pauline. 2003. Language in Jamaica. Kingston: Arawak. Claes, Jeroen. 2016. Cognitive, Social, and Individual Constraints on Linguistic Variation: A Case Study of Presentational, ‘Haber’ Pluralization in Caribbean Spanish. Cognitive Linguistics Research. Berlin: Mouton de Gruyter. Claridge, Claudia. 2000. Multi-Word Verbs in Early Modern English: A Corpus-Based Study. Language and Computers. Amsterdam: Rodopi. Clarke, Sandra. 2010. Newfoundland and Labrador English. Dialects of English. Edinburgh: Edinburgh University Press. Cohen, Jacob. 1992. A Power Primer. Psychological Bulletin, 112(1), 155–159. Collins, Peter. 1995. The Indirect Object Construction in English: An Informational Approach. Linguistics, 33, 35–49. Collins, Peter (ed.). 2015. Grammatical Change in English World-Wide. Studies in Corpus Linguistics. Amsterdam: Benjamins. Cuyckens, Hubert, D’hoedt, Frauke, and Szmrecsanyi, Benedikt. 2014. Variability in Verb Complementation in Late Modern English: Finite vs. Non-Finite Patterns. Pages 182–204 of Late Modern English Syntax. Cambridge: Cambridge University Press. Cysouw, Michael. 2013. Disentangling Geography from Genealogy. In Auer, Peter, Hilpert, Martin, Stukenbrock, Anja, and Szmrecsanyi, Benedikt (eds.), Space in Language and Linguistics. Berlin: Mouton de Gruyter. Dabrowska, ˛ Ewa. 2012. Different Speakers, Different Grammars: Individual Differences in Native Language Attainment. Linguistic Approaches to Bilingualism, 2(3), 219–253. Dabrowska, ˛ Ewa. 2015. Individual Differences in Grammatical Knowledge. Pages 650–668 of Dabrowska, ˛ Ewa, and Divjak, Dagmar (eds.), Handbook of Cognitive Linguistics. Berlin: Mouton de Gruyter. Dabrowska, ˛ Ewa. 2016a. Cognitive Linguistics’ Seven Deadly Sins. Cognitive Linguistics, 27(4), 479–491. Dabrowska, ˛ Ewa. 2016b. Language in the Mind and in the Community. In Daems, Jocelyne, Zenner, Eline, Heylen, Kris, Speelman, Dirk, and Cuyckens, Hubert (eds.), Change of Paradigms – New Paradoxes. Berlin: Mouton de Gruyter. Daelemans, Walter, and van den Bosch, Antal. 2005. Memory-Based Language Processing. Studies in Natural Language Processing. Cambridge: Cambridge University Press. D’Arcy, Alexandra. 2020. The Relevance of World Englishes for Variationist Sociolinguistics. Pages 436–458 of Schreier, Daniel, Schneider, Edgar W., and Hundt, Marianne (eds.), The Cambridge Handbook of World Englishes. Cambridge Handbooks in Language and Linguistics. Cambridge: Cambridge University Press. D’Arcy, Alexandra, and Tagliamonte, Sali A. 2015. Not Always Variable: Probing the Vernacular Grammar. Language Variation and Change, 27(03), 255–285. D’Arcy, Alexandra, Haddican, Bill, Richards, Hazel, Tagliamonte, Sali A., and Taylor, Ann. 2013. Asymmetrical Trajectories: The Past and Present of – Body/–One. Language Variation and Change, 25(3), 287–310. Davies, Mark. 2013. Corpus of Global Web-Based English: 1.9 Billion Words from Speakers in 20 Countries. https://www.english-corpora.org/glowbe

196

References

Davies, Mark, and Fuchs, Robert. 2015. Expanding Horizons in the Study of World Englishes with the 1.9 Billion Word Global Web-based English Corpus (GloWbE). English World-Wide, 36(1), 1–28. Davydova, Julia. 2019. Quotative like in the Englishes of the Outer and Expanding Circles. World Englishes, 38(4), 578–592. De Cuypere, Ludovic. 2015. The Old English to-Dative. English Language and Linguistics, 19(1), 1–26. De Cuypere, Ludovic, and Verbeke, Saartje. 2013. Dative Alternation in Indian English: A Corpus-Based Analysis. World Englishes, 32(2), 169–184. de Marneffe, Marie-Catherine, Grimm, Scott, Arnon, Inbal, Kirby, Susannah, and Bresnan, Joan. 2012. A Statistical Model of the Grammatical Choices in Child Production of Dative Sentences. Language and Cognitive Processes, 27(1), 25–61. De Smet, Hendrik, D’hoedt, Frauke, Fonteyn, Lauren, and Van Goethem, Kristel. 2018. The Changing Functions of Competing Forms: Attraction and Differentiation. Cognitive Linguistics, 29(2), 197–234. de Swart, Peter, Lamers, Monique, and Lestrade, Sander. 2008. Animacy, Argument Structure, and Argument Encoding. Lingua, 118(2), 131–140. De Troij, Robbert. in preparation. North and South, bottom to top. Using big data to model syntactic variation in Belgian and Netherlandic Dutch. Ph.D. thesis, KU Leuven. De Vaus, D. A. 2002. Analyzing Social Science Data. London Sage Publications. Dehé, Nicole. 2002. Particle Verbs in English: Syntax, Information Structure, and Intonation. Amsterdam: Benjamins. Deshors, Sandra C. 2014. A Case for a Unified Treatment of EFL and ESL: A Multifactorial Approach. English World-Wide, 35(3), 277–305. Deshors, Sandra C., and Gries, Stefan Th. 2016. Profiling Verb Complementation Constructions across New Englishes: A Two-Step Random Forests Analysis of Ing vs. to Complements. International Journal of Corpus Linguistics, 21(2), 192–218. Deterding, David. 2010. Singapore English. Edinburgh: Edinburgh University Press. Deuber, Dagmar. 2014. English in the Caribbean: Variation, Style and Standards in Jamaica and Trinidad. Studies in English Language. Cambridge: Cambridge University Press. Dewaele, Jean-Marc. 2018. Why the Dichotomy “L1 Versus LX User” Is Better than “Native Versus Non-native Speaker.” Applied Linguistics, 39(2), 236–240. Diessel, Holger, and Tomasello, Michael. 2005. Particle Placement in Early Child Language: A Multifactorial Analysis. Corpus Linguistics and Linguistic Theory, 1, 89–112. Divjak, Dagmar, and Gries, Stefan Thomas (eds.). 2012. Frequency Effects in Language Representation. Trends in Linguistics: Studies and Monographs. Berlin: Mouton de Gruyter. Divjak, Dagmar, Dabrowska, Ewa, and Arppe, Antti. 2016. Machine Meets Man: Evaluating the Psychological Reality of Corpus-Based Probabilistic Models. Cognitive Linguistics, 27(1), 1–33. Dollinger, Stefan. 2019. English in Canada. Pages 52–69 of Nelson, Cecil L., Proshina, Z. G., Davis, Daniel R., Kachru, Braj B., and Kachru, Yamuna (eds.), Handbook of World Englishes, 2nd edition. Chichester: Wiley-Blackwell. Dryer, Matthew S., and Haspelmath, Martin (eds.). 2013. WALS Online. Leipzig: Max Planck Institute for Evolutionary Anthropology.

References

197

Dubois, Tanguy, Paquot, Magali, and Szmrecsanyi, Benedikt. 2022. Alternation phenomena and language proficiency: the genitive alternation in the spoken language of EFL learners. Corpus Linguistics and Linguistic Theory. Eckert, Penelope. 2008. Variation and the Indexical Field. Journal of Sociolinguistics, 12(4), 453–476. Edwards, Alison. 2017. ICE Age 3: The Expanding Circle: ICE Age 3: The Expanding Circle. World Englishes, 36(3), 404–426. Edwards, Alison, and Laporte, Samantha. 2015. Outer and Expanding Circle Englishes: The Competing Roles of Norm Orientation and Proficiency Levels. English WorldWide, 36(2), 135–169. Ehret, Katharina, Wolk, Christoph, and Szmrecsanyi, Benedikt. 2014. Quirky Quadratures: On Rhythm and Weight as Constraints on Genitive Variation in an Unconventional Data Set. English Language and Linguistics, 18(02), 263–303. Elenbaas, Marion B. 2013. The Synchronic and Diachronic Status of English Light Verbs. Linguistic Variation, 13(1), 48–80. Ellis, Nick C., Römer, Ute, and O’Donnell, Matthew Brook (eds.). 2016. UsageBased Approaches to Language Acquisition and Processing: Cognitive and Corpus Investigations of Construction Grammar. Language Learning Monograph Series. Chichester: Wiley-Blackwell. Engel, Alexandra, Grafmiller, Jason, Rosseel, Laura, and Szmrecsanyi, Benedikt. 2022. Assessing the complexity of lectal competence: the register-specificity of the dative alternation after give. Cognitive Linguistics. Evans, Stephen. 2014. The evolutionary dynamics of postcolonial Englishes: A Hong Kong case study. Journal of Sociolinguistics, 18(5), 571–603. Fang, Yu, and Liu, Haitao. 2021. Predicting Syntactic Choice in Mandarin Chinese: A Corpus-Based Analysis of Ba Sentences and SVO Sentences. Cognitive Linguistics, 32(2), 219–250. Ferreira, Victor S., and Bock, Kathryn. 2006. The Functions of Structural Priming. Language and Cognitive Processes, 21(7–8), 1011–1029. Filppula, Markku. 2004. Irish English: Morphology and Syntax. Pages 73–101 of Kortmann, Bernd, Schneider, Edgar, Burridge, Kate, Mesthrie, Raj, and Upton, Clive (eds.), A Handbook of Varieties of English, vol. 2: Morphology and Syntax. Berlin: Mouton de Gruyter. Filppula, Markku, Klemola, Juhani, and Paulasto, Heli. 2009. Vernacular Universals and Language Contacts: Evidence from Varieties of English and Beyond. London: Routledge. Fischer, Olga, and van der Wurff, Wim. 2006. Syntax. Pages 109–198 of Hogg, Richard, and Denison, David (eds.), A History of the English Language. Cambridge: Cambridge University Press. Ford, Marilyn, and Bresnan, Joan. 2013. Using Convergent Evidence from Psycholinguistics and Usage. Pages 295–312 of Krug, Manfred, and Schlüter, Julia (eds.), Research Methods in Language Variation and Change. Cambridge: Cambridge University Press. Ford, Marilyn, and Bresnan, Joan. 2015. Generating Data as a Proxy for Unavailable Corpus Data: The Contextualized Sentence Completion Task. Corpus Linguistics and Linguistic Theory, 11(1), 187–224. Fox, John. 2003. Effect Displays in R for Generalised Linear Models. Journal of Statistical Software, 8(15), 1–27.

198

References

Franco, Karlien, and Tagliamonte, Sali A. 2020. New -Way(s) with -Ward(s): Lexicalization, Splitting and Sociolinguistic Patterns. Language Variation and Change, 32(2), 217–239. Fraser, Bruce. 1965. An Examination of the Verb-Particle Construction in English. Unpublished Ph.D. thesis, Massachusetts Institute of Technology. Fraser, Bruce. 1976. The Verb-Particle Combination in English. New York: Academic Press. Futrell, Richard, Mahowald, Kyle, and Gibson, Edward. 2015. Large-Scale Evidence of Dependency Length Minimization in 37 Languages. Proceedings of the National Academy of Sciences, Aug., 201502134. Gardner, Dee, and Davies, Mark. 2007. Pointing out Frequent Phrasal Verbs: A CorpusBased Analysis. TESOL Quarterly, 41(2), 339–359. Gardner, Matt Hunt. 2017. Grammatical Variation and Change in Industrial Cape Breton. Ph.D. thesis, University of Toronto. Gardner, Matt Hunt, Denis, Derek, Brook, Marisa, and Tagliamonte, Sali A. 2020. Be like and the Constant Rate Effect: From the Bottom to the Top of the S-Curve. English Language and Linguistics, 25(2), 281–324. Gardner, Matt Hunt, Uffing, Eva, Van Vaeck, Nicholas, and Szmrecsanyi, Benedikt. 2021. Variation isn’t that hard: Morphosyntactic choice does not predict production difficulty. PLOS ONE, 16(6), e0252602. Garretson, Gregory, O’Connor, M. Catherine, Skarabela, Barbora, and Hogan, Marjorie. 2004. Coding Practices Used in the Project Optimality Typology of Determiner Phrases. http://npcorpus.edu/documentation/index.html Geeraerts, Dirk, Kristiansen, Gitte, and Peirsman, Yves (eds.). 2010. Advances in Cognitive Sociolinguistics. Berlin: Mouton de Gruyter. Gelman, Andrew. 2008. Scaling Regression Inputs by Dividing by Two Standard Deviations. Statistics in Medicine, 27(15), 2865–2873. Gelman, Andrew, and Hill, Jennifer. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models. Analytical Methods for Social Research. Cambridge: Cambridge University Press. Gerwin, Johanna. 2013. Give It Me!: Pronominal Ditransitives in English Dialects. English Language and Linguistics, 17(03), 445–463. Gerwin, Johanna. 2014. Ditransitives in British English Dialects. Topics in English Linguistics. Berlin: Mouton de Gruyter. Gibson, Edward. 2000. The Dependency Locality Theory: A Distance-Based Theory of Linguistic Complexity. Pages 95–126 of Image, Language, Brain. Cambridge, MA: MIT Press. Gilquin, Gaëtanelle. 2011. Corpus Linguistics to Bridge the Gap between World Englishes and Learner Englishes. Pages 638–642 of Miyares, Leonel Ruiz, and Silva, María Rosa Álvarez (eds.), Comunicación Social En El Siglo XXI, Vol. II. Santiago de Cuba: Centro de Lingüística Aplicada. Gilquin, Gaëtanelle. 2015a. At the Interface of Contact Linguistics and Second Language Acquisition Reserch: New Englishes and Learner Englishes Compared. English World-Wide, 36(1), 90–123. Gilquin, Gaëtanelle. 2015b. The Use of Phrasal Verbs by French-speaking EFL Learners. A Constructional and Collostructional Corpus-Based Approach. Corpus Linguistics and Linguistic Theory, 11(1), 51–88. Gilquin, Gaëtanelle, and Gries, Stefan Th. 2009. Corpora and Experimental Methods: A State-of-the-Art Review. Corpus Linguistics and Linguistic Theory, 5(1), 1–26.

References

199

Givón, T. 1993. English Grammar: A Function-Based Introduction. Volume I. Amsterdam: Benjamins. Goebl, Hans. 1982. Dialektometrie: Prinzipien Und Methoden Des Einsatzes Der Numerischen Taxonomie Im Bereich Der Dialektgeographie. Wien: Österreichische Akademie der Wissenschaften. Goebl, Hans. 2008. Le Laboratoire de Dialectométrie de l’Université de Salzbourg. Zeitschrift für französische Sprache und Literatur, 118(1), 35–55. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Goldberg, Adele E. 2002. Surface Generalizations: An Alternative to Alternations. Cognitive Linguistics, 13–14, 327–356. Goldberg, Adele E. 2019. Explain Me This: Creativity, Competition, and the Partial Productivity of Constructions. Princeton: Princeton University Press. Görlach, Manfred. 1999. Regional and Social Variation. Pages 459–538 of Lass, Roger (ed.), The Cambridge History of the English Language, Volume 3 (1476–1776). Cambridge: Cambridge University Press. Grafmiller, Jason. 2014. Variation in English Genitives across Modality and Genres. English Language and Linguistics, 18(03), 471–496. Grafmiller, Jason, and Szmrecsanyi, Benedikt. 2018. Mapping Out Particle Placement in Englishes around the World: A Study in Comparative Sociolinguistic Analysis. Language Variation and Change, 30(03), 385–412. Grafmiller, Jason, Szmrecsanyi, Benedikt, and Hinrichs, Lars. 2016. Restricting the Restrictive Relativizer. Corpus Linguistics and Linguistic Theory, 14(2), 309–355. Grafmiller, Jason, Szmrecsanyi, Benedikt, Röthlisberger, Melanie, and Heller, Benedikt. 2018. General Introduction: A Comparative Perspective on Probabilistic Variation in Grammar. Glossa: a journal of general linguistics, 3(1). Green, Georgia. 1974. Semantics and Syntactic Regularity. Bloomington: Indiana University Press. Greenbaum, Sidney. 1991. ICE: The International Corpus of English. English Today, 7(04), 3. Greenbaum, Sidney. 1996. Comparing English Worldwide: The International Corpus of English. Oxford: Clarendon Press. Gries, Stefan Th. 1999. Particle Movement: A Cognitive and Functional Approach. Cognitive Linguistics, 10, 105–145. Gries, Stefan Th. 2002. The Influence of Processing on Syntactic Variation: Particle Placement in English. Pages 269–288 of Dehé, Nicole, Jackendorff, Ray, McIntyre, Andrew, and Urban, Silke (eds.), Verb-Particle Explorations. Berlin: Mouton de Gruyter. Gries, Stefan Th. 2003. Multifactorial Analysis in Corpus Linguistics: A Study of Particle Placement. New York: Continuum Press. Gries, Stefan Th. 2005. Syntactic Priming: A Corpus-based Approach. Journal of Psycholinguistic Research, 34(4), 365–399. Gries, Stefan Th. 2011. Acquiring Particle Placement in English: A Corpus-Based Perspective. Pages 235–263 of Morphosyntactic Alternations in English: Functional and Cognitive Perspectives. London: Equinox. Gries, Stefan Th. 2017. Syntactic Alternation Research: Taking Stock and Some Suggestions for the Future. Belgian Journal of Linguistics, 31(Dec.), 8–29. Gries, Stefan Th., and Bernaisch, Tobias. 2016. Exploring Epicentres Empirically: Focus on South Asian Englishes. English World-Wide, 37(1), 1–25.

200

References

Gries, Stefan Th., and Deshors, Sandra C. 2014. Using Regressions to Explore Deviations between Corpus Data and a Standard/Target: Two Suggestions. Corpora, 9(1), 109–136. Gries, Stefan Th., and Deshors, Sandra C. 2015. EFL and/vs. ESL?: A Multi-Level Regression Modeling Perspective on Bridging the Paradigm Gap. International Journal of Learner Corpus Research, 1(1), 130–159. Gries, Stefan Th., and Kootstra, Gerrit Jan. 2017. Structural Priming within and across Languages: A Corpus-Based Perspective. Bilingualism: Language and Cognition, 20(02), 235–250. Gries, Stefan Th., and Wulff, Stefanie. 2005. Do Foreign Language Learners Also Have Constructions? Evidence from Priming, Sorting, and Corpora. Annual Review of Cognitive Linguistics, 3, 182–200. Gries, Stefan Th, Bernaisch, Tobias, and Heller, Benedikt. 2018. A Corpus-Linguistic Account of the History of the Genitive Alternation in Singapore English. Pages 245–279 of Deshors, Sandra C., and Kautzsch, Alexander (eds.), Modeling World Englishes: Assessing the Interplay of Emancipation and Globalization of ESL Varieties. Varieties of English around the World (VEAW). Amsterdam: Benjamins. Grieve, Jack. 2016. Regional Variation in Written American English. Studies in English Language. Cambridge: Cambridge University Press. Grieve, Jack, Biber, Douglas, Friginal, Eric, and Nekrasova, Tatiana. 2010. Variation among Blog Text Types: A Multidimensional Analysis. Pages 303–322 of Mehler, Alexander, Sharoff, Serge, and Santini, Marina (eds.), Genres on the Web: Corpus Studies and Computational Models. New York: Springer. Grondelaers, Stefan, and Speelman, Dirk. 2007. A Variationist Account of Constituent Ordering in Presentative Sentences in Belgian Dutch. Corpus Linguistics and Linguistic Theory, 3(2), 161–193. Gropen, Jess, Pinker, Steven, Hollander, Michelle, Goldberg, Richard, and Wilson, Ronald. 1989. The Learnability and Acquisition of the Dative Alternation in English. Language, 65(2), 203–257. Guy, Gregory R. 2013. The Cognitive Coherence of Sociolects: How Do Speakers Handle Multiple Sociolinguistic Variables? Journal of Pragmatics, 52(June), 63–71. Guy, Gregory R. 2014. Linking Usage and Grammar: Generative Phonology, Exemplar Theory, and Variable Rules. Lingua, 142(Apr.), 57–65. Guy, Gregory R. 2015. Coherence, Constraints and Quantities. In NWAV44. Guy, Gregory R., and Hinskens, Frans. 2016. Linguistic Coherence: Systems, Repertoires and Speech Communities. Lingua, 172–173(Mar.), 1–9. Hackert, Stephanie, and Wengler, Diana. 2022. Recent Grammatical Change in Postcolonial Englishes: A Real-time Study of Genitive Variation in Caribbean and Indian News Writing. Journal of English Linguistics, 50(1), 3–38. Haddican, Bill, and Johnson, Daniel Ezra. 2012. Effects on the Particle Verb Alternation across English Dialects. Pages 31–40 of University of Pennsylvania Working Papers in Linguistics 18. University of Pennsylvania. Haddican, Bill, Johnson, Daniel Ezra, Wallenberg, Joel, and Holmberg, Anders. 2021. Variation and Change in the Particle Verb Alternation across English Dialects. Pages 205–228 of Advancing Socio-grammatical Variation and Change. London: Routledge. Haddican, Bill, Cutler, Cecelia, Newman, Michael, and Tortora, Christina. 2022. CrossSpeaker Covariation across Six Vocalic Changes in New York City English. American Speech, Mar., 1–27.

References

201

Haiman, John. 1980. The Iconicity of Grammar: Isomorphism and Motivation. Language, 56(3), 515. Hansen, Beke. 2018. Corpus Linguistics and Sociolinguistics: A Study of Variation and Change in the Modal Systems of World Englishes. Language and Computers: Studies in Digital Linguistics. Leiden: Brill Rodopi. Hawkins, John A. 1994. A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Hawkins, John A. 2004. Efficiency and Complexity in Grammars. Oxford Linguistics. Oxford: Oxford University Press. Heine, Bernd, and Kuteva, Tania. 2005. Language Contact and Grammatical Change. Cambridge Approaches to Language Contact. Cambridge: Cambridge University Press. Heller, Benedikt. 2018. Stability and Fluidity in Syntactic Variation World-Wide: The Genitive Alternation Across Varieties of English. Ph.D. thesis, KU Leuven. Heller, Benedikt, and Szmrecsanyi, Benedikt. 2019. Possessives World-Wide: Genitive Variation in Varieties of English. Pages 315–335 of Yáñez-Bouza, Nuria, Moore, Emma, van Bergen, Linda, and Hollmann, Willem B. (eds.), Categories, Constructions, and Change in English Syntax, first edn. Cambridge: Cambridge University Press. Heller, Benedikt, Bernaisch, Tobias, and Gries, Stefan Th. 2017a. Empirical Perspectives on Two Potential Epicenters: The Genitive Alternation in Asian Englishes. ICAME Journal, 41(1), 111–144. Heller, Benedikt, Szmrecsanyi, Benedikt, and Grafmiller, Jason. 2017b. Stability and Fluidity in Syntactic Variation World-Wide: The Genitive Alternation Across Varieties of English. Journal of English Linguistics, 45(1), 3–27. Hernández, Nuria. 2010. Personal Pronouns in the Dialects of England: A CorpusDriven Study of Grammatical Variation in Spontaneous Speech. Ph.D. thesis, University of Freiburg. Hickey, Raymond. 2010. The Englishes of Ireland: Emergence and Transportation. Pages 76–95 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. London: Routledge. Hilpert, Martin. 2008. The English Comparative – Language Structure and Language Use. English Language and Linguistics, 12(03), 395. Hiltunen, Risto. 1983. The Decline of the Prefixes and the Beginnings of the English Phrasal Verb: The Evidence from Some Old and Early Middle English Texts. Turku: University of Turku. Hiltunen, Turo, McVeigh, Joe, and Säily, Tanja (eds.). 2017. Big and Rich Data in English Corpus Linguistics: Methods and Explorations. Studies in Variation, Contacts and Change in English. VARIENG. Hinrichs, Lars, and Szmrecsanyi, Benedikt. 2007. Recent Changes in the Function and Frequency of Standard English Genitive Constructions: A Multivariate Analysis of Tagged Corpora. English Language and Linguistics, 11, 437–474. Hinrichs, Lars, Szmrecsanyi, Benedikt, and Bohmann, Axel. 2015. Which-Hunting and the Standard English Relative Clause. Language, 91(4), 806–836. Hoffmann, Thomas. 2011. Preposition Placement in English: A Usage-Based Approach. Cambridge: Cambridge University Press. Hoffmann, Thomas. 2014. The Cognitive Evolution of Englishes: The Role of Constructions in the Dynamic Model. Pages 160–180 of Buschfeld, Sarah, Hoffmann,

202

References

Thomas, Huber, Magnus, and Kautzsch, Alexander (eds.), Varieties of English Around the World. Amsterdam: Benjamins. Horch, Stephanie. 2019. Complementing Corpus Analysis with Web-Based Experimentation in Research on World Englishes. English World-Wide, 40(1), 25–53. Hothorn, Torsten, Hornik, Kurt, and Zeileis, Achim. 2006. Unbiased Recursive Partitioning: A Conditional Inference Framework. Journal of Computational and Graphical Statistics, 15(3), 651–674. Huddleston, Rodney, and Pullum, Geoffrey K. 2002. The Cambridge Grammer of the English Language. Cambridge: Cambridge University Press. Hudson, Richard. 1984. Word Grammar. Oxford: Blackwell. Hughes, Arthur, Trudgill, Peter, and Watt, Dominic James Landon. 2012. English Accents & Dialects: An Introduction to Social and Regional Varieties of English in the British Isles. Fifth edition London: Hodder Education. Hundt, Mairanne. 2013. The Diversification of English: Old, New and Emerging Epicentres. Pages 182–203 of Schreier, Daniel, and Hundt, Marianne (eds.), English as a Contact Language. Studies in English Language. Cambridge: Cambridge University Press. Hundt, Marianne. 2004. Animacy, Agentivity, and the Spread of the Progressive in Modern English. English Language and Linguistics, 8(1), 47–69. Hundt, Marianne. 2020. On Models and Modelling. World Englishes, 40(3), 298–317. Hundt, Marianne, and Gut, Ulrike (eds.). 2012. Mapping Unity and Diversity WorldWide: Corpus-Based Studies of New Englishes. Varieties of English around the World. Amsterdam: Benjamins. Hundt, Marianne, and Leech, Geoffrey. 2012. “Small Is Beautiful”: On the Value of Standard Reference Corpora for Observing Recent Grammatical Change. Pages 175–188 of Nevalainen, Terttu, and Traugott, Elizabeth Closs (eds.), The Oxford Handbook of the History of English. Oxford: Oxford University Press. Hundt, Marianne, and Szmrecsanyi, Benedikt. 2012. Animacy in Early New Zealand English. English World-Wide, 33, 241–263. Hundt, Marianne, Röthlisberger, Melanie, and Seoane, Elena. 2021. Predicting Voice Alternation across Academic Englishes. Corpus Linguistics and Linguistic Theory, 17(1), 189–222. Hundt, Marianne, Rautionaho, Paula, and Strobl, Carolin. 2020. Progressive or Simple? A Corpus-Based Study of Aspect in World Englishes. Corpora, 15(1), 77–106. Huson, Daniel H. 1998. SplitsTree: A Program for Analyzing and Visualizing Evolutionary Data. Bioinformatics, 14(10), 68–73. Huson, Daniel H., and Bryant, David. 2006. Application of Phylogenetic Networks in Evolutionary Studies. Molecular Biology Evolution, 23(2), 254–267. Jain, Anil K., Murty, M. Narasimha, and Flynn, Patrick J. 1999. Data Clustering: A Review. ACM Computing Surveys, 31(3), 264–323. James, Ariel N., Fraundorf, Scott H., Lee, Eun-Kyung, and Watson, Duane G. 2018. Individual Differences in Syntactic Processing: Is There Evidence for Reader-Text Interactions? Journal of Memory and Language, 102(Oct.), 155–181. Janda, Laura A. 2017. The Quantitative Turn. Pages 498–514 of Dancygier, Barbara (ed.), The Cambridge Handbook of Cognitive Linguistics. Cambridge Handbooks in Language and Linguistics. Cambridge: Cambridge University Press. Jankowski, Bridget L., and Tagliamonte, Sali. 2021. Urban-Rural Dimensions to Variable –Body/–One: The Case of Ontario, Canada. Pages 141–158 of Ziegler, Arne, Edler, Stefanie, and Oberdorfer, Georg (eds.), Urban Matters: Current Approaches

References

203

in Variationist Sociolinguistics. Studies in Language Variation (SILV). Amsterdam: Benjamins. Jankowski, Bridget L., and Tagliamonte, Sali A. 2014. On the Genitive’s Trail: Data and Method from a Sociolinguistic Perspective. English Language and Linguistics, 18(02), 305–329. Jankowski, Bridget Lynn. 2013. A Variationist Approach to Cross-Register Language Variation and Change. Ph.D. thesis, University of Toronto. Jones, Megan, and Tagliamonte, Sali. 2004. From Somerset to Samaná: Preverbal Did in the Voyage of English. Language Variation and Change, 16(02), 93–126. Kaan, Edith. 2014. Predictive Sentence Processing in L2 and L1: What Is Different? Linguistic Approaches to Bilingualism, 4(2), 257–282. Kaan, Edith, and Chun, Eunjin. 2018. Priming and Adaptation in Native Speakers and Second-Language Learners. Bilingualism: Language and Cognition, 21(2), 228–242. Kachru, Braj B. 1985. Standards, Codification and Sociolinguistic Realism: The English Language in the Outer Circle. Pages 11–30 of Quirk, Randolph, and Widdowson, Henry G. (eds.), English in the World: Teaching and Learning the Language and Literatures. Cambridge: Cambridge University Press. Kachru, Braj B. (ed.). 1992. The Other Tongue: English across Cultures. 2nd edition. English in the Global Context. Urbana: University of Illinois Press. Kachru, Braj B. 1997. World Englishes and English-Using Communities. Annual Review of Applied Linguistics, 17(Mar.), 66–87. Kallen, Jeffrey L. (ed.). 1997. Focus on Ireland. Varieties of English around the World, Amsterdam: Benjamins. Kastronic, Laura, and Poplack, Shana. 2014. The (North) American English Mandative Subjunctive in the 21st Century: Revival or Remnant? University of Pennsylvania Working Papers in Linguistics, 20(2), 9. Kempen, Gerard, and Harbusch, Karin. 2004. A Corpus Study into Word Order Variation in German Subordinate Clauses: Animacy Affects Linearization Independently of Grammatical Function Assignment. Pages 173–181 of Pechmann, Thomas, and Habel, Chrisopher (eds.), Multidisciplinary Approaches to Language Production. Berlin: Mouton de Gruyter. Kendall, Tyler, Bresnan, Joan, and Van Herk, Gerard. 2011. The Dative Alternation in African American English: Researching Syntactic Variation and Change across Sociolinguistic Datasets. Corpus Linguistics and Linguistic Theory, 11, 361–389. Kennedy, Arthur. 1920. The Modern English Verb-Adverb Combination. Stanford: Stanford University Press. Kennison, Shelia M., and Messer, Rachel H. 2014. Psycholinguistics. Oxford: Oxford University Press. Kim, Chaeyoon, Reddy, Sravana, Stanford, James N., Wyschogrod, Ezra, and Grieve, Jack. 2019. Bring on the Crowd! Using Online Audio Crowd-Sourcing for LargeScale New England Dialectology and Acoustic Sociophonetics. American Speech, 94(2), 151–194. Kim, Su Nam, and Baldwin, Timothy. 2010. How to Pick out Token Instances of English Verb-Particle Constructions. Language Resources and Evaluation, 44(1–2), 97–113. Kirk, John, and Nelson, Gerald. 2018. The International Corpus of English Project: A Progress Report. World Englishes, 37(4), 697–716. Kirkpatrick, Andy. 2010. The Routledge Handbook of World Englishes. Routledge Handbooks in Applied Linguistics. London: Routledge.

204

References

Klavan, J., and Divjak, D. 2016. The Cognitive Plausibility of Statistical Classification Models: Comparing Textual and Behavioral Evidence. Folia Linguistica, 50(2), 355–384. Klavan, Jane. 2020. Pitting Corpus-Based Classification Models against Each Other: A Case Study for Predicting Constructional Choice in Written Estonian. Corpus Linguistics and Linguistic Theory, 16(2), 363–391. Klein, Wolfgang, and Perdue, Clive. 1997. The Basic Variety (or: Couldn’t Natural Languages Be Much Simpler?). Second Language Research, 13, 301–347. Koopman, Willem. 1990. Word Order in Old English. Ph.D. thesis, University of Amsterdam. Kortmann, Bernd. 2002. New Prospects for the Study of Dialect Syntax: Impetus from Syntactic Theory and Language Typology. Pages 185–213 of Barbiers, Sief, Cornips, Leonie, and van der Kleij, Susanne (eds.), Syntactic Microvariation. Amsterdam: Meertens Instituut. Kortmann, Bernd, and Lunkenheimer, Kerstin (eds.). 2012. The Mouton World Atlas of Variation in English. Berlin: Mouton de Gruyter. Kortmann, Bernd, and Lunkenheimer, Kerstin (eds.). 2013. eWAVE. Leipzig: Max Planck Institute for Evolutionary Anthropology. Kortmann, Bernd, and Schröter, Verena. 2017. Varieties of English. Pages 304–330 of Hickey, Raymond (ed.), The Cambridge Hanbook of Areal Linguistics. Analytical Methods for Social Research. Cambridge: Cambridge University Press. Kortmann, Bernd, and Szmrecsanyi, Benedikt. 2004. Global Synopsis: Morphological and Syntactic Variation in English. Pages 1142–1202 of Kortmann, Bernd, Schneider, Edgar, Burridge, K, Mesthrie, R, and Upton, C (eds.), A Handbook of Varieties of English, vol. 2. Berlin: Mouton de Gruyter. Kortmann, Bernd, and Szmrecsanyi, Benedikt. 2009. World Englishes between Simplification and Complexification. Pages 265–285 of Siebers, Lucia, and Hoffmann, Thomas (eds.), World Englishes – Problems, Properties and Prospects: Selected Papers from the 13th IAWE Conference. Amsterdam: Benjamins. Kortmann, Bernd, and Wolk, Christoph. 2012. Morphosyntactic Variation in the Anglophone World: A Global Perspective. In Kortmann, Bernd, and Lunkenheimer, Kerstin (eds.), The Mouton World Atlas of Variation in English. Berlin: Mouton de Gruyter. Kortmann, Bernd, Schneider, Edgar, Burridge, Kate, Mesthrie, Raj, and Upton, Clive (eds.). 2004. A Handbook of Varieties of English. Berlin: Mouton de Gruyter. Kristiansen, Gitte, and Geeraerts, Dirk (eds.). 2013. Contexts of Use in Cognitive Sociolinguistics. Thematic Issue of Journal of Pragmatics 52. Kruisinga, Etsko, and Erades, P. A. 1953. An English Grammar. Vol. I. Groningen: P. Noordhoff. Kruskal, Joseph B., and Wish, Myron. 1978. Multidimensional Scaling. Newbury Park, London: Sage Publications. La Peruta, Roberta. 2022. Using VADIS to weigh competing epicentral influence. World Englishes, 41(3), 400–413. Labov, William. 1963. The Social Motivation of a Sound Change. Word, 19, 273–309. Labov, William. 1966. The Social Stratification of English in New York City. Washington, DC: Center for Applied Linguistics. Labov, William. 1969. Contraction, Deletion, and Inherent Variability of the English Copula. Language, 45, 715–762. Labov, William. 1972. Sociolinguistic Patterns. Philadelphia: University of Philadelphia Press.

References

205

Labov, William. 1978. Where Does the Linguistic Variable Stop? A Response to Beatriz Lavandera. Pages 1–22 of Working Papers in Sociolinguistics, No. 44. Austin, TX: Southwest Educational Development Lab. Labov, William. 2001. Principles of Linguistic Change: Volume 2: Social Factors. Oxford: Blackwell. Labov, William. 2007. Transmission and Diffusion. Language, 83(2), 344–387. Lalla, Barbara, and D’Costa, Jean. 1990. Language in Exile: Three Hundred Years of Jamaican Creole. Tuscaloosa: University of Alabama Press. Lass, Roger. 2004. South African English. Pages 363–386 of Hickey, Raymond (ed.), Legacies of Colonial English: Studies in Transported Dialects. Cambridge: Cambridge University Press. Lavandera, Beatriz. 1978. Where Does the Sociolinguistic Variable Stop? Language in Society, 7, 171–183. Leemann, Adrian, Kolly, Marie-José, Purves, Ross, Britain, David, and Glaser, Elvira. 2016. Crowdsourcing Language Change with Smartphone Applications. PLOS ONE, 11(1), e0143060. Leimgruber, Jakob R. E. 2013. Singapore English: Structure, Variation, and Usage. Cambridge: Cambridge University Press. Levey, Stephen. 2010. The Englishes of Canada. Pages 113–131 of Kirkpatrick, Andy (ed.), Routledge Handbook of World Englishes. London: Routledge. Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Levshina, Natalia. 2015. How to Do Linguistics with R: Data Exploration and Statistical Analysis. Amsterdam: Benjamins. Levshina, Natalia. 2018. Probabilistic Grammar and Constructional Predictability: Bayesian Generalized Additive Models of Help + (to) Infinitive in Varieties of Web-Based English. Glossa: a journal of general linguistics, 3(1), 55. Levshina, Natalia, Geeraerts, Dirk, and Speelman, Dirk. 2013. Towards a 3D-grammar: Interaction of Linguistic and Extralinguistic Factors in the Use of Dutch Causative Constructions. Journal of Pragmatics, 52(June), 34–48. Levy, Roger, and Jaeger, T Florian. 2007. Speakers Optimize Information Density through Syntactic Reduction. Pages 849–856 of Schölkopf, B, Platt, J, and Hoffman, T (eds.), Advances in Neural Information Processing Systems 19. Cambridge, MA: MIT Press. Li, D.C.S. 1999. The Function and Status of English in Hong Kong: A Post-1997 Update. English World-Wide, 20(1), 67–110. Liao, Yan, and Fukuya, Yoshinori J. 2004. Avoidance of Phrasal Verbs: The Case of Chinese Learners of English: Language Learning. Language Learning, 54(2), 193–226. Löffler, Heinrich. 2003. Dialektologie. Eine Einführung. Tübingen: Narr. Lohse, Barbara, Hawkins, John A., and Wasow, Thomas. 2004. Domain Minimization in English Verb-Particle Constructions. Language, 80(2), 238–261. Loureiro-Porto, Lucía. 2017. ICE vs GloWbE: Big Data and Corpus Compilation. World Englishes, 36(3), 448–470. Low, Ee-Ling. 2010. English in Singapore and Malaysia: Differences and Similarities. Pages 229–246 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. London: Routledge. MacDonald, Maryellen C. 2013. How Language Production Shapes Language Form and Comprehension. Frontiers in Psychology, 4, 1–16.

206

References

Macglagan, Margaret. 2010. The English(es) of New Zealand. Pages 152–164 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. Routledge Handbooks in Applied Linguistics. London: Routledge. MacKenzie, Laurel. 2013. Variation in English Auxiliary Realization: A New Take on Contraction. Language Variation and Change, 25(1), 17–41. MacWhinney, Brian. 1997. Second Language Acquisition and the Competition Model. Pages 113–142 of De Groot, Anette M.B., and Kroll, Judit F. (eds.), Tutorials in Bilingualism: Psycholinguistic Perspectives. Mahawa, NJ: Lawrence Erlbaum Associates. Mahon, Bradford Z., and Caramazza, Alfonso. 2009. Concepts and Categories: A Cognitive Neuropsychological Perspective. Annual review of psychology, 60, 27–51. Mair, Christian. 2002. Creolisms in an Emergent Standard: Written English in Jamaica. English World-Wide, 23, 31–58. Mair, Christian. 2003. Gerundial Complements after Begin and Start: Grammatical and Sociolinguistic Factors, and How They Work against Each Other. Pages 329–346 of Rohdenburg, Günter, and Mondorf, Britta (eds.), Determinants of Grammatical Variation in English. Berlin: Mouton de Gruyter. Mair, Christian. 2013. The World System of Englishes: Accounting for the Transnational Importance of Mobile and Mediated Vernaculars. English World-Wide, 34(3), 253–278. Malchukov, Andrej L. 2008. Animacy and Asymmetries in Differential Case Marking. Lingua, 118, 203–221. Martin, Clara D., Branzi, Francesca M., and Bar, Moshe. 2018. Prediction Is Production: The Missing Link between Language Production and Comprehension. Scientific Reports, 8(1), 1079. McArthur, Tom. 1992. Models of English. English Today, 8(4), 12–21. McArthur, Tom. 1998. The English Languages. Cambridge: Cambridge University Press. McArthur, Tom. 2003. World English, Euro-English, Nordic English? English Today, 19(1), 54–58. McFadden, Thomas. 2002. The Rise of the To-Dative in Middle English. Pages 107– 123 of Lightfoot, David (ed.), Syntactic Effects of Morphological Change. Oxford: Oxford University Press. McMahon, April, Heggarty, Paul, McMahon, Robert, and Maguire, Warren. 2007. The Sound Patterns of Englishes: Representing Phonetic Similarity. English Language and Linguistics, 11(1), 113–142. Melchers, Gunnel, and Shaw, Philip. 2011. World Englishes. London: Hodder Education. Meriläinen, Lea, Paulasto, Heli, and Rautionaho, Paula. 2017. Extended Uses of the Progressive Form in Inner, Outer and Expanding Circle Englishes. Pages 191–216 of Extended Uses of the Progressive Form in Inner, Outer and Expanding Circle Englishes. Berlin: Mouton de Gruyter. Mesthrie, Rajend, and Bhatt, Rakesh Mohan. 2008. World Englishes: The Study of New Linguistic Varieties. Key Topics in Sociolinguistics. Cambridge: Cambridge University Press. Meyerhoff, Miriam. 2016. Methods, Innovations and Extensions: Reflections on Half a Century of Methodology in Social Dialectology. Journal of Sociolinguistics, 20(4), 431–452.

References

207

Modiano, Marko. 1999. Standard English(es) and Educational Practices for the World’s Lingua Franca. English Today, 15(4), 3–13. Mufwene, Salikoko. 2001. The Ecology of Language Evolution. Cambridge: Cambridge University Press. Mukherjee, Joybrato. 2005. English Ditransitive Verbs: Aspects of Theory, Description and a Usage-based Model. Language and Computers. Amsterdam: Rodopi. Mukherjee, Joybrato. 2007. Steady States in the Evolution of New Englishes: Presentday Indian English as an Equilibrium. Journal of English Linguistics, 35(2), 157–187. Mukherjee, Joybrato. 2010a. Corpus-Based Insights into Verb-Complementational Innovations in Indian English: Cases of Nativised Semantico-Structural Analogy. Pages 219–241 of Lenz, Alexandra N., and Plewnia, Albrecht (eds.), Grammar between Norm and Variation. Frankfurt am Main: Peter Lang. Mukherjee, Joybrato. 2010b. The Development of the English Language in India. Pages 167–180 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. London: Routledge. Mukherjee, Joybrato, and Gries, Stefan Th. 2009. Collostructional Nativisation in New Englishes: Verb-construction Associations in the International Corpus of English. English World-Wide, 30(1), 27–51. Mukherjee, Joybrato, and Hoffmann, Sebastian. 2006. Describing VerbComplementational Profiles of New Englishes: A Pilot Study of Indian English. English World-Wide, 27, 147–173. Mukherjee, Joybrato, and Hundt, Marianne (eds.). 2011. Exploring Second Language Varieties of English and Learner Englishes. Amsterdam: Benjamins. Mustanoja, Tauno F. 1960. A Middle English Syntax. Vol. I. Helsinki: Société Néophilologique. Nagy, Naomi, and Hoffman, Michol F. 2017. Variation at the Crossroads: Advancing Theory by Integrating Methods. Canadian Journal of Linguistics/Revue canadienne de linguistique, 62(04), 517–518. Nelson, Cecil L., Proshina, Z. G., Davis, Daniel R., Kachru, Braj B., and Kachru, Yamuna (eds.). 2019. Handbook of World Englishes. 2nd edition edn. Chichester: Wiley-Blackwell. Nelson, Gerald. 1996. The Design of the Corpus. Pages 3–12 of Greenbaum, Sidney (ed.), Comparing English Worldwide: The International Corpus of English. Oxford: Clarendon Press. Nerbonne, John, Heeringa, Wilbert, and Kleiweg, Peter. 1999. Edit Distance and Dialect Proximity. Pages v–xv of Sankoff, David, and Kruskal, Joseph (eds.), Time Warps, String Edits and Macromolecules: The Theory and Practice of Sequence Comparison. Stanford: CSLI Press. Oehrle, Richard Thomas. 1976. The Grammatical Status of the English Dative Alternation. Ph.D. thesis, MIT. Olavarría de Ersson, Eugenia, and Shaw, Philip. 2003. Verb Complementation Patterns in Indian Standard English. English World-Wide, 24, 137–161. Onysko, Alexander. 2016. Modeling World Englishes from the Perspective of Language Contact. World Englishes, 35(2), 196–220. Orton, Harold, and Dieth, Eugen. 1962. Survey of English Dialects. Leeds: E. J. Arnold. Osselton, Noel. 1988. Thematic Genitives. Pages 138–144 of Nixon, Graham, and Honey, John (eds.), An Historic Tongue: Studies in English Linguistics in Memory of Barbara Strang. London: Routledge.

208

References

Osthoff, Hermann, and Brugmann, Karl. 1878. Morphologische Untersuchungen auf dem Gebiete der indogermanischen Sprachen. 1. Theil. Leipzig. Oushiro, Livia. 2016. Social and Structural Constraints in Lectal Cohesion. Lingua, 172(Mar.), 116–130. Paquot, Magali, Grafmiller, Jason, and Szmrecsanyi, Benedikt. 2019. Particle Placement Alternation in EFL Learner vs. L1 Speech: Assessing the Similarity of Probabilistic Grammars. Pages 71–92 of Abel, Andrea, Glaznieks, Aivars, and Lyding, Verena (eds.), Widening the Scope of Learner Corpus Research. Selected Papers from the Fourth Learner Corpus Research Conference. Louvain-la-Neuve: Catholic University of Louvain Press. Patrick, Peter. 2004. Jamaican Creole: Morphology and Syntax. Pages 407–439 of Kortmann, Bernd, Schneider, Edgar W., Upton, Clive, Mesthrie, Rajend, and Burridge, Kate (eds.), A Handbook of Varieties of English, vol. 2: Morphology and Syntax. Berlin: Mouton de Gruyter. Payne, John, Pullum, Geoffrey K., Scholz, Barbara C., and Berlage, Eva. 2013. Anaphoric One and Its Implications. Language, 89(4), 794–829. Pearl, Judea, TotalBoox, and TBX. 1988. Probabilistic Reasoning in Intelligent Systems. Cambridge, MA: Morgan Kaufmann Publishers Inc. Pefianco Martin, Isabel. 2010. Periphery ELT: The Politics and Practice of Teaching English in the Philippines. Pages 247–264 of Kirkpatrick, Andy (ed.), The Routledge Handbook of World Englishes. London: Routledge. Pefianco Martin, Isabel. 2014. Beyond Nativization? Philippine English in Schneider’s Dynamic Model. Pages 70–85 of Buschfeld, Sarah, Hoffmann, Thomas, Huber, Magnus, and Kautzsch, Alexander (eds.), The Evolution of Englishes: The Dynamic Model and Beyond. Varieties of English Around the World. Amsterdam: Benjamins. Perek, Florent, and Goldberg, Adele E. 2015. Generalizing beyond the Input: The Functions of the Constructions Matter. Journal of Memory and Language, 84(Oct.), 108–127. Perek, Florent, and Goldberg, Adele E. 2017. Linguistic Generalization on the Basis of Function and Constraints on the Basis of Statistical Preemption. Cognition, 168(Nov.), 276–293. Pickering, Martin J., and Gambi, Chiara. 2018. Predicting While Comprehending Language: A Theory and Review. Psychological Bulletin, 144(10), 1002–1044. Pierrehumbert, Janet B. 2006. The next Toolkit. Journal of Phonetics, 34(4), 516–530. Pijpops, Dirk, and Van de Velde, Freek. 2018. A Multivariate Analysis of the Partitive Genitive in Dutch. Bringing Quantitative Data into a Theoretical Discussion. Corpus Linguistics and Linguistic Theory, 14(1), 1–33. Pinker, Steven. 1989. Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA: MIT Press. Poplack, Shana. 2018. Categories of Grammar and Categories of Speech: When the Quest for Symmetry Meets Inherent Variability. Pages 7–34 of Shin, Naomi L., and Erker, Daniel (eds.), Studies in Functional and Structural Linguistics, vol. 76. Amsterdam: Benjamins. Poplack, Shana, and Dion, Nathalie. 2009. Prescription vs. Praxis: The Evolution of Future Temporal Reference in French. Language, 557–587. Poplack, Shana, and Tagliamonte, Sali. 1989. There’s No Tense Like the Present: Verbal -s Inflection in Early Black English. Language Variation and Change, 1(1), 47–84. Poplack, Shana, and Tagliamonte, Sali. 1999. The Grammaticization of Going to in (African American) English. Language Variation and Change, 11(3), 315–342.

References

209

Poplack, Shana, and Tagliamonte, Sali. 2001. African American English in the Diaspora. Language in Society. Oxford: Blackwell. Prat-Sala, Mercè, and Branigan, Holly P. 2000. Discourse Constraints on Syntactic Processing in Language Production: A Crosslinguistic Study in English and Spanish. Journal of Memory and Language, 42(2), 168–182. Preston, Dennis Richard, and Niedzielski, Nancy A. (eds.). 2010a. Introduction: Sociophonetic Studies of Language Variety. Pages 1–14 of Preston, Dennis Richard, and Niedzielski, Nancy A. (eds.), A Reader in Sociophonetics. Trends in Linguistics. Studies and Monographs. Berlin: Mouton de Gruyter. Preston, Dennis Richard, and Niedzielski, Nancy A. (eds.). 2010b. A Reader in Sociophonetics. Trends in Linguistics. Studies and Monographs. Berlin: Mouton de Gruyter. Pütz, Martin, Robinson, Justyna, and Reif, Monika (eds.). 2014. Cognitive Sociolinguistics: Social and Cultural Variation in Cognition and Language Use. Thematic Issue of Review of Cognitive Linguistics 10 (2012). Amsterdam: Benjamins. Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey, and Svartvik, Jan. 1985. A Comprehensive Grammar of the English Language. London: Longman. R Core Team. 2014. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. Rodríguez-Puente, Paula. 2013. The Development of Phrasal Verbs in British English from 1650 to 1990: A Corpus-based Study. Ph.D. thesis, University of Santiago de Compostela. Rodríguez-Puente, Paula. 2016. Particle Placement in Late Modern English and Twentieth-Century English: Morpho-syntactic Variables. Folia Linguistica, 37, 145– 175. Rosenbach, Anette. 2002. Genitive Variation in English: Conceptual Factors in Synchronic and Diachronic Studies. Berlin: Mouton de Gruyter. Rosenbach, Anette. 2005. Animacy versus Weight as Determinants of Grammatical Variation in English. Language, 81(3), 613–644. Rosenbach, Anette. 2006. Descriptive Genitives in English: A Case Study on Constructional Gradience. English Language and Linguistics, 10(1), 77–118. Rosenbach, Anette. 2014. English Genitive Variation – the State of the Art. English Language and Linguistics, 18(02), 215–262. Rosenbach, Anette. 2017. Constraints in Contact: Animacy in English and Afrikaans Genitive Variation – a Cross-Linguistic Perspective. Glossa: a Journal of General Linguistics, 2(1), 72. Röthlisberger, Melanie. 2014. Guidelines for the Dative Alternation. Röthlisberger, Melanie. 2018a. The Dative Dataset of World Englishes. https://doi .org/10.5281/zenodo.2553357. Röthlisberger, Melanie. 2018b. Regional Variation in Probabilistic Grammars: A Multifactorial Study of the English Dative Alternation. Ph.D. thesis, KU Leuven. Röthlisberger, Melanie, and Szmrecsanyi, Benedikt. 2019. Dialect Typology: Recent Advances. Pages 1–26 of Brunn, Stanley D, and Kehrein, Roland (eds.), Handbook of the Changing World Language Map. Cham: Springer International Publishing. Röthlisberger, Melanie, and Tagliamonte, Sali A. 2020. The Social Embedding of a Syntactic Alternation: Variable Particle Placement in Ontario English. Language Variation and Change, 32(3), 317–348. Röthlisberger, Melanie, Grafmiller, Jason, and Szmrecsanyi, Benedikt. 2017. Cognitive Indigenization Effects in the English Dative Alternation. Cognitive Linguistics, 28(4), 673–710.

210

References

Sailaja, Pingali. 2009. Indian English. Dialects of English. Edinburgh: Edinburgh University Press. Sand, Andrea. 2013. Singapore Weblogs: Between Speech and Writing. In Huber, Magnus, and Mukherjee, Joybrato (eds.), Corpus Linguistics and Variation in English: Focus on Non-Native Englishes, vol. 13. VARIENG. Sankoff, David. 1988. Sociolinguistics and Syntactic Variation. Pages 140–161 of Newmeyer, F. (ed.), Linguistics: The Cambridge Survey. Cambridge: Cambridge University Press. Schilk, Marco, Bernaisch, Tobias, and Mukherjee, Joybrato. 2012. Mapping Unity and Diversity in South Asian English Lexicogrammar: Verb-Complementational Preferences across Varieties. Pages 137–165 of Hundt, Marianne, and Gut, Ulrike (eds.), Mapping Unity and Diversity World-Wide: Corpus-Based Studies of New Englishes. Amsterdam: Benjamins. Schilk, Marco, Mukherjee, Joybrato, and Nam, Christopher. 2013. Complementation of Ditransitive Verbs in South Asian Englishes: A Multifactorial Analysis. Corpus Linguistics and Linguistic Theory, 9(2), 1–40. Schmid, Hans-Jörg (ed.). 2017. Entrenchment and the Psychology of Language Learning: How We Reorganize and Adapt Linguistic Knowledge. Berlin: Mouton de Gruyter. Schmid, Hans-Jörg, Würschinger, Quirin, Fischer, Sebastian, and Küchenhoff, Helmut. 2020. That’s Cool. Computational Sociolinguistic Methods for Investigating Individual Lexico-Grammatical Variation. Frontiers in Artificial Intelligence, 3. Schneider, Edgar. 2003. The Dynamics of New Englishes: From Identity Construction to Dialect Birth. Language, 79(2), 233–281. Schneider, Edgar. 2007. Postcolonial English: Varieties Around the World. Cambridge: Cambridge University Press. Schneider, Edgar. 2011. English Around the World: An Introduction. [S.l.]: Cambridge University Press. Schneider, Edgar. 2014. New Reflections on the Evolutionary Dynamics of World Englishes:. World Englishes, 33(1), 9–32. Schreier, Daniel, Hundt, Mairanne, and Schneider, Edgar W. 2019. World Englishes: An Introduction. Pages 1–22 of Schreier, Daniel, Hundt, Marianne, and Schneider, Edgar W. (eds.), The Cambridge Handbook of World Englishes, first edn. Cambridge Handbooks in Language and Linguistics. Cambridge: Cambridge University Press. Séguy, Jean. 1971. La Relation Entre La Distance Spatiale et La Distance Lexicale. Revue de Linguistique Romane, 35, 335–357. Seoane, Elena. 2016. World Englishes Today. Pages 1–16 of Seoane, Elena, and Suárez Gómez, Cristina (eds.), World Englishes: New Theoretical and Methodological Considerations. Varieties of English around the World. Amsterdam: Benjamins. Setter, Jane, Wong, Cathy S., and Chan, Brian H.S. 2010. Hong Kong English. Edinburgh: Edinburgh University Press. Seuren, Pieter, and Wekker, Herman. 1986. Semantic Transparency as a Factor in Creole Genesis. Pages 57–70 of Muysken, Pieter, and Smith, Norval (eds.), Substrata versus Universals in Creole Genesis. Amsterdam: Benjamins. Shannon, Claude E. 1948. A Mathematical Theory of Communication. Bell System Technical Journal, 27, 379–423. Sharma, Devyani. 2009. Typological Diversity in New Englishes. English World-Wide, 30(2), 170–195.

References

211

Shih, Stephanie, Grafmiller, Jason, Futrell, Richard, and Bresnan, Joan. 2015. Rhythm’s Role in Genitive Construction Choice in Spoken English. In Vogel, Ralf, and Vijver, Ruben (eds.), Rhythm in Cognition and Grammar. Berlin: Mouton de Gruyter. Silva-Corvalán, Carmen. 1994. Language Contact and Change: Spanish in Los Angeles. Oxford Studies in Language Contact. Oxford: Clarendon Press. Siyanova, Anna, and Schmitt, Norbert. 2007. Native and Nonnative Use of MultiWord vs. One-Word Verbs. IRAL – International Review of Applied Linguistics in Language Teaching, 45(2). Spruit, Marco René, Heeringa, Wilbert, and Nerbonne, John. 2009. Associations among Linguistic Levels. Lingua, 119(11), 1624–1642. Stephens, Nola. 2015. Dative Constructions and Givenness in the Speech of Four-YearOlds. Linguistics, 53(3), 405–442. Strang, Barbara. 1968. Modern English Structure. Second edition. London: Arnold. Strobl, Carolin, Boulesteix, Anne-Laure, Kneib, Thomas, Augustin, Thomas, and Zeileis, Achim. 2008. Conditional Variable Importance for Random Forests. BMC Bioinformatics, 9(1), 307. Strobl, Carolin, Malley, James, and Tutz, Gerhard. 2009. An Introduction to Recursive Partitioning: Rationale, Application, and Characteristics of Classification and Regression Trees, Bagging, and Random Forests. Psychological Methods, 14(4), 323–348. Szmrecsanyi, Benedikt. 2005. Language Users as Creatures of Habit: A Corpus-Based Analysis of Persistence in Spoken English. Corpus Linguistics and Linguistic Theory, 1, 113–149. Szmrecsanyi, Benedikt. 2006. Morphosyntactic Persistence in Spoken English. Berlin: Mouton de Gruyter. Szmrecsanyi, Benedikt. 2010. The English Genitive Alternation in a Cognitive Sociolinguistics Perspective. Mouton de Gruyter. Pages 141–166. Szmrecsanyi, Benedikt. 2013a. Grammatical Variation in British English Dialects: A Study in Corpus-Based Dialectometry. Studies in English Language. Cambridge: Cambridge University Press. Szmrecsanyi, Benedikt. 2013b. The Great Regression: Genitive Variability in Late Modern English News Texts. Pages 59–88 of Börjars, Kersti, Denison, David, and Scott, Alan (eds.), Morphosyntactic Categories and the Expression of Possession. Amsterdam: Benjamins. Szmrecsanyi, Benedikt. 2016. About Text Frequencies in Historical Linguistics: Disentangling Environmental and Grammatical Change. Corpus Linguistics and Linguistic Theory, 12(1), 153–171. Szmrecsanyi, Benedikt. 2017. Variationist Sociolinguistics and Corpus-Based Variationist Linguistics: Overlap and Cross-Pollination Potential. Canadian Journal of Linguistics/Revue canadienne de linguistique, 62(4), 685–701. Szmrecsanyi, Benedikt. 2019. Register in Variationist Linguistics. Register Studies, 1(1), 76–99. Szmrecsanyi, Benedikt. 2022. Measuring distance-based coherence. Pages 127–142 of Beaman, Karen V., and Guy, Gregory R. (eds.), The Coherence of Linguistic Communities: Orderly Heterogeneity and Social Meaning. London: Routledge. Szmrecsanyi, Benedikt, and Hinrichs, Lars. 2008. Probabilistic Determinants of Genitive Variation in Spoken and Written English: A Multivariate Comparison across Time, Space, and Genres. Pages 291–309 of Nevalainen, Terttu, Taavitsainen, Irma,

212

References

Pahta, Päivi, and Korhonen, Minna (eds.), The Dynamics of Linguistic Variation: Corpus Evidence on English Past and Present. Amsterdam: Benjamins. Szmrecsanyi, Benedikt, and Kortmann, Bernd. 2009. The Morphosyntax of Varieties of English Worldwide: A Quantitative Perspective. Lingua, 119(11), 1643–1663. Szmrecsanyi, Benedikt, and Röthlisberger, Melanie. 2019. World Englishes from the Perspective of Dialect Typology. Pages 534–558 of Schreier, Daniel, Hundt, Marianne, and Schneider, Edgar W. (eds.), The Cambridge Handbook of World Englishes. Cambridge: Cambridge University Press. Szmrecsanyi, Benedikt, and Wolk, Christoph. 2011. Holistic Corpus-Based Dialectology. Brazilian Journal of Applied Linguistics / Revista Brasileira de Linguística Aplicada, 11(2), 561–592. Szmrecsanyi, Benedikt, Rosenbach, Anette, Bresnan, Joan, and Wolk, Christoph. 2014. Culturally Conditioned Language Change? A Multivariate Analysis of Genitive Constructions in ARCHER. Pages 133–152 of Hundt, Marianne (ed.), Late Modern English Syntax. Studies in English Language. Cambridge: Cambridge University Press. Szmrecsanyi, Benedikt, Grafmiller, Jason, Heller, Benedikt, and Röthlisberger, Melanie. 2016a. Around the World in Three Alternations: Modeling Syntactic Variation in Varieties of English. English World-Wide, 37(2), 109–137. Szmrecsanyi, Benedikt, Biber, Douglas, Egbert, Jesse, and Franco, Karlien. 2016b. Toward More Accountability: Modeling Ternary Genitive Variation in Late Modern English. Language Variation and Change, 28(1), 1–29. Szmrecsanyi, Benedikt, Grafmiller, Jason, Bresnan, Joan, Rosenbach, Anette, Tagliamonte, Sali, and Todd, Simon. 2017. Spoken Syntax in a Comparative Perspective: The Dative and Genitive Alternation in Varieties of English. Glossa: a Journal of General Linguistics, 2(1), 86. Szmrecsanyi, Benedikt, Grafmiller, Jason, and Rosseel, Laura. 2019. Variation-Based Distance and Similarity Modeling: A Case Study in World Englishes. Frontiers in Artificial Intelligence, 2(Nov.), 23. Tagliamonte, Sali. 2000. The Grammaticalization of the Present Perfect in English: Tracks of Change and Continuity in a Linguistic Enclave. Pages 329–354 of Fischer, Olga, Rosenbach, Anette, and Stein, Dieter (eds.), Pathways of Change: Grammaticalization in English. Amsterdam: Benjamins. Tagliamonte, Sali. 2001. Comparative Sociolinguistics. Pages 729–763 of Chambers, Jack, Trudgill, Peter, and Schilling-Estes, Natalie (eds.), Handbook of Language Variation and Change. Oxford: Blackwell. Tagliamonte, Sali. 2009. “There Was Universals; Then There Weren’t”: A Comparative Sociolinguistic Perspective on “Default Singulars”. Pages 103–129 of Filppula, Markku, Klemola, Juhani, and Paulasto, Heli (eds.), Vernacular Universals and Language Contacts: Evidence from Varieties of English and Beyond. London: Routledge. Tagliamonte, Sali. 2012. Variationist Sociolinguistics: Change, Observation, Interpretation. Language in Society. Chichester: Wiley-Blackwell. Tagliamonte, Sali. 2013. Comparative Sociolinguistics. Pages 130–156 of Chambers, J. K., and Schilling, Natalie (eds.), Handbook of Language Variation and Change, second edn. Chichester: John Wiley & Sons. Tagliamonte, Sali, and Baayen, Harald. 2012. Models, Forests and Trees of York English: “Was/Were” Variation as a Case Study for Statistical Practice. Language Variation and Change, 24, 135–178.

References

213

Tagliamonte, Sali, and Denis, Derek. 2010. The Stuff of Change: General Extenders in Toronto, Canada. Journal of English Linguistics, 38(4), 335–368. Tagliamonte, Sali, and Smith, Jennifer. 2002. “Either It Isn’t or It’s Not”: Neg/Aux Contraction in British Dialects. English World Wide, 23(2), 251–281. Tagliamonte, Sali, and Smith, Jennifer. 2005. No Momentary Fancy! The Zero “Complementizer” in English Dialects. English Language and Linguistics, 9(2), 289–309. Tagliamonte, Sali A. 2014. A Comparative Sociolinguistic Analysis of the Dative Alternation. Pages 297–318 of Torres-Cacoullos, Rena, Dion, Nathalie, and Lapierre, André (eds.), Linguistic Variation: Confronting Fact and Theory. London: Routledge. Tagliamonte, Sali A., and D’Arcy, Alexandra. 2009. Peaks Beyond Phonology: Adolescence, Incrementation, and Language Change. Language, 85(1), 58–108. Tagliamonte, Sali A, and Smith, Jennifer. 2006. Layering, Competition and a Twist of Fate: Deontic Modality in Dialects of English. Diachronica, 23(2), 341–380. Tagliamonte, Sali A., D’Arcy, Alexandra, and Jankowski, Bridget. 2010. Social Work and Linguistic Systems: Marking Possession in Canadian English. Language Variation and Change, 22(1), 149–173. Tagliamonte, Sali A., D’Arcy, Alexandra, and Louro, Celeste Rodríguez. 2016a. Outliers, Impact, and Rationalization in Linguistic Change. Language, 92(4), 824–849. Tagliamonte, Sali A., In collaboration with Dylan Uscher, Lawrence Kwok, and students from HUM199Y 2009 and 2010. 2016b. So Sick or so Cool? The Language of Youth on the Internet. Language in Society, 45(1), 1–32. Tamaredo, Iván, Röthlisberger, Melanie, Grafmiller, Jason, and Heller, Benedikt. 2019. Probabilistic Indigenization Effects at the Lexis–Syntax Interface. English Language and Linguistics, 24(2), 413–440. Tamminga, Meredith. 2019. Interspeaker Covariation in Philadelphia Vowel Changes. Language Variation and Change, 31(2), 119–133. Tamminga, Meredith, MacKenzie, Laurel, and Embick, David. 2016a. The Dynamics of Variation in Individuals. Linguistic Variation, 16(2), 300–336. Tamminga, Meredith, Ahern, Christopher, and Ecay, Aaron. 2016b. Generalized Additive Mixed Models for Intraspeaker Variation. Linguistics Vanguard, 2(s1). Tanaka, Mikihiro N., Branigan, Holly P., McLean, Janet F., and Pickering, Martin J. 2011. Conceptual Influences on Word Order and Voice in Sentence Production: Evidence from Japanese. Journal of Memory and Language, 65(3), 318–330. Taylor, John. 1989. Possessive Genitives in English. Linguistics, 27, 663–686. Taylor, John R. 1996. Possessives in English. Oxford: Clarendon Press. Temperley, David. 2007. Minimization of Dependency Length in Written English. Cognition, 105, 300–333. Temperley, David. 2008. Dependency-Length Minimization in Natural and Artificial Languages. Journal of Quantitative Linguistics, 15(3), 256–282. Theijssen, Daphne. 2012. Making Choices Modelling the English Dative Alternation. Nijmegen: UB Nijmegen. Theijssen, Daphne, ten Bosch, Louis, Boves, Lou, Cranen, Bert, and van Halteren, Hans. 2013. Choosing Alternatives: Using Bayesian Networks and Memory-Based Learning to Study the Dative Alternation. Corpus Linguistics and Linguistic Theory, 9(2), 227–262. Thim, Stefan. 2012. Phrasal Verbs: The English Verb-Particle Construction and Its History. Berlin: Mouton de Gruyter.

214

References

Thomas, Russel. 1931. Syntactical Processes Involved in the Development of the Adnominal Periphrastic Genitive in the English Language. Ph.D. dissertation, University of Michigan. Thompson, Sandra A. 1990. Information Flow and Dative Shift in English Discourse. Pages 239–253 of Edmondson, Jerold A., Feagin, Crawford, and Mühlhäusler, Peter (eds.), Development and Diversity: Language Variation across Time and Space. A Festschrift for Charles-James N. Bailey. Arlington: SIL and University of Texas. Thothathiri, Malathi, and Rattinger, Michelle G. 2016. Acquiring and Producing Sentences: Whether Learners Use Verb-Specific or Verb-General Information Depends on Cue Validity. Frontiers in Psychology, 7. Tiv, Mehrgol, Gonnerman, Laura, Whitford, Veronica, Friesen, Deanna, Jared, Debra, and Titone, Debra. 2019. Figuring Out How Verb-Particle Constructions Are Understood During L1 and L2 Reading. Frontiers in Psychology, 10, 1733. Tooley, Kristen M., and Bock, Kathryn. 2014. On the Parity of Structural Persistence in Language Production and Comprehension. Cognition, 132(2), 101–136. Torres Cacoullos, Rena, and Travis, Catherine E. 2014. Prosody, Priming and Particular Constructions: The Patterning of English First-Person Singular Subject Expression in Conversation. Journal of Pragmatics, 63(Mar.), 19–34. Torres Cacoullos, Rena, and Walker, Torres Cacoullos. 2009. The Present of the English Future: Grammatical Variation and Collocations in Discourse. Language, 85(2), 321–354. Travis, Catherine E., and Torres Cacoullos, Rena. 2012. What Do Subject Pronouns Do in Discourse? Cognitive, Mechanical and Constructional Factors in Variation. Cognitive Linguistics, 23(4), 711–748. Trudgill, Peter. 1996. Dialect Typology: Isolation, Social Network and Phonological Structure. Page 3 of Guy, Gregory R., Feagin, Crawford, Schiffrin, Deborah, and Baugh, John (eds.), Towards a Social Science of Language: Papers in Honor of William Labov. Volume 1: Variation and Change in Language and Society. Amsterdam: Benjamins. Trudgill, Peter. 2001. Contact and Simplification: Historical Baggage and Directionality in Linguistic Change. Linguistic Typology, 5, 371–374. Trudgill, Peter. 2004. Dialect Contact and New-Dialect Formation: The Inevitability of Colonial Englishes. Edinburgh: Edinburgh University Press. Trudgill, Peter. 2008. Colonial Dialect Contact in the History of European Languages: On the Irrelevance of Identity to New-Dialect Formation. Language in Society, 37, 241–280. Trudgill, Peter. 2009. Vernacular Universals and the Sociolinguistic Typology of English Dialects. Pages 302–329 of Filppula, Marrku, Klemola, Juhani, and Paulasto, Heli (eds.), Vernacular Universals and Language Contacts: Evidence from Varieties of English and Beyond. London: Routledge. Trudgill, Peter. 2011. Sociolinguistic Typology: Social Determinants of Linguistic Complexity. Oxford: Oxford University Press. van Hout, Roeland, and Muysken, Pieter. 2016. Taming Chaos. Chance and Variability in the Language Sciences. Pages 249–266 of Landsman, Klaas, and van Wolde, Ellen (eds.), The Challenge of Chance. Cham: Springer International Publishing. van Kemenade, Ans. 1987. Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Verhagen, Véronique, Mos, Maria, Backus, Ad, and Schilperoord, Joost. 2018. Predictive Language Processing Revealing Usage-Based Variation. Language and Cognition, 10(2), 329–373.

References

215

Verhoeven, Elisabeth. 2014. Thematic Prominence and Animacy Asymmetries. Evidence from a Cross-Linguistic Production Study. Lingua, 143(May), 129–161. Verner, Karl. 1877. Eine Ausnahme der ersten Lautverschiebung. Zeitschrift für vergleichende Sprachforschung auf dem Gebiete der Indogermanischen Sprachen, 23(2. H), 97–130. Visser, Frederikus Theodorus. 1963. An Historical Syntax of the English Language. Leiden: Brill. Wälchli, Bernhard. 2014. Algorithmic Typology and Going from Known to Similar Unknown Categories within and across Languages. In Szmrecsanyi, Benedikt, and Wälchli, Bernhard (eds.), Aggregating Dialectology, Typology, and Register Analysis. Berlin: Mouton de Gruyter. Walker, Abby. 2019. The Role of Dialect Experience in Topic-Based Shifts in Speech Production. Language Variation and Change, 31(2), 135–163. Walker, Abby, and Campbell-Kibler, Kathryn. 2015. Repeat What after Whom? Exploring Variable Selectivity in a Cross-Dialectal Shadowing Task. Frontiers in Psychology, 6. Walker, James A. 2007. “There’s Bears Back There”: Plural Existentials and Vernacular Universals in (Quebec) English. English World-Wide. A Journal of Varieties of English, 28(2), 147–166. Ward, Joe H. Jr. 1963. Hierarchical Grouping to Optimize an Objective Function. Journal of the American Statistical Association, 58, 236–244. Wasow, Thomas, and Arnold, Jennifer. 2003. Post-Verbal Constituent Ordering in English. Pages 119–154 of Rohdenburg, Günter, and Mondorf, Britta (eds.), Determinants of Grammatical Variation in English. Amsterdam: Mouton de Gruyter. Waters, Cathleen, and Tagliamonte, Sali A. 2017. Is One Innovation Enough? Leaders, Covariation, and Language Change. American Speech, 92(1), 23–40. Wee, Lionel. 2014. The Evolution of Singlish in Late Modernity: Beyond Phase 5? Pages 126–141 of Buschfeld, Sarah, Hoffmann, Thomas, Huber, Magnus, and Kautzsch, Alexander (eds.), The Evolution of Englishes: The Dynamic Model and Beyond. Varieties of English Around the World. Amsterdam: Benjamins. Weiner, Judith, and Labov, William. 1983. Constraints on the Agentless Passive. Journal of Linguistics, 19, 29–58. Weinreich, Uriel, Labov, William, and Herzog, Marvin. 1968. Empirical Foundations for a Theory of Language Change. Pages 98–127, 150–151, 183–195 of Lehmann, Winfred, and Malkiel, Yakov (eds.), Directions for Historical Linguistics. Austin: University of Texas Press. Wekker, Herman. 1996a. Creole Languages and Language Acquisition. Berlin: Mouton de Gruyter. Wekker, Herman. 1996b. Creolization and the Acquisition of English as a Second Language. Pages 139–149 of Wekker, Herman (ed.), Creole Languages and Language Acquisition. Berlin: Mouton de Gruyter. Williams, Jessica. 1987. Non-Native Varieties of English: A Special Case of Language Acquisition. English World-Wide, 8(2), 161–199. Williams, Robert S. 1994. A Statistical Analysis of English Double Object Alternation. Issues in Applied Linguistics, 5(1), 37–58. Wolfram, Walt. 1976. Towards a Description of a-Prefixing in Appalachian Englisn. American Speech, 51(1/2), 45–56. Wolfram, Walt, and Schilling-Estes, Natalie. 1998. American English: Dialects and Variation. Oxford: Blackwell.

216

References

Wolk, Christoph, Bresnan, Joan, Rosenbach, Anette, and Szmrecsanyi, Benedikt. 2013. Dative and Genitive Variability in Late Modern English: Exploring CrossConstructional Variation and Change. Diachronica, 30(3), 382–419. Wulff, Stefanie, and Gries, Stefan Th. 2019. Particle Placement in Learner Language. Language Learning, 69(4), 873–910. Wulff, Stefanie, Lester, Nicholas, and Martinez-Garcia, Maria T. 2014. That-Variation in German and Spanish L2 English. Language and Cognition, 6(02), 271–299. Zaenen, Annie, Carletta, Jean, Garretson, Gregory, Bresnan, Joan, Koontz-Garboden, Andrew, Nikitina, Tatiana, O’Connor, Mary Catherine, and Wasow, Thomas. 2004. Animacy Encoding in English: Why and How. Pages 118–125 of Byron, Donna, and Webber, Bonnie (eds.), Proceedings of the 2004 ACL Workshop on Discourse Annotation, Barcelona, July 2004. East Strouddsburg: Association for Computational Linguistics. Zipp, Lena, and Bernaisch, Tobias. 2012. Particle Verbs across First and Second Language Varieties of English. Pages 167–196 of Hundt, Marianne. and Gut, Ulrike (eds.), Mapping Unity and Diversity World-Wide: Corpus-Based Studies of New Englishes. Amsterdam: Benjamins. Zuur, Alain F., Ieno, Elena N., Walker, Neil J., Saveliev, Anatoly A., and Smith, Graham M. 2009. Mixed Effects Models and Extensions in Ecology with R. Cham: Springer International Publishing. Zwicky, Arnold M. 1987. Suppressing the Zs. Journal of Linguistics, 23, 133–148.

Index

acquisition, 27, 30, 37, 42, 131, 173–176, 180 alternation, 1–3, 6, 10–13, 15–33, 82–84, 87, 89, 92–94, 96–99, 101–104, 106, 108–112, 114–117, 119–132, 134–137 animacy, 3, 10, 19, 20, 22, 24–26, 28–30, 45, 76, 83, 85–91, 107–109, 115, 116, 136, 169, 172, 185–186 annotation, 10, 31, 84, 85, 94, 100 British English, 1, 10, 16, 19–21, 24, 25, 29, 31, 46, 47, 85–95, 97–102, 108–110, 112, 122, 124, 125, 128, 129, 132–134, 142, 144, 152, 174, 175 Canadian English, 1, 6, 10, 19, 29, 47, 85–87, 89, 90, 92–95, 100, 101, 108, 109, 122, 124, 128, 129, 133, 134, 175 coherence, 11, 112, 114, 120, 123–129, 136, 137, 170, 172, 181–183 comparative linguistics, 1, 5, 19, 20, 23, 113, 170 Comparative Sociolinguistics, 3, 5, 10, 19, 29, 112–115, 117, 127, 135, 137 corpus linguistics, 4, 58–60, 82, 112, 131 Corpus of Global Web-based English, 2, 10, 21, 22, 61–65, 81, 82, 84, 93, 98, 122, 126, 165, 168 dative (alternation), 2, 17, 19–22, 25, 27–31, 33, 68–69, 92–98, 101, 108–111, 115–117, 121, 122, 124, 126–130, 136, 137, 168, 169, 177–179 definiteness, 24, 26, 28–30, 76, 84, 94, 97, 98, 100, 116 dialect typology, 1, 6, 9, 10, 42–46, 171–174, 185 dialectology, 1, 3, 5, 6, 12, 14–15, 132 dialectometry, 6, 10, 11, 112, 113, 124, 131, 135

Distance Based Coherence (DBC), 123, 124, 126, 127, 136, 137, 140 distance(s), 6, 90–92, 98, 99, 101, 105–107, 112, 117–121, 123–138 Dynamic Model, 37–41, 46, 110, 137, 167, 173, 174 end weight, see length final sibilancy, 24, 26, 80, 85, 87–92, 108, 109, 116 frequency, 9, 23, 79, 83, 85, 86, 94, 98, 103, 113, 116, 138 genitive (alternation), 2, 17, 19–27, 33, 45, 46, 66–68, 80, 84–92, 101, 108–111, 115–117, 121–123, 126, 127, 129, 130, 135–137, 168, 169, 172, 176–177 genre, 25, 28, 59, 61, 63, 64, 88, 95, 101, 102, 168, 169 givenness, 28, 30, 32, 77, 94, 100, 115, 116 Hong Kong English, 1, 6, 9, 10, 21, 25, 48, 85, 86, 88–95, 97, 98, 100, 101, 108, 109, 122, 128, 132, 133, 176 Indian English, 1, 6, 10, 20, 21, 25, 28, 29, 49, 85, 86, 88, 89, 93–95, 97, 98, 100–103, 108, 109, 112, 122, 125, 128, 129, 132, 133, 142, 152, 177–179 indigenization, 3, 21, 44, 49, 50, 52, 54, 82, 110, 111, 135, 172, 173, 179, 189 Inner Circle, 3, 9, 10, 82, 85–88, 91–95, 98–101, 103–110, 112, 122, 123, 129–133, 136, 137, 167–174 International Corpus of English, 10, 21, 25, 28, 31, 61–65, 81, 82, 122, 126, 144, 145, 165, 168

217

218

Index

Irish English, 1, 10, 31, 50, 85–87, 93–95, 97, 100, 101, 103, 122, 129, 132, 133, 174 Jamaican English, 1, 10, 51, 85, 86, 88, 93–95, 97, 100–102, 122, 129, 132, 134, 176 language change, 37–39, 173 language contact, 37–40, 42, 45, 173, 176–179 Language Variation and Change (LVC), see variationist sociolinguistics length, 3, 10, 20, 22, 24–26, 28, 29, 32, 78, 85–89, 94–98, 100–109, 116, 136, 142, 151, 152, 169–171 multidimensional scaling (MDS), 128–132, 137, 138, 140, 171 nativization, see indigenization NeighborNet, 132–134, 137 New Zealand English, 1, 6, 9, 10, 19, 20, 25, 28, 52, 85–87, 89, 90, 93–95, 100, 101, 103, 122, 125, 128, 129, 132–134, 142, 152, 175 NP type, 77, 85–89, 116 Outer Circle, 9, 10, 21, 82, 85–88, 91–95, 98–101, 103–106, 108–110, 112, 122, 123, 129, 131–133, 136, 137, 167–173, 176–180 particle placement (alternation), 2, 3, 17, 21, 22, 28, 30–33, 70–73, 80, 81, 98–106, 108–111, 116, 117, 121, 122, 124–127, 129, 130, 136, 137, 142, 164, 168, 169, 179–180 Philippines English, 1, 10, 21, 25, 53, 85, 86, 88–91, 93–95, 97, 98, 100–102, 108, 109, 122, 129, 132–134 probabilistic grammar, 1, 3, 5, 7–8, 10, 11, 26, 31, 44–46, 82–84, 88, 90, 98, 108–110, 112–115, 120, 128–131, 135, 138, 166–167, 173–175, 183–186

probabilistic linguistics, 1, 7–8, 172, 175 pronominality, 17, 27, 28, 30, 31, 94, 97, 98, 107–109, 116, 169, 170 psycholinguistics, 1, 3, 8–9, 27, 109, 115, 163–164, 168, 173, 174 random forest, 10, 21, 25, 29, 31, 82, 83, 106, 119, 168 rating task, 2, 3, 5, 8, 9, 28, 143, 171 register, see genre regression, 8, 10, 11, 19–21, 24, 25, 28, 29, 31, 82–84, 88–92, 95, 97–99, 103–110, 115–120, 138, 151, 169 semantic compositionality, 80, 101, 103–105, 108, 116 semantic relation, 23, 24, 26, 45, 80, 86, 87, 116 Singapore English, 1, 10, 16, 21, 25, 46, 53, 85, 86, 88–95, 100, 101, 103, 108, 109, 122, 129, 131–134, 142, 152 surprisal, 8, 81, 101–104, 116, 169 thematicity, 79 Three Circles Model, 35–37 transmission, see acquisition universals, 42–43, 85, 135, 137, 171–173, 175, 184–185 usage-based linguistics, 1, 7, 9, 11, 27, 135, 141, 174–175 variable context, 17, 23, 65–74, 84, 93, 99, 168 Variation-Based Distance and Similarity Modeling (VADIS), 11, 112–117, 121–127, 135, 137–140, 170–172, 174, 187, 189 variationist sociolinguistics, 1, 4–5, 7, 8, 10–12, 14–15, 57–58, 112–114, 116, 123, 133, 180–186 World Englishes, 3, 9–11, 20, 34–55, 135, 172