Contrastive Register Variation: A Quantitative Approach to the Comparison of English and German 9783110238594, 9783110238587

Table of contents :
Acknowledgements
List of tables and figures
List of abbreviations
Chapter 1. Introduction
1 The concept of variation in linguistics
2 Goal of the study
3 Methodology for the investigation of variation
4 Organisation of the study
Part 1: Theoretical foundations
Chapter 2. State of the art
1 Texts: a view on text linguistics
2 Register: situating texts in the language system
3 Variation: comparing linguistic systems
3.1 Contrastive linguistics —
3.2 Lexico-grammatical variation across languages —
3.3 Register variation across languages —
4 Translation: influences from language mediation
4.1 Register analysis in translation studies —
4.2 Translation properties —
5 Envoi
Chapter 3. On empirical methods in linguistics
1 Conceptual and methodological links between linguistics and social sciences
2 Empirical methods
2.1 Qualitative and quantitative approaches
2.2 From abstract concepts to observable features
3 Quality criteria for empirical research
4 Limits of cross-linguistic empirical research
5 Summary
Chapter 4. Indicators of register classification
1 Field of discourse
1.1 Experiential domain
1.2 Goal orientation
2 Tenor of discourse
2.1 Agentive roles
2.2 Social role relationship
2.3 Social distance
2.4 Appraisal
3 Mode of discourse
3.1 Language role
3.2 Channel
3.3 Medium
4 Summary
Part 2: Empirical analysis
Chapter 5. Research design
1 The corpus
1.1 Design
1.2 Enrichment
2 Exploiting the corpus
2.1 Query tools
2.2 Statistics
3 Corpus contrasts
3.1 Overview of the contrasts
3.2 Relative register values
4 Hypotheses
4.1 Language-internal variation in originals
4.2 Cross-linguistic variation in originals
4.3 Variation between originals and translations
5 Summary
Chapter 6. English intralingual register variation
1 Field of discourse
1.1 Experiential domain
1.2 Goal orientation
2 Tenor of discourse
2.1 Social role relationship
2.2 Social distance
3 Mode of discourse
3.1 Language role
3.2 Medium
4 Summary
Chapter 7. German intralingual register variation
1 Field of discourse
1.1 Experiential domain
1.2 Goal orientation
2 Tenor of discourse
2.1 Social role relationship
2.2 Social distance
3 Mode of discourse
3.1 Language role
3.2 Medium
4 Summary
Chapter 8. Cross-linguistic register comparison
1 Field of discourse
1.1 Experiential domain
1.2 Goal orientation
2 Tenor of discourse
2.1 Social role relationship
2.2 Social distance
3 Mode of discourse
3.1 Language role
3.2 Medium
4 Summary
Chapter 9. Variation between originals and translations
1 Field of discourse
1.1 Experiential domain
1.2 Goal orientation
2 Tenor of discourse
2.1 Social role relationship
2.2 Social distance
3 Mode of discourse
3.1 Language role
3.2 Medium
4 Summary
Chapter 10. Insights on register variation
1 Variation in the CroCo registers
1.1 Variation in the English registers
1.2 Variation in the German registers
1.3 Cross-linguistic variation
1.4 Variation between the originals and translations
2 The hypotheses revisited
3 Towards a model of register variation
Chapter 11. Conclusions
1 Summary of the results
2 Evaluation of the methodology
3 Outlook
References
Appendix: Linguistic features in alphabetic order
Index

Citation preview

Stella Neumann Contrastive Register Variation

Trends in Linguistics Studies and Monographs

Editor Volker Gast Editorial Board Walter Bisang Jan Terje Faarlund Hans Henrich Hock Natalia Levshina Heiko Narrog Matthias Schlesewsky Amir Zeldes Niina Ning Zhang Editor responsible for this volume Volker Gast

Volume 251

Stella Neumann

Contrastive Register Variation A Quantitative Approach to the Comparison of English and German

ISBN 978–3-11–023858–7 e-ISBN 978–3-11–023859–4 ISSN 1861–4302 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliografische Information der Deutschen Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutschen Nationalbibliografie; detailed bibliographic data are available in the internet http://dnb.dnb.de. © 2014 Walter de Gruyter GmbH, Berlin/Boston Typesetting: Frank Benno Junghanns, Berlin Printing: Hubert & Co. GmbH & Co. KG, Göttingen ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

To Stefan

Acknowledgements Variation may be, as we will soon see, a pervasive motif in language. It is, however, completely inappropriate to describe the invariant support I have received from many people. This corpus of supporters can be segmented into a number of subgroups, which do, very much to the dismay of the taxonomist, overlap to a large extent. At the same time, there are some subgroups which stand out significantly, being only made up of one single item. First and foremost, there is the ERICH subgroup consisting of a single, even singular item, my mentor Erich Steiner. The item in this group has a knack for creating a working environment which is at the same time cheerful and profoundly academic. Next, there is the SILVI subgroup, again consisting of one single item, my colleague and dear friend (cf. the FRIEND subgroup) Silvia Hansen-Schirra with whom I will never share an office again not because we would not like to but because she is such a gifted scholar that she has long outgrown the limited confines of our small office. The members of the ERICH and SILVI groups are also founding parts of the CROCO subgroup and as such were my partners in conceiving many exciting analyses of our translation corpus. The present study owes its existence to this subgroup. Yet another single item group is PROOF, comprising my meticulous proof reader, Mary Mondt, who spent many a day (and night) straightening out my complicated sentences. I am very lucky to categorise the member of this subgroup also as part of the FRIEND subgroup. The COLLEAG subgroup was not only instrumental in conducting the present research but, more importantly, was just great fun to work with. Here are the core members: Oliver Čulo (see also SCRIPT), Peggy Daut, Kerstin Kunz, Karin Maksymski, Silke Theison (see also SCRIPT), Mihaela Vela (see also SCRIPT). Affiliated to this group is also Vahram Atayan who, among other things, took over my class at a decisive point in time. The members of the SCRIPT group are not script girls running around movie sets but rather a number of very talented people, most of them computational linguists, who helped me a great deal with getting all the linguistic information out of the corpus. Apart from the members already mentioned in connection with the COLLEAG subgroup, an important member is Christopher Brosch (cf. FRIEND). All the quantities of linguistic output extracted from the corpus with the help of the SCRIPT subgroup would be meaningless without subjecting them to some wicked statistical tests. This is where the STATS group comes into the picture: Its members Florian Frötscher and Stefan Evert provided me with some significant help on statistics. 1 Acknowledgements

viii 

 Acknowledgements

The CROCO subgroup was already mentioned before. It also comprises several generations of great student assistants: Lina Beckmann, Marlene Kast, Annette Klinger, Kristin Lorson, Katrin Menzel, Jones Ofori, Christine Ortinau, Heike Przybyl, Annett Späthe, Andrea Stockero. Special thanks go to Sarah Cieslik, Sandra Hansen and Yvonne Klein. Connected with the CROCO subgroup is also the DFG group whom I would like to thank for supporting my research as part of the project Linguistic properties of translations – a corpus-based investigation for the language pair EnglishGerman (STE 840/5–1 and -2 and HA 5457/1–2) and in particular for allowing me to leave the project for a visit to Sydney which had an important impact on the final shape of this study. This brings me to the next subgroup, the SYD group: The members of this subgroup, particularly Christian Matthiessen (now at The Hong Kong Polytechnic University), David Butt, Annabelle Lukin and Alison Moore and the CLSL people at Macquarie University as well as Jon Patrick at the University of Sydney gave me some food for thought. The INT group contains a number of international colleagues, particularly Adriana Pagano, who were very patient and supportive while waiting for my contributions. More recently, the IFAAR group consists of the department of English, American and Romance studies with great colleagues and a fantastic team. I should like to mention in particular Alexandra Leistner and Sebastian Meß to whom I am grateful for their help with finalising the manuscript. I am particularly indebted to the GRUY subgroup: to Birgit Sievert for her support and coordination, to an anonymous reviewer for comments which helped me clarify some important ideas, and notably to Volker Gast, the series editor, for his patient and careful supervision of this book project. The FRIEND subgroup has many overlaps with other groups mentioned before. I would like to identify some members who were particularly instrumental in writing this study. The classification here requires the introduction of another intermediate level of categories: MONO includes Silvia Hansen-Schirra, Wiebke Janßen, Mary Mondt, Heike Scheuerpflug and Mareike Schmitt and can be characterised by its being instrumental in helping me enjoy life. Furthermore, there is the AREND group consisting of Elisabeth and Peter Arend with Daniel and Sandra who provided me with a home in every sense of the word. I couldn’t possibly ever give back all the support and luxurious food that I have received from this subgroup. The FAMI subgroup comprises a number of loving people, in particular my sister Anna Neumann (or should I say my personal acknowledgements coach?). FAMI also includes my mother Franziska Neumann, a sociologist, who actually started my interest in empirical research already when I was a child and has kept me going ever since. My brother Niko with Nicole, Eric, Phillip, Jan and the little

Acknowledgements 

 ix

one are always there to make me happy. Had my father Nikolaus Neumann lived to see this piece of writing, he would be more proud of it than I could ever be. Finally, there is one remaining subgroup requiring our undivided attention: the NEUBAU subgroup containing one very special item, my partner Stefan Baumann who is incredibly funny and a great linguist. Besides his invariant support, he helped me at every stage of this research and in particular with wrapping up writing this study. This book is dedicated to him in deep gratitude. I would like to thank the members of all subgroups for their various contributions to the completion of this research. All remaining weaknesses are, of course, solely mine.

Contents Acknowledgements — vii vii List of tables and figures — xv List of abbreviations — xx Chapter 1 Introduction — 1 1 The concept of variation in linguistics — 1 2 Goal of the study — 3 3 Methodology for the investigation of variation — 4 4 Organisation of the study — 5 Part 1: Theoretical foundations Chapter 2 State of the art — 9 1 Texts: a view on text linguistics — 9 2 Register: situating texts in the language system  — 14 3 Variation: comparing linguistic systems — 19 3.1 Contrastive linguistics — 19 3.2 Lexico-grammatical variation across languages  — 22 3.3 Register variation across languages — 24 4 Translation: influences from language mediation — 27 4.1 Register analysis in translation studies  — 28 4.2 Translation properties — 29 5 Envoi — 32 Chapter 3 On empirical methods in linguistics — 34 1 Conceptual and methodological links between linguistics and social sciences — 34 2 Empirical methods — 37 2.1 Qualitative and quantitative approaches — 39 2.2 From abstract concepts to observable features  — 43 3 Quality criteria for empirical research  — 45 4 Limits of cross-linguistic empirical research — 46 5 Summary — 47

xii 

 Contents

Chapter 4 Indicators of register classification — 48 1 Field of discourse — 48 1.1 Experiential domain — 49 1.2 Goal orientation — 54 2 Tenor of discourse — 61 2.1 Agentive roles — 63 2.2 Social role relationship — 63 2.3 Social distance — 67 2.4 Appraisal  — 71 3 Mode of discourse — 72 3.1 Language role — 72 3.2 Channel — 75 3.3 Medium — 75 4 Summary — 78 Part 2: Empirical analysis Chapter 5 Research design — 83 1 The corpus  — 83 1.1 Design — 83 1.2 Enrichment — 88 2 Exploiting the corpus — 90 2.1 Query tools — 90 2.2 Statistics — 93 3 Corpus contrasts — 100 3.1 Overview of the contrasts — 100 3.2 Relative register values — 101 4 Hypotheses — 102 4.1 Language-internal variation in originals — 102 4.2 Cross-linguistic variation in originals — 103 4.3 Variation between originals and translations — 103 5 Summary — 105 Chapter 6 English intralingual register variation  — 106 1 Field of discourse — 106 1.1 Experiential domain — 106 1.2 Goal orientation — 122 2 Tenor of discourse — 136

Contents 

Social role relationship — 136 Social distance — 148 Mode of discourse — 156 Language role — 156 Medium — 160 Summary — 164

2.1 2.2 3 3.1 3.2 4

Chapter 7 German intralingual register variation — 166 1 Field of discourse — 166 1.1 Experiential domain — 166 1.2 Goal orientation — 180 2 Tenor of discourse — 192 2.1 Social role relationship — 192 2.2 Social distance — 202 3 Mode of discourse — 208 3.1 Language role — 208 3.2 Medium — 210 4 Summary — 213 Chapter 8 Cross-linguistic register comparison — 215 1 Field of discourse — 215 1.1 Experiential domain — 216 1.2 Goal orientation — 227 2 Tenor of discourse — 233 2.1 Social role relationship — 233 2.2 Social distance — 243 3 Mode of discourse — 246 3.1 Language role — 246 3.2 Medium — 251 4 Summary — 255 Chapter 9 Variation between originals and translations — 258 1 Field of discourse — 258 1.1 Experiential domain — 258 1.2 Goal orientation — 275 2 Tenor of discourse — 282 2.1 Social role relationship — 282 2.2 Social distance — 294 3 Mode of discourse — 298

1 Contents

 xiii

xiv 

 Contents

3.1 3.2 4

Language role — 298 Medium — 300 Summary — 306

Chapter 10 Insights on register variation  — 307 1 Variation in the CroCo registers — 307 1.1 Variation in the English registers  — 308 1.2 Variation in the German registers — 312 1.3 Cross-linguistic variation — 316 1.4 Variation between the originals and translations — 318 2 The hypotheses revisited — 320 3 Towards a model of register variation — 322 Chapter 11 Conclusions — 325 1 Summary of the results — 325 2 Evaluation of the methodology — 328 3 Outlook — 330 References — 333 Appendix: Linguistic features in alphabetic order — 349 Index — 358

List of tables and figures Tables Overall size of the CroCo corpus — 86 Summary of the corpus comparisons — 101 The 10 most frequent lexical items in EO — 107 Number of types per register in EO — 110 Subgroups computed by the Tukey HSD test — 113 The 10 most frequent lexical verbs in EO — 115 Verb statistics for EO — 118 Key figures for lexical chains in two EO registers — 120 Significant results for argumentation indicators in EO — 126 Nominalisations with the suffix -ism in ER — 128 Significant results for exposition indicators in EO — 129 Significant results for instruction indicators in EO — 132 Significant results for narration indicators in EO — 135 Grammatical density in two English original registers — 146 Overview of the social distance indicators in EO — 148 Significant positive and negative indicators of neutral style in EO — 150 Significant positive and negative indicators of casual style in EO — 152 Significant positive and negative indicators of consultative style in EO — 153 Grammatical intricacy in the English FICTION and SHARE registers — 162 The 10 most frequent lexical items in GO — 167 Number of types per register in GO — 169 The 10 most frequent lexical verbs in GO — 172 Verb statistics for GO — 175 Key figures for lexical chains in two GO registers — 177 Significant results for argumentation indicators in GO — 182 Significant results for exposition indicators in GO — 184 Significant results for instruction indicators in GO — 187 Significant results for narration indicators in GO — 190 Grammatical density in two German original registers — 199 Overview of the social distance indicators in GO — 202 Significant positive and negative indicators of neutral style in GO — 203 Significant positive and negative indicators of casual style in GO — 205 Significant positive and negative indicators of consultative style in GO — 206

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

1 List of tables and figures

xvi  34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

 List of tables and figures Grammatical intricacy in the German FICTION and SHARE registers — 211 The 10 most frequent lexical items in contrastrive comparison — 217 Lexical types per register in EO and GO compared to the register mean — 218 The 10 most frequent lexical verbs in the contrastive registers — 222 Difference of verb type-token relationships between contrastive registers and ­register mean — 224 Key figures for lexical chains in two contrastive registers — 226 Registers assigned to goal types in both languages — 228 Comparison of narration indicators in FICTION — 230 Comparison of exposition indicators in SHARE — 231 Grammatical density in two contrastive registers — 241 Grammatical intricacy in the contrastive registers — 252 Percentage of selected functions in theme position in FICTION and SHARE — 254 The 10 most frequent lexical items in original and translated registers — 259 Comparison of the number of lexical types per register — 264 The 10 most frequent lexical verbs in original and translated registers — 268 Difference between subcorpus and register mean in lexical verb statistics — 270 Key figures for lexical chains in originals and translations — 273 Means and interquartile range for the difference between register and reference corpora — 283 Grammatical density of originals and translations in FICTION and SHARE — 292 Means in relation to the reference corpora for the positive indicators of casual style in FICTION — 295 Means in relation to the reference corpora for the positive indicators of neutral style in SHARE — 297 Lexical density in relation to the reference corpora in FICTION and SHARE originals and translations — 301 Grammatical intricacy in FICTION and SHARE originals and translations — 302 Percentage of selected functions in theme position in FICTION and SHARE originals and translations — 303 Personal pronouns in relation to the reference corpora in FICTION and SHARE ­originals and translations — 304 Profiles of the English original registers — 309 Profiles of the German original registers — 313

List of tables and figures 

 xvii

Figures 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Relation between different modalities in an example from EO_INSTR — 74 Overview of the CroCo corpus design — 88 System network of the basic options for mood — 91 Example of a query for imperative mood — 92 Example of a box plot for personal pronouns per all tokens in English originals (EO) — 95 Box plot of the lexical item frequency ratio across registers in EO — 112 Mean values of the lexical item frequency ratio in EO — 113 Concordance lines for relational uses of is — 117 Process type differences between EO and ER — 118 Relational processes in EO compared to ER — 123 Box plot of passives per sentences in EO and ER — 127 Material processes in EO compared to ER — 131 Mental and verbal processes in EO compared to ER — 133 Mean values of declaratives per no. of sentences in EO — 138 Mean values of imperatives per no. of sentences in EO — 139 Mean values of interrogatives per no. of sentences in EO — 140 Mean values of modal verbs per no. of sentences in EO — 142 Mean values of modal lexis per no. of tokens in EO — 143 Key words per tokens in EO — 145 Mean values of nominalisations per no. of tokens in EO — 147 Example of an ancillary role of language in EO_INSTR — 157 Example of a subsidiary role of figures in EO_INSTR — 157 Box plot of the lexical item frequency ratio across registers in GO — 170 Mean values of the lexical item frequency ratio in GO — 171 Process type differences between GR and GO — 176 Relational processes in GO compared to GR — 180 Material processes in GO compared to GR — 186 Mental and verbal processes in GO compared to GR — 188 Mean values of declaratives per no. of sentences in GO — 193 Mean values of imperatives per no. of sentences in GO — 194 Mean values of interrogatives per no. of sentences in GO — 195 Mean values of modal verbs per no. of sentences in GO — 196 Mean values of modal lexis per no. of tokens in GO — 197 Key words per tokens in GO — 198 Mean values of nominalisations per no. of tokens in GO — 201 Example of a subsidiary role of pictures in GO_TOU — 210

xviii  37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

 List of tables and figures Mean values of the lexical item frequency ratio in EO and GO — 219 Box plot of the lexical item frequency ratio across registers in EO and GO — 221 Process type differences between the reference corpora and the contrastive registers — 225 Difference between frequency of declaratives in two registers and the reference corpora — 234 Difference between frequency of imperatives in two registers and the reference corpora — 235 Difference between frequency of interrogatives in two registers and the reference corpora — 237 Difference between frequency of modal verbs in two registers and the reference corpora — 238 Difference between frequency of modal lexis in two registers and the reference corpora — 239 Key words per tokens in FICTION and SHARE — 240 Difference between frequency of nominalisations in two registers and the reference corpora — 242 Difference between frequency of imperatives in two registers and the reference corpora — 247 Difference between lexical density in two registers and the reference corpora — 249 Frequency of personal pronouns in relation to the reference corpora in FICTION and SHARE — 255 Comparison of the lexical item frequency ratio in the FICTION subcorpora — 265 Comparison of the lexical item frequency ratio in the SHARE subcorpora — 267 Process type differences between the reference corpora and the FICTION subcorpora — 272 Process type differences between the reference corpora and the SHARE subcorpora — 272 Relational processes in FICTION and SHARE compared to the reference corpora — 276 Material processes in FICTION and SHARE compared to the reference corpora — 279 Mental and verbal processes in FICTION and SHARE compared to the reference corpora — 280 Frequency of declaratives in the FICTION subcorpora compared to the reference corpora — 284

List of tables and figures 

58 60 61 62 63 64 65 66 67

 xix

Frequency of declaratives in the SHARE subcorpora compared to the reference corpora — 284 Frequency of modal verbs in the SHARE subcorpora compared to the reference corpora — 287 Frequency of modal lexis in the FICTION subcorpora compared to the reference corpora — 288 Frequency of modal lexis in the SHARE subcorpora compared to the reference corpora — 289 Key words per tokens in the FICTION and SHARE subcorpora — 289 Frequency of nominalisations in the FICTION subcorpora compared to the reference corpora — 293 Frequency of nominalisations in the SHARE subcorpora compared to the reference corpora — 293 Juxtaposition of the number of verb types and verb type-token ratio in EO — 310 Juxtaposition of the number of verb types and verb type-token ratio in GO  — 315

List of abbreviations ANOVA

analysis of variance

EO

English original texts

ER

English reference corpus

ESSAY

subcorpus of the CroCo corpus containing political essays

ETrans

English translations

F

error term (in the analysis of variance)

FICTION

subcorpus of the CroCo corpus containing literary texts

GO

German original texts

GR

German reference corpus

GTrans

German translations

INSTR

subcorpus of the CroCo corpus containing instructional texts

IQR

interquartile range

LD

lexical density

LSP

languages for specific purposes

M

mean value

p

probability (of the type I error)

POPSCI

subcorpus of the CroCo corpus containing popular scientific texts

SFL

Systemic Functional Linguistics

SHARE

subcorpus of the CroCo corpus containing corporate communication

SPEECH

subcorpus of the CroCo corpus containing prepared political speeches

TOU

subcorpus of the CroCo corpus containing tourism brochures

TTR

type-token ratio

WEB

subcorpus of the CroCo corpus containing websites

Chapter 1 Introduction

1 The concept of variation in linguistics Variation can be regarded as an intrinsic characteristic of language and is a problem relevant to all linguistic levels (Oesterreicher 2001: 1571). But what does the term ‘variation’ actually mean? In general terms, it refers to a difference or change from the usual amount or form of something (DCE 1995 s.v. variation), thus necessitating a common basis from which it departs. In terms of language, we can examine variation from several different, but possibly connected perspectives: Languages have different variants, depending on where they are spoken, resulting in different dialects (see Auer and di Luzio 1988 with an integrated view on regional and social dialects, Kortmann 2004 on dialects in a cross-linguistic perspective) and sometimes also national varieties (see, for instance, Hundt 1998). This is usually described as regional variation and is the main object of research in dialectology and the study of varieties. Languages are also used differently depending on the social stratification of societies. This area of social variation has been widely studied in sociolinguistics (Labov 1966, Trudgill 1974, Ammon et al. 2004). A further perspective of variation is the historical one. Every language changes in the course of time, simply as a matter of evolution or as a result of interaction with other languages. The study of language change (e.g. Croft 2000) thus includes grammaticalisation and lexicalisation processes (e.g. Traugott and Dasher 2005, Mair 2006, Brinton and Traugott 2005) and language contact (Thomason and Kaufman 1988 etc.). Language contact, in turn, is closely related to questions of contrastive linguistics (e.g. Johansson 2007) and language typology (e.g. Caffarel, Martin and Matthiessen 2004, Comrie 1989): Linguistic changes as a result of contact between languages presuppose a certain degree of contrastive and/or typological variation. Another form of linguistic variation is functional variation, which is determined by the activity in which language is functioning (Halliday and Hasan 1989: 41). Halliday (1978) distinguishes this type of variation according to language use from other social aspects determining variation which depend on the language user’s socio-geographic background, thus referring to dialects and sociolects. The distinction between these types of variation is also taken up by

2 

 Introduction

Quirk et al. (1985: 16), who list region and social group as variation relating to the language user and field of discourse, medium and attitude (i.e. variants of Halliday’s field, mode and tenor of discourse) as variation attributed to language use. Citing the romance linguist Eugenio Coseriu, Oesterreicher (2001: 1565) describes a similar classification of variation into ‘diatopic’ (relating to geographical and thus dialectal differences), ‘diastratic’ (referring to socio-cultural differences) and ‘diaphasic’ (relating to different styles in differing communicative situations) differences. Register variation as functional or diaphasic variation determined by the situational context is at the centre of this study. Whether an option provided by the language system is selected by speakers is not only influenced by preferences that have evolved and are constantly evolving over time in a given linguistic community. It is more specifically controlled by the specific situation in which speakers interact, creating a functional differentiation. This ‘context of situation’ is narrowed down in register theory (e.g. Halliday, McIntosh and Strevens 1964, Halliday 1978, Halliday and Hasan 1989) in terms of the part of reality that is negotiated (‘field of discourse’), the relationship between the participants in the interaction (‘tenor of discourse’) and the way language is organised to convey the intended meaning (‘mode of discourse’). The rationale underlying the concept of register variation is the assumption that people are never totally free in the way they use language, but rather that recurring situations predetermine their use of language. Speakers may choose from the options provided by the language system and talk about the same subject matter quite differently, depending on the situation in which they interact. For instance, a researcher will use language quite differently when discussing the research of a colleague in a seminar or any other formal setting than when discussing the same issues over lunch. Therefore registers were introduced into linguistic description as subsystems which filter the language system and amplify or boost certain options of the system, while at the same time blocking other options (Matthiessen 1993: 258–262). This filter on the language potential may even allow certain lexico-grammatical constructions specific to a given register that do not apply to the majority of other registers. Viewed from this top-down perspective, registers are described as subsystems of the language system. In a bottom-up perspective, registers can also be described as groupings of texts sharing similar features that can be explained in turn by the shared context of situation. The comparison of languages is typically carried out on the level of the language system in contrastive linguistics. That is, the potential offered by languages is explored and compared (e.g. König and Gast 2012 in the language pair English-German). This is a valuable exercise and has enriched our knowledge of language in general. Nevertheless, there is a well known difference between

Goal of the study 

 3

the potential provided by language systems and the choices made by speakers of a given language, particularly in terms of the more specific options filtered by different registers as described above. This becomes particularly obvious when non-native speakers express themselves in a way that is comprehensible but does not sound natural to a native speaker. These cases are not necessarily covered by contrastive linguistics as they pertain to usage. Since all of these diverse perspectives of the study of variation are realised by different linguistic means on the various linguistic levels or strata ranging from phonetics to the study of texts and discourses, it appears appropriate to call variation a central problem of linguistics (Labov 2004: 6).1 The present study concentrates on variation resulting from different situational contexts of language use, i.e. functional variation as described above. As will be shown in the following, it incorporates a contrastive comparison of functional variation and the specific type of functional variation resulting from language mediation in translation.

2 Goal of the study The goal of the study is to gain insight into the quantitative distribution of indicators that characterise language in use in three perspectives: –– the intralingual perspective which serves to compare different registers within the two languages English and German, –– the contrastive perspective which is intended to contrast similar registers across the two languages and, finally, –– translations which display a specific type of variation in comparison to nontranslated texts. The study is thus not so much interested in exploring the full extent of language systems, but in the way this potential is typically used and particularly which options are preferred in given situations. Preference here is understood as frequency of occurrence which prompts a quantitative approach to the study of variation. A quantitative approach to linguistic description entails the use of empirical methods and includes the derivation of valid indicators and the appropriate use of statistical techniques. In this perspective, the study follows the 1 Oesterreicher (2001: 1564) points out that linguists have long regarded variation as an interfering factor in linguistic reasoning which has to be eliminated (see also Labov 2004), and that this attempt in itself confirms the existence of linguistic variation.

4 

 Introduction

example set by Biber’s seminal work on the quantitative description of variation (e.g. Biber 1988, 1995, Biber et al. 1999). At the same time, however, the study is based on the assumption that the process of describing language must proceed in a dialectic fashion, building on theoretical concepts that are tested against language in use. The results thus obtained serve to modify the theory which, in turn, has to be tested again so as to arrive at a better theoretical description of language. This approach builds on the assumption that no scholarly enterprise can free itself from theoretical presuppositions (see also Oesterreicher 2001: 1564). Building on an explicit theoretical basis can serve as a valuable framework of the linguistic analysis as long as a plausible connection in terms of operationalisation between abstract concepts and concrete, observable indicators is provided. Systemic functional language theory (Halliday and Matthiessen 2004) seems well suited for this purpose, particularly in the context of cross-linguistic investigations (see chapter 3 of Teich 2003 for an outline of systemic functional linguistics and its application to crosslinguistic investigation). More concretely, the present study intends to exemplify the application of a quantitative methodology to the theoretical framework afforded by register theory as part of the systemic functional language theory. The analysis will show that this study represents a first step towards quantitative register classification rather than offering conclusive evidence. Some individual analyses presented here are more exhaustive than others, thus illustrating the dialectic process of gradually improving theoretical concepts and their operationalisations. For instance, some analyses have to be regarded as pilot studies and consequently bar the application of statistical tests in order not to suggest that the results are conclusive (see chapters 8 and 9).

3 Methodology for the investigation of variation Investigating registers as subsystems of language systems means, as mentioned earlier, examining language in use, not options of the system. We are thus concerned with analysing frequencies of occurrences of given linguistic features. This requires the analysis of a wide range of linguistic features covering all descriptive categories in large numbers of texts. This is already a difficult undertaking within one language, and the task becomes even more complicated when registers in more than one language including translations are compared. It is further complicated by questions concerning the contrastive comparability of the data and the features. These prerequisites entail specific decisions in terms of manageability of the research that will also be addressed in this study.

Organisation of the study 

 5

As mentioned above, a main methodological decision concerns the application of quantitative methods, thus where necessary trading in features that can be effectively queried at the expense of higher informativity of the feature. For instance, since a reliable quantitative annotation of process types in the systemic functional sense is not available, we use a much more restricted concept of classifying lexical verb types that is inspired by the systemic functional concept but can by no means be interpreted in terms of this highly elaborated theoretical concept. Of course, we strive to retrieve as much information as possible from the corpus under investigation claiming to cover a middle ground of explanatory power: while the analyses made in the present study are more informative than the exploitation of raw corpora, they cannot – and do not attempt to – give as much insight as a manual analysis of the same corpus would be capable of providing. This is the typical trade-off between qualitative example-based research, which focuses on the generation of hypotheses and, on the other side, quantitative corpus-based research, whose main objective is to test hypotheses. It is hoped that the present study will contribute to the testing of hypotheses within the framework of register theory. In a different perspective, the analyst is forced to choose between covering a wide range of features that are then only studied on a rather coarse level or delving into the details of a limited selection of features. Here, the present study adopts the former approach. This appears particularly suited for gaining a comprehensive overview of register variation that provides interfaces for future studies, both of more registers and/or languages and of more detailed categories of register variation. The wide range of issues covered by the study requires us to concentrate not only on the major categories of register variation, but also to exemplify the interpretation of indicators to varying degrees. The quantitative research design is coupled with the use of statistical techniques employed to describe the linguistic evidence and to determine whether the results represent systematic features of the texts or whether they can be attributed to chance. However, statistics only have an auxiliary function in the present methodology due to the important role that is accorded to the interpretation of the quantitative findings that was addressed in section 2 of this chapter.

4 Organisation of the study The remainder of the study is organised as follows. The following three chapters prepare the theoretical ground for the study. Chapter 2 gives an overview of the state of the art. It is organised in terms of increasingly large units of linguistic investigation. Starting with a brief look at the study of texts as the unit of

6 

 Introduction

variation, it proceeds to examine previous work on registers before turning to contrastive comparisons as well as the study of variation in translations. This discussion will identify some desiderata that are addressed in the present study. As explained above, the description of variation in the view taken up by the present study requires an empirical research design. Chapter 3 explores the foundations of empirical methods in linguistics, discusses qualitative versus quantitative methods, some quality criteria of empirical research and finally some specific limitations of cross-linguistic empirical studies. Chapter 4 is the final theoretical chapter and marks the transition to the empirical analysis presented in this volume. It takes up the review of register theory given in the state of the art and discusses in detail the derivation of indicators derived from the general register variables. This derivation requires examining the various subdimensions that have been addressed in different register studies in order to link them both to the abstract register variables and to observable indicators in the text. Still in the area of methodology, chapter 5 introduces the research design of the present study including the corpus design, the enrichment of the corpus, its exploitation and the principles followed in interpreting the findings. After the elaboration of the theoretical and methodological background of the study in the previous chapters, chapter 5 concludes with the discussion of hypotheses that will be tested in the empirical analysis. On the basis of the observable indicators derived in chapter 4 and after having explained the research design in chapter 5, chapters 6 to 9 will examine the corpus in all due detail. They are organised according to the three types of variation explored in this study: intralingual register variation in English (chapter 6) and German (chapter 7), cross-linguistic variation (chapter 8) and finally variation between originals and translations (chapter 9). Chapter 10 will generalise some characteristics of the different types of variations from the results of the empirical analysis. It will also discuss some new hypotheses with respect to variation in general that arise from the findings of the present study. The study will end with a conclusion in chapter 11, summarising the main findings, assessing the chosen methodology as well as providing an outlook on future work on the basis of this study. The main part of the study is complemented by the references and an appendix containing a detailed description of the queries for the indicators. Additional material such as all results of the statistical tests is available upon request.

Part 1: Theoretical foundations

Chapter 2 State of the art The present chapter aims at discussing the theoretical background of this study. It moves from the study of smaller, monolingual units to increasingly larger, cross-linguistic units. This implies the problem of research traditions in the different research communities. The German-speaking linguistics community has focused on the (individual) text as the relevant unit of research – an approach which has resulted in a distinct term for this area of research (text linguistics). The English-speaking research community, on the other hand, takes a different perspective on similar topics. Texts are often approached from the viewpoint of lexico-grammar or under the heading of discourse analysis. Unfortunately, there is only limited exchange between the research traditions because German-speaking linguists often do not read Anglophone literature and vice versa (see Muntigl and Gruber 2005 addressing this topic with respect to approaches to genre). The following discussion does not seek to reconcile the different approaches but rather to present those aspects relevant to the present study. The chapter is organised as follows: It moves from smaller, more limited research foci (the text or group of texts) to more complex considerations (register), the inclusion of more than one language (variation) and finally specific aspects of translation. The chapter will conclude by identifying open research issues resulting from this discussion to be addressed in the present study.

1 Texts: a view on text linguistics From the point of view of text linguistics two main aspects are of interest for the present research: proposals for text analysis as well as approaches to grouping texts into types. The following section will look at proposals for the analysis of texts as the highest unit of analysis in linguistics and single out two approaches. Although they are not assumed to be representative of the two research traditions addressed above, they exemplify the differing approaches to the study of texts. Approaches to text classification in the framework of text linguistics are then examined briefly. Why do we need to look at text analysis models if our work is concerned with register variation? Each register can and must be described according to typical features of texts. Registers understood as text types, i.e. texts grouped together

10 

 State of the art

on the basis of their similarities in terms of registerial/situational features, can be described by analysing each text assigned to the register under investigation. We thus apply text analysis to each of the texts. In what follows we will therefore discuss models for textual analysis as described in text linguistics. The German text linguistics research tradition (Schoenke 2000) concentrates on written texts while Anglophone research often takes into account spoken texts, if only in the criteria applied to the analysis (Thiele 2000). Text analysis in this tradition is often called discourse analysis, a term coined by Harris (1952). Brown and Yule (1983: 24) argue that discourse analysis is more concerned with the process of interaction between the participants than with the text as a product. If the process of interaction is the object of research, this must have consequences for the research design. It means that we do not study language but language users, a fact that does not become clear from Brown and Yule’s examples, which still represent linguistic products of the interaction between language users. The researcher has to gain access to the participants in the process and apply appropriate research methods like psycholinguistic experiments or interviews. If, however, process is supposed to mean the sequence of utterances, thus suggesting a focus on spoken interaction, this entails issues of transcribing spoken data, of taking into account overlaps of turns etc. (Schmidt 2004). Whether the process or the product is under scrutiny, most works occupied with text or discourse analysis investigate textual features, i.e. those features that help structure words, clauses and sentences into texts.2 Often, the overriding parameter is the theme of the text in the sense of the Prague School (Daneš 1970). Brinker (2005), as a representative of the German tradition of text linguistics, assumes the text’s theme as one of two main categories for text analysis, his other main category being the text’s function (see Heinemann and Heinemann 2002 for a similar approach). Brinker (2005: 100) bases his understanding of text function on speech act theory and defines it as the communication purpose expressed in the text by means of certain conventionalised means that are well established in the community. The emittent thus provides the recipient with instructions on how to understand the text, e.g. as an informative or an appellative text (my translation). He concedes that a text may be characterised by several functions but maintains that the “mode of communication” is determined by one dominating communicative function (Brinker 2005: 89). ­Brinker’s 2 A remarkable exception are works like Martin and Rose (2003) who cover a comprehensive range of features for the analysis of discourse on all linguistic levels understood as contributing to the understanding of discourse to a given area of interest (in their case the reconciliation process in South Africa).

Texts: a view on text linguistics 

 11

concept of function thus resembles that of Bühler (1934), who assumes an overriding function at the expense of other functions (cf. discussion of various approaches to language functions in House 1997). In comparison to Bühler, Brinker’s notion of function refers to a more concrete level of intention pursued by the sender of the text in the sense of the goal the text is supposed to achieve (see also our discussion of goal types and their relation to text types in section 1.2 of chapter 4). Brinker proposes drawing on concrete linguistic indicators in order to determine the text function, although he does so in a vague manner. He lists linguistic indicators like performative formulae, modality, evaluative elements, extra-linguistic aspects like layout as well as contextual factors such as background knowledge. However, he does not explicate which indicator points to which function and how he decides whether an indicator pointing to a given function determines the dominant function or an ancillary function. This makes his discussion more a hermeneutic interpretation of the texts than an analysis of linguistic evidence. Furthermore, Brinker concentrates on just two textual aspects (function and thematic structure). This results in a restricted description instead of a comprehensive characterization of the text, which leaves aside any factors exploring the relationship between the interactants or other interpersonal/pragmatic aspects. While ‘discourse’ for the purpose of analysis is mostly seen as more or less equivalent to ‘text’, albeit with a stronger focus on the inclusion of spoken language and on the process of interaction between sender and recipient, Widdowson (2004) assumes ‘text’ to be an epiphenomenon of ‘discourse’. According to Widdowson, it is the overt linguistic trace of a process of negotiating the passage of intended meaning, the pragmatic process of discourse realization, whereby the resources of the language code are used to engage with the context of beliefs, values, assumptions that constitute the user’s social and individual intent. (2004: 14)

Widdowson thus sees text as the symptom of pragmatic intent, the manifestation of linguistic data carrying the implication of discourse. Without referring to discourse as the pragmatic process which realizes it, any analysis of the textual product is said to be pointless (2004: 14). With reference to Halliday (1994b) and broadly to the systemic functional research tradition, he dismisses the analysis of texts, particularly one that focuses on grammar because it does not relate text to context – as the study of discourse does – and should be separated from the study of grammar (Widdowson 2004: 34). Unfortunately, Widdowson only refers to Halliday’s (1994b) Introduction to Functional Grammar, which is not concerned with the study of register or discourse but simply with grammar, and fails to make any reference to the many volumes from the systemic functional spectrum

12 

 State of the art

discussing context (cf. in particular Ruqaiya Hasan’s work on context, e.g. 1995, 1999, as well as Ghadessy 1999) and its analysis in the form of register analysis (Halliday, McIntosh and Strevens 1964, Hasan 1973, Halliday 1978, Halliday and Hasan 1989, Ghadessy 1988, 1993, Martin 1992, Steiner 2004b to name but a few). The analysis Widdowson proposes is mainly psycholinguistic in nature in that it involves, for instance, asking subjects from different socio-cultural backgrounds to read a text and rewrite it from memory (Widdowson 2004: 170). While experiments of this type may yield interesting insights into the way the subjects interpret the text in relation to their personal background, the findings obtained would necessarily be generalisable to the text’s actual reader and the validity of any conclusions drawn from the data would be limited to potential readers, i.e. the ones chosen for the purposes of the experiment. They are not any more authentic than the analyst herself, the experiment thus does not provide more insight into the actual discourse process between the sender and the actual readers (not to mention the intended reader). Nevertheless, such an experiment would give us an idea of different readings of a given text. If this remains the only method3 applied interpreting a text in its contextual setting, the procedure would fail to account for other aspects impacting on the make-up of the text besides its effect on the reader. Ideally, texts are analysed employing a suite of (empirical) methods ranging from grammatical analysis and interviews to psycholinguistic experiments. Approaches of the above type are mostly concerned with the analysis of individual texts. The next and closely related step is to investigate groups made up of similar texts and thereby arrive at a categorisation of texts. There is a considerable terminological confusion over how to call these groups of texts. In the Anglophone research community, the terms text type, register and genre seem to be prevalent with genre appearing to have different, not completely commensurate readings referring either to a rather theory-neutral concept, which seems related to the German Textsorte (e.g. Swales 1990, Bhatia 1993, Paltridge 1995) or to a concept within systemic functional linguistics that is superimposed to the concept of register (see Muntigl and Gruber 2005 and section 2 of this chapter). The German research tradition (Heinemann 2000a, b) differentiates between Textsorten and Texttypen, which broadly correspond to bottom-up and top-down classifications. Genre in the German context refers to literary or other cultural categories. ‘Genre’ is used in the present study as a broad cover term for bottomup classifications. It seems to be the term most similar to the German Textsorte 3 See Widdowson’s important distinction between “approach” (built on theoretical ideas and founded on ideological assumptions) and “method” (as a set of operational procedures, replicable in their application) (Widdowson 2004: 158).

Texts: a view on text linguistics 

 13

(Muntigl and Gruber 2005, Lee 2001). Text type seems to be used in similar ways in both research traditions addressing top-down categories on a high level of abstraction (among others Werlich 1976, Isenberg 1983). Over the course of the years, many scholars have proposed sets of criteria for categorising texts. Among these are approaches concentrating on the function of a text (e.g. Gülich and Raible 1975, Werlich 1976, Brinker 2005, Reiß 1976, Rolf 1993), approaches offering a more or less comprehensive set of features (e.g. Sandig 1975, Lux 1981), often focussing on one language (e.g. English, Gläser 1990, Swales 1990, Bhatia 1993) or a restricted domain (e.g. science and engineering, Göpferich 1995, also with reference to translation. Most of these approaches do not go beyond the study of individual texts to underpin their categorisation.4 An alternative approach is a “semantics of genre labels” (Dimter 1981: 34, our translation). In his empirical study, Dimter (1981) does not use linguistic features for his classification of German genres but counts and analyses labels of genres in a dictionary of the German language and interviews with native speakers who are asked to assign nonsense texts without understanding the contents to labels of the respective group of texts. Existing approaches to text typology (e.g. Isenberg 1983 for German) often derive their classification as an intellectual exercise conceiving abstract categories. The link to observable criteria for the assignment of concrete texts to the abstract types is often weak or missing altogether. A classification that is suited for empirical research requires a sound link between the various levels of abstraction to the indicators that are observable and that fulfil quality criteria of empirical research (see section 3 of chapter 3). Those approaches that describe groups of texts, i.e. text types, genres etc. (e.g. contributions in Adamzik 2000) in a bottom-up way tend to be corpus-based and typically use descriptive categories that are particularly useful for specific groups of texts have a limited range of application (e.g. Engberg 1997). For the most part, these approaches merely describe one genre or register (e.g. Neumann 2003, Teich 2003) or a set of closely related genres like, e.g., Trumpp (1998).5 While this is, of course, an important exercise, the combination of a range of individual corpus-based studies does not provide a coherent overview of the variation in a

4 Interestingly, Heinemann (2000a: 525) calls the early approaches to Textsorten “empirical”. His reading of the word ‘empirical’, however, does not refer to the method of classification but rather to the character of the groups classified. The linguistic classifications of these groups are based on folk labels for groups of texts. 5 Engberg (1997), Neumann (2003), Teich (2003) and Trumpp (1998) describe the respective genre in a contrastive perspective; see also Fix et al. (2001) and Lenk amd Chesterman (2005) for collections of comparative descriptions.

14 

 State of the art

given language (pair) because each study employs a different set of categories. Biber’s work (1988, 1995) is an example of how to approach a comprehensive classification of genres (or registers in his terms). However, his strictly inductive methodology results in features which are not geared to cross-linguistic comparisons (see section 3.3 of chapter 2). This can be avoided by using a set of categories that is based on functions, thus allowing a contrastive comparison. The categories described in register theory (see section 2 of this chapter; chapter 4) appear particularly well suited to these requirements not only because they are based on functions but also because they cover a broad range of criteria in referential, pragmatic and textual perspective relevant to many, if not all registers. Another area of research is text categorisation for computational purposes. Typically, this is an object of data mining and concerns the assignment of text documents to categories on the basis of the contents of the texts for example for filtering or managing texts (Sebastiani 2002). This branch of research takes a pragmatic approach to the classification, since it is not about interpreting the features that distinguish different groups of texts but serves ulterior computational goals, which makes it less interesting for linguistic research questions. More generally, text categorisation may also be used for corpus creation (Sinclair and Ball 1996), which is in turn closely related to linguistic research using corpora. Most of the classifications of texts are derived independently of a general theory of language. The placement of text classification as part of a general language theory allows linking the classification to general assumptions about language, i.e. an upward link, as well as to concrete instances and their linguistic features, i.e. a downward link. This is particularly the case for those approaches linked to the systemic functional register theory like the early attempt by Lux (1981). Approaches to register classification will be discussed below in more detail.

2 Register: situating texts in the language system The study of texts in classical text linguistics as shown above focuses on individual texts, often more specifically on individual features in texts or groups of texts. These features can be linguistic in nature, like the focus on the topic of a text (e.g. Brinker 2005), or social, like the “social and political thought relevant to discourse and language” in Critical Discourse Analysis (Fairclough 1992). Register theory, by contrast, situates texts in a general framework that views registers as a concretion of the language potential that incorporates the situational

Register: situating texts in the language system  

 15

context in which language in use is embedded and that is realised by means of lexico-grammar (Matthiessen 1993). In the systemic functional framework, genre (Hasan 1996, Bateman 2006) is often used as a category superimposed on the concept of register (see, for instance, Martin 1992, 1997, Eggins and Martin 1997; see Ventola 2005 for a concise overview of the development of register and genre theory in SFL). Martin (1992: 503) views genres as social processes (whereas Halliday 1978 equates register with social processes). In this interpretation, the notion of genre refers to stages in a social activity, among other things, and represents the link between register and the context of culture. Citing his own and Ventola’s previous work, Martin (1992: 505) explains: Genre networks would thus be formulated on the basis of similarities and differences between text structures which thereby define text types. As part of the realisation process, generic choices would preselect field, mode and tenor options associated with particular elements of text structure. Text structure is referred to as schematic structure in Martin’s model, with genre defined as a staged, goal oriented social process realised through ­register.

He justifies this higher level concept by claiming that “no culture combines field, mode and tenor variables freely” (1992: 562). Ventola (2005: 29–30) adds that the values for field, tenor and mode cannot be expected to remain unchanged throughout all the stages of an interaction, thus narrowing down the meaning of the three register variables to a very specific domain, where the experiential domain is, for instance, limited to buying goods but not wrapping them. The present study focuses on the analysis of texts as wholes, and consequently does not take into account staging within texts and instead concentrates on register, thus leaving aside any potential category above this level. The concept of register originates from the idea that context determines language use. Originally described by the anthropologist Bronislav Malinowski as two complementary sets of influences or constraints on linguistic exchanges, i.e. the context of situation and the context of culture (Malinowski 1935), context was introduced to linguistics by Malinowski’s colleague J.R. Firth as a determining factor of language (see e.g. Firth 1957, particularly chapter 14) and has developed into what has become known as British Contextualism (for an introduction see Steiner 1983, 2000a). Register emerged as the linguistic framework for describing the context of situation. For the description of a register three parameters were introduced (e.g. Halliday, McIntosh and Strevens 1964, Halliday 1978, Gregory and Carroll 1978, Halliday and Hasan 1989, Gregory 1988): field of discourse specifying the topic of the linguistic exchange in the given situation, tenor of discourse characterising

16 

 State of the art

the relationship between the participants in the situation and finally mode of discourse describing the way in which the exchange is transmitted.6 These three parameters correspond to the three metafunctions of language assumed to cover human experience as the ideational metafunction, enacting personal and social relationships as the interpersonal metafunction and organising the discursive flow (Halliday and Matthiessen 2004: 29–30). Just as situations tend to recur and thus form types, registers represent recurring ways of using language in a given situation. The language system can even be grouped into typical co-occurrences and non-occurrences according to the respective situation. Registers can thus be described as sub-systems of the language system or, when viewed from below, as types of instantiated texts reflecting a similar situation (cf. Matthiessen 1993). The concept of types (of situations or of instantiated texts) implies a certain frequency of recurrence of features or patterns. In a methodological sense this means that, strictly speaking, a description of these types requires a quantification of their characteristic features. The analysis of groups of instantiations can thus be regarded as a requirement of the theory. Otherwise, we can only describe a given specimen of the assumed type, which does not permit any statements on the type itself. The global definition of the three parameters remains essentially unchanged with some work having been carried out in this area aimed to provide a more detailed elaboration (Matthiessen 1993 gives a comprehensive overview of the development of register theory). The parameters are, however, highly abstract and cannot be observed directly in texts as instances of a given register. Therefore, from an early stage in the development of register theory (Halliday, ­McIntosh and Strevens 1964) subdimensions specifying the three register variables, like the experiential domain under field of discourse, social hierarchy under tenor of discourse and medium under mode of discourse, were discussed and gradually developed. Although these subdimensions provide necessary concretions of the highly abstract and broad register parameters, they are still too abstract to be directly observable in instances. In social research, variables like these are termed latent variables. Halliday (1978) also discusses possible subdimensions specifying the three variables. These subdimensions are briefly introduced in Halliday and Hasan (1989) and subsequently discussed further (e.g. Poynton 1985, Martin 1992, Lavid 1993, Hasan 1978, 1995, 1999, House 1997, Steiner 2004b) and taken up and sometimes modified in several descriptions of individual registers (cf. for 6 Crystal and Davy (1969: 61ff.) propose an alternative structure that was in part taken up by House (1997). Ure and Ellis (1977: 199–202) identify four “dimensions of variation” with the corresponding “language patterns” Mode, Formality, Field and Role.

Register: situating texts in the language system  

 17

instance several contributions in Ghadessy 1988 and 1993). Steiner (2004b) discusses various subdimensions. Subdimensions represent a crucial step in deriving concrete indicators in terms of features observable in texts. While an example-based study involving these latent variables may rely on direct human interpretation (even if this may leave the relationship between indicators and variables somewhat implicit), a quantitative study necessitates operationalisations in the form of observable indicators (see section 2.1 of chapter 3), i.e. an additional intermediate step of derivation as shown, for instance, by Steiner (2004b). An example from tenor of discourse (the full derivation of indicators will be discussed in chapter 4) will serve to illustrate this problem: The subdimension of social hierarchy, sometimes also referred to as power or social role relationship, has been named as one concretion of tenor (e.g. Poynton 1985, House 1997, Steiner 2004b) and interpreted as either equal or unequal. However, we cannot observe directly in linguistic data whether the relation between the interactants is equal or unequal. The assignment to one of the options is based on the conscious or unconscious perception of concrete features in the text. Furthermore, social role is a rather complex sociological notion that can only be described by approximation via an additional range of concepts like authority, gender etc. (see section 2.2 of chapter 4). Therefore, the level of concrete, measurable indicators is only accessible via a further level of abstraction. The indicators thus determined may be interpreted as pointing to either an equal or an unequal social relationship. One of the criteria used for the subclassification of the abstract construct is level of expertise (cf. Steiner 2004b), which is still not an observable variable but requires operationalisations in terms of lexico-grammatical realisations, for instance in the form of technical vocabulary. Ideally, the analyst also states which value obtains for a given interpretation, e.g. frequent use of technical vocabulary as a manifestation of a high level of expertise. The explication of all intermediate steps results in a transparent derivation of concrete, observable (and potentially also measurable) linguistic data capable of serving as evidence for the abstract construct (see chapter 4). Martin (1992), for instance, names subcategories – particularly for tenor – and relates them to concrete features on the different strata. Steiner (2004b) describes the subdimensions and even lists potential operationalisations. His (and also Martin’s 1992) discussion of indicators, however, does not elaborate on the exact relationship between the operationalisations and the subdimensions in the sense of establishing a hypothetical link (see section 2.2 of chapter 3). Steiner does illustrate the relationship by offering examples of concrete texts to which certain realisations of features apply. This helps to understand how he

18 

 State of the art

interprets the features, but it is difficult to replicate his interpretation with different texts on the basis of this description. The following example may help demonstrate this problem: The mood system may be assumed to be an operationalisation of the distinction between the spoken and the written medium (as a subdimension of mode of discourse). Since mood is a feature of every finite clause, this statement does not give us much information. Furthermore, it does not suffice to state that the imperative mood helps distinguish between spoken and written medium because only relating the imperative to medium does not tell us whether a high frequency of imperatives is indicative of a spoken or written medium. As linguists, we might have some intuitions about this, but the explication of this final step would at least benefit the intersubjective verifiability of this. Hence, it is necessary to state as precisely as possible not only which indicator can be interpreted in view of which subdimension but also which value (e.g. high frequency or relative absence) supports this interpretation. Chapter 4 attempts to do this for the present study. To summarise, we claim that there is a difference between giving an example for an intellectual application of a given category of register analysis and supplying an operationalisation for an observable indicator of the same concept. While an example only serves as an illustration of the interpretation which might be difficult to be reproduced (see section 3 of chapter 3), an operationalisation should allow its replication in the sense of an application of the same operationalisation to a different instantiation and – provided the linguistic constellation is comparable – yield the same results. The steps required to derive observable indicators resemble the inter-stratal realisations explicated by Matthiessen (1993). Conceptualising the theoretical framework of register analysis, he describes the way in which language in context is interpreted as a system of systems ordered in symbolic abstraction. (…) Each system has its own internal organization (…) but it is related to other systems in a realizational chain: it realizes a higher system (unless it is the highest system) and it is realized by a lower one (unless it is the lowest system). This chain of inter-stratal realizations bridges the gap between semiotic in high-level cultural meanings and the material, either in speaking or in writing, through a series of intermediate strata. (Matthiessen 1993: 226)

In the same way the different strata bridge the gap between the high-level meanings and low-level material, the different methodological steps in the deduction process reduce the distance between the abstract construct and the concrete linguistic evidence. Examples of these are provided, for instance, by HansenSchirra, Neumann and Steiner (2007).

Variation: comparing linguistic systems 

 19

Given these methodological aspects are taken into consideration in a study, register analysis is particularly well suited for a comprehensive analysis of texts and groups of texts because it provides a transparent derivation of indicators that clearly reflect abstract concepts. The functional orientation of the framework in combination with concepts that can be assumed to apply languageindependently makes register analysis particularly well suited for multilingual studies (see Neumann 2003, Teich 2003). In the following section, aspects of multilingual research will be discussed in more detail.

3 Variation: comparing linguistic systems As discussed in the introduction, variation can be understood in different ways. When we look at a certain amount of texts within one language system, variation represents the differences between these texts and possibly also between whole groups of texts that share similarities. Beyond this, variation can also refer to the differences (and similarities) between language systems. The following section discusses various approaches to (contrastive) variation. For the sake of completeness: Variation may also refer to diachronic changes or regional differences within a given language. The present study, however, concentrates on a synchronic consideration of the standard variety of the two languages involved. Therefore the present section concentrates on studies covering cross-linguistic synchronic variation. We will first look at contrastive linguistics and how it typically investigates interlingual variation before turning to two individual studies that are of specific importance to the present approach to variation. These studies will be discussed following the discussion of contrastive linguistics.

3.1 Contrastive linguistics Contrastive linguistics is the scientific discipline concerned with the comparison of language systems. Whereas language typology (e.g. Greenberg 2005, Comrie 1989, Haspelmath 2005, Caffarel, Martin and Matthiessen 2004) is chiefly concerned with the identification of general similarities between languages (universals) and the classification of languages into general types, the aim of contrastive linguistics is to identify commonalities and differences between two given languages (cf. König 1990). Studies also include comparisons of individual features (frequently using corpus data). The initial purpose of contrastive linguistics was to use the findings in foreign language teaching as it was hoped that

20 

 State of the art

errors could be predicted on the basis of the differences between two given languages (see, for instance, Fisiak 1981, 1984). However, soon two main obstacles were identified: firstly, it was argued that not all errors that were predictable from the differences would actually be made by learners, and secondly, errors could occur that were not attributable to language differences (see König and Gast 2012: 1–2). The comparison of languages is, however, especially relevant to translation and interpreting, since differences between the languages involved are potential sources of problems in the translation process and knowledge of them will help – if not predict but at least – explain discrepancies between source and target texts (see section 4.2 of chapter 2). In the early days of translation studies, a strong research focus was placed on these problems (cf. the stylistiques comparées, the contrastive stylistics by Vinay and Darbelnet 1958 and Malblanc 1961). Contrastive linguistics has thus exerted an influence of translation studies and, conversely, translations are taken into account in contrastive linguistics. The problem of mapping comparable linguistic units is often solved by employing translations as a basis of comparison (e.g. James 1980: 178, see also Johansson 2003: 35).7 This seems to be a somewhat adventurous approach considering the fact that translators may resort to altogether new structures not related to the respective structure in the source text when confronted with contrastive divergences (see the large volume of work in translation studies on so called translation procedures, e.g. Vinay and Darbelnet 1958, Catford 1965, Newmark 1988, Leuven-Zwart 1989 etc.). Johansson (2003: 35), however, points out that the use of balanced corpora improves the validity and reliability (see section 3 of chapter 3) of this type of research (cf. Malmkjaer 1998 on the interaction between contrastive linguistics and translation studies). Another adaptation of this method is employed by Doherty (1991, 1996, 1999, 2002, 2006), who discusses revised translations (what the author calls “control paraphrases”) to underpin assumptions about the languages involved. The author’s main interest is in the contrastive aspects of information structure. While Doherty’s focus is on changes in translation, the implications are discussed in terms of contrastive differences in the English and German language. Doherty’s work is an important contribution to both, contrastive linguistics and translation studies, deepening our knowledge on the contrastive differences in information structure in English and German.8 7 Granger (2003: 20), by contrast, points out that contrastive linguists often view translations as unreliable due to interference in the translation process. 8 A minor observation concerns the use of Doherty’s “discourse appropriate translations”. If used as the preferred translation strategy by a majority of translators, “discourse appropriate

Variation: comparing linguistic systems 

 21

A more traditional approach to the comparison of a given language pair is to analyse and contrast options afforded by the two language systems and compare them (Hellinger 1977, Bublitz 1978, Erdmann 1990, Mair 1995, Legenhausen and Rohdenburg 1995, Steiner and Teich 2004 etc.). König and Gast (32012) provide the most recent account of the comparison of English and German. They base the comparison on semantic notions like “‘temporal relations’, ‘inalienable possession’, ‘co-reference’ and ‘understood subjects’” and argue that nearly all of the formal and semantic distinctions used by them are also used in typological studies (2012: 5). They give a detailed account of the main areas of differences between the two language systems from the phoneme inventory to a wide range of aspects of sentence grammar. The authors do not attempt – and are somewhat critical of (2012: 319) – far reaching generalisations beyond the description of the contrasts discussed in detail. This, by contrast, is the guiding principle of Hawkins’ (1986) comparative typology of English and German. Hawkins (1986) assumes that there is a general difference between the English and the German language in the way the two languages map semantics and syntax onto each other. (…) where the surface structures (morphology and syntax) of English and German contrast, English regularly exhibits greater ‘distance’ between form and meaning in specifiable ways. I.e. English surface structures exhibit less correspondence with their semantic representations than do those of German, given fairly standard assumptions about the precise nature of these two levels. (Hawkins 1986: 6)

Hawkins (1986) attempts to prove this principle in a wide range of grammatical areas, thus subsuming the individual contrasts under the umbrella of this overarching principle. He argues that “there are precise proper subset relations between most of the contrasting structures” in English and German and that these prove his general claim that German has more surface variation, whereas English displays “a greater ambiguity (and/or vagueness) of surface forms” (1986: 121). Although this general theory holds a certain appeal, particularly because Hawkins extends it to a general hypothesis about typological variation between languages, the general validity of this theory, as claimed by Hawkins, has been disproved (see, for instance, Rohdenburg 1990, Hawkins’ reaction 1992 and König and Gast 2012: 244, 319). A final approach that should be mentioned here is the specifically translation-oriented literature on contrasts between the two languages (e.g. Friederich­ translations” would bring about a dramatic increase of levelling-out (Baker 1996) with respect to information distribution, i.e. an even stronger surge of translations to gravitate towards a mean of linguistic options while originals cover a much wider range of linguistic variation.

22 

 State of the art

1977, Königs 2000). These works particularly aim at practical use in cases of doubt during translation. They list areas of contrastive differences, typically organising their discussion by the individual phenomena described. While these works are often rich in concrete examples including translations, they make little use of linguistic terminology and do not give any explanations that might assist comprehension beyond the particular case. Let us briefly summarise the contribution of contrastive linguistics to the study of linguistic variation: It provides the foundation for the investigation of variation across languages. The use of corpora for this investigation has been championed by Johansson in particular (e.g. Johansson 1998, 2003, 2004). How­ ever, quantitative approaches typically concentrate on individual features, a more comprehensive picture of the range of quantitative variation in a given language pair can only be achieved by consolidating individual studies. More comprehensive approaches to the description of contrasts between a language pair are typically limited to the investigation of the options existing in the language system, thus they often fail to take into consideration the frequencies of the contrasted features or, in general, preferences in usage as reflected by (differing) frequencies.

3.2 Lexico-grammatical variation across languages A notable exception to the literature in contrastive linguistics limited to the comparison of the options in the language system is Elke Teich’s (2003) study. She does not only give a comprehensive and comparative overview of the main grammatical systems in the English and German language based on systemic functional linguistics, but also quantifies the differences and commonalities for one register. The fact that she employs a functional grammar theory allows her to use the function of the respective grammatical structure as the basis of comparison for the contrastive analysis. Since the author also includes translations in the corpus analysis, the findings of the study are extended beyond contrasting the two languages to statements on how the contrastive differences and commonalities impact the language in translation. Teich’s (2003) main goal is to develop “a methodology for the analysis of the lexico-grammatical properties of translations and otherwise multilingually comparable texts”, as she puts it in the preface to her study. She exemplifies the methodology for the comparison of English and German lexico-grammar in a corpus of popular-scientific texts. Like Doherty (2006), Teich is interested in findings not only relevant to translation studies but also to language typology and contrastive linguistics – albeit more explicitly: “what is at issue here is bringing together translation studies, language typology and contrastive lin-

Variation: comparing linguistic systems 

 23

guistics in order to be able to deal with the particular research questions posed” (Teich 2003: 3). She thus compares not only English originals and their German translations but also texts in the opposite translation direction as well as the originals in both languages. This broadens the range of insights to be gained because it helps isolate features of translated texts that can be ascribed to the influence of the translation process (cf. the growing amount of work on properties of translated texts, e.g. Baker 1996, Olohan and Baker 2000, Hansen 2003, Steiner 2008c, Hansen-Schirra, Neumann and Steiner 2007 etc., see also section 4.2 of this chapter). A major part of the study is devoted to a comparative description of the main grammatical systems, transitivity, mood/modality and theme, of the English and German language. Teich then proceeds to the level of instantiation, or language in use, analysing lexico-grammatical features in popular-scientific texts in the two languages. This empirical analysis is aimed at testing two hypotheses about the nature of translations, namely source language shining through and target language normalisation, for the features under investigation. The author thus combines a deductive approach to the description of the contrasted language systems with an inductive approach to corpus analysis. This is one of the main merits of Teich’s study. By combining a clear theoretical framework in the form of systemic functional grammar, which she elaborates for both, English and German, with the analysis of a balanced corpus of textual instances Teich overcomes the typical limitations of the contrastive approach often restricted to general statements on the language systems concerned (see section 3.1 of this chapter). Furthermore, it goes beyond the limitations with respect to possible general statements of a strictly inductive approach like Biber’s (see section 3.3 of this chapter). This combined approach advances our knowledge of the two languages as it provides quantitative evidence for the register-specific realisations of options of the two language systems involved and especially because it shows in detail how contrasting language systems impact on translations. This permits Teich to predict the impact on translations from diverging options available in the language systems involved: Where the target language system offers more options than the source language system, it can accommodate interference from the source language system. Where the target language system options are more restricted, Teich predicts a trend towards compensation and potentially normalisation (Teich 2003: 218). It would be interesting to test this assumption by looking at another language pair providing differing options and to assess the effect of these divergences on translations. Teich’s description of the German language is highly insightful. All of the grammatical systems she compares contain important specifications for the

24 

 State of the art

German language that deviate from those in the English system networks (based mainly on Halliday 1994b and Matthiessen 1995). For instance, the description of the German imperative as well as the voice system shed light on some clear contrastive differences in the more delicate areas of the system networks.9 However, particularly when working in a given theoretical framework like systemic functional linguistics, as done by Teich, the obvious procedure is to use an existing description (English) as a blueprint for the description of a new language (here, German). As a consequence, features that play a marginal role in the model language may be neglected although they represent a more significant factor in the other language (see also Hansen-Schirra and Neumann 2003). This applies, for instance, to the German subjunctive, or in more general terms, verb mood. Teich’s description of German does not account for the subjunctive although it is a part of the grammatical system of mood and also has a bearing on the system of modality in the German language. From the point of view of the present study, her use of register analysis ap­­ pears somewhat superficial in that it is restricted to a rough characterisation of texts serving as a basis of comparison to establish comparability of the English and German texts. This should not be held against the study since it does not claim to provide a fully-fledged account of the register of popular-scientific writing. Teich’s work is mainly concerned with the influence of contrasting language systems on translations. She is not chiefly interested in the more general variation of registers in a language pair. This aspect is, however, an interesting and important field that provides a whole new level of linguistic insight into variation within one language and across languages. Douglas Biber’s work in this area remains unparalleled and will be discussed in detail in the following section.

3.3 Register variation across languages When carrying the investigation beyond not only individual registers, but also individual languages, there is no alternative to a quantitative, corpus-based research design. Douglas Biber’s work (1988, 1989, 1995) in this field has had a major impact – and has also attracted criticism (cf. McEnery and Hardie 2012: 111–115 for a summary). In 1988, Douglas Biber published a corpus-based study on variation across speech and writing. This study proposes a genuinely inductive approach to the investigation of the continuum of spoken and written discourse in English. Biber selected 67 mainly lexical and grammatical features 9 This results in her assumption that the two languages are comparable in the more indelicate ends of the system networks with the discrepancies appearing only in the more delicate areas.

Variation: comparing linguistic systems 

 25

mentioned in the relevant literature as indicators for the spoken-written continuum. He combined the LOB corpus (Johansson et al. 1978), the London-Lund corpus (Greenbaum and Svartvik 1990) and – in Biber (1995) – a collection of personal and professional letters and annotated the whole corpus with linguistic information on the 67 features. Biber proceeded to conduct a factor analysis of the frequencies of the features to determine their co-occurrences. The resulting groupings of co-occurring features are interpreted in the form of factors or dimensions. Factor analysis is one of the multivariate statistical techniques serving to reduce large data sets to a smaller number of factors that are assumed to reflect patterns of relationships in the data. It thus works largely inductively since, apart from the theoretical assumptions implicit in the selection of the linguistic features analysed in the data sets, factor analysis does not rely on any further derivation from abstract concepts. The whole study, from the linguistic features to their interpretation in terms of dimensions, is taken up in Biber (1995) under the heading of register variation.10 It needs to be stressed here that Biber does not alter the research layout in this more recent study: In fact, he does not introduce any new features but rather reports the results of the 1988 study of spoken and written language, changing the focus of the interpretation to a more general comparison of registers in English. He also retains the same seven dimensions used in the 1988 study, i.e. Involved versus Informational Production, Narrative versus Non-Narrative Concerns, Explicit versus Situation-Dependent Reference, Overt Expression of Persuasion, Abstract versus Non-Abstract Information, On-line Informational Elaboration, Academic Hedging. Biber explains that the variation between different registers is reflected in their position along these seven dimensions. The main focus of the 1995 study is on the comparison of four different languages: English, Nukulaelae Tuvaluan, Korean and Somali. After a discussion of the dimensions identified independently for each language, Biber compares the dimensions across the four languages. Due to his inductive method the dimensions are not directly comparable across the four languages. Biber therefore intro­ duces so-called communicative functions without, however, further clarifying this term. With respect to the oral/literate dimension11 he names four types marking 10 Note that Biber’s notion of register is very broad and does not incorporate the detailed framework of concepts employed in systemic functional register analysis as discussed in section 2 of this chapter. 11 “The term oral refers to stereotypically spoken discourse – that is, conversation – while the term literate refers to stereotypically written discourse – that is, informational exposition.” (Biber 1995: 238)

26 

 State of the art

interactiveness, production circumstances, stance and finally functions particular to each language. Two additional functions refer to narration and argumentation/ persuasion. Biber then compares the monolingual dimensions, their relevant features and characteristic registers in the four languages along these functional types. In the next step, he compares what he labels ‘equivalent registers’ (1995: 237) across the languages on the basis of their respective position on the dimension. For instance, face-to-face conversations are compared along the six functional types. A final set of intralingual interpretations is concerned with the internal variation of the registers and with what Biber calls ‘text types’, i.e. groups of texts based on their similarity in terms of linguistic features. Biber thus offers a far-reaching discussion of different aspects of intra- and interlingual register variation. His approach is innovative in that it does not only entail quantitative, bottom-up analyses and detailed interpretations, but it also introduces statistical techniques into linguistic interpretation. His work, however, also raises a number of questions: the linguistic features Biber uses to carry out the general analysis of English registers are taken from the 1988 study concerned with the spoken-written continuum. This results in a bias of the whole study towards this continuum, yet many of the features under investigation are not restricted to the spoken-written continuum or, to put it in systemic terms, to medium of discourse. Moreover – a more serious reservation – certain important features serving as indicators of other aspects of register variation are excluded. From the point of view of statistical techniques a question arises with respect to Biber’s – very convincing – findings, namely which level of measurement applies to the data (see section 2.2 of chapter 5). The level of measurement determines the statistical technique apt for the purpose of investigating the respective data (cf. Butler 1985: 12). As discussed in Baroni and Evert (2009; see also section 2 of chapter 5), frequency counts may not reach the measurement level required for parametric statistical techniques like factor analysis (as well as the tests used in this study). Biber (1988, 1995) does not mention whether he computes proportions based on the frequency counts that would allow advanced tests. Therefore, factor analysis might not be suitable for Biber’s data. Vermunt and Magidson (2005) remark that factor analysis is frequently applied to variables that do not achieve the interval level and may consequently bias the parameter estimates. Butler (1985: 12–13) points out that linguists commonly assume a higher level of measurement because “parametric tests suitable for interval data are more powerful than non-parametric tests. As often happens in the application of statistical methods to real problems, practical considerations frequently outweigh the concerns of the theoretical purist.” Biber’s inductive approach poses more problems in the contrastive perspective – not to mention the general issue of using features deduced in other theory-

Translation: influences from language mediation 

 27

based studies, which again highlights the fact that there is no such thing as an exclusively inductive study. Biber has to introduce communicative functions which are in turn inferred inductively from the findings supplied by his contrastive comparisons. As a result of the functions being derived from these findings, their scope is limited to generalisations that merely apply to the features originally investigated. A combined deductive and inductive methodology – particularly one that is functional12 – would ensure a systematic and comprehensive contrastive comparison, while still taking into account relevant insights from the study of empirical evidence. A theoretical framework well-suited to complement and motivate bottom-up analyses is the systemic functional one, certain aspects of which are related to Biber’s work, for instance, the functional view of language and the role of contextual information. Teich’s (2003) analysis of one register exemplifies how the systemic functional framework can be employed in studies on linguistic variation (see section 3.2 of this chapter). As will be shown in this study, the concept of register in systemic functional terms requires some adaptation to make it suitable for quantitative studies, for which Biber’s approach sets a valuable example.

4 T  ranslation: influences from language mediation Translations represent an interesting special case of linguistic variation: Comparable texts in two languages belonging to contrastive registers reflect the language-specific peculiarities of a given context of situation and vary accordingly. Translations combine and consequently vary reflecting the influences of both contexts. One way of approaching this kind of variation is by contrasting the register of both originals and translations. The next section will take stock of register analysis in translation studies. A strand of research that is particularly concerned with variation in translations is the study of translation properties. It will be reviewed in the following section. The present discussion of the state of the art in translation studies concentrates on research concerned with variation in the widest sense of the word. However, it does not cover the full range of research in translation studies, the bulk of which currently focuses on the process of translation (cf. Alves 2003, Hansen 2006, Jakobsen 1999). With some notable exceptions (Catford 1965, Matthiessen 2001, Steiner 2004a,b, see also Matthiessen 2007 etc.), translation research is often conducted independently 12 See Neumann (2003) for a discussion of the advantages of founding contrastive comparisons on the systemic functional approach.

28 

 State of the art

of a general theory of language capable of providing theoretical concepts that would enable the derivation of hypotheses and substantiate the interpretation of findings (see also section 1 of this chapter). The scope of the research is, therefore, limited by the absence of such a theoretical foundation.

4.1 Register analysis in translation studies Text analysis of the source text (e.g. Nord 1995, Newmark 1988 etc.) is an area of translation studies that is often advocated with a view to application without a firm basis in a more general theoretical framework (see Steiner’s 2004b critique of these approaches). Register analysis, introduced to translation studies mainly to serve as a tool for text analysis or to solve text type-specific translation problems (see the contributions in Trosborg 1997), has helped overcome this weakness. It is often treated rather cursorily with only the three register variables being mentioned. This should not, however, be held against the authors since register analysis is typically recognised as a methodological necessity, but does not represent a central aspect of their research. Moreover, the more detailed framework was not readily available for earlier authors, but had to be collected from various individual sources. Therefore, it comes as no surprise that register analysis has received scant treatment; Baker (1992: 15–17), for instance, dedicates less than two pages to lexical meaning and discusses register analysis under this heading. This is somewhat surprising since one would assume that one major advantage of register analysis over other methods of text analysis is precisely that it is not restricted to lexical, or more generally, to experiential meaning. Hatim and Mason (1990: 36–54 and 1997: 20–24) provide a more detailed discussion and incorporate a small-scale register analysis in their model of text analysis for translation. Finally, Teich (2003) employs register as the concept to identify text exemplars for her corpus and to establish comparability between the cross-linguistic exemplars. Register linguistics figures more prominently in House’s (1997) translation quality assessment, which uses a modified register analysis including a range of subdimensions serving as a grid. The analysis proceeds interpretatively characterising the texts in a hermeneutic way and sometimes citing examples to corroborate interpretation. House does not count the frequencies of the features taken into consideration and thus does not provide an empirical basis to support her assumptions about the texts under investigation. Steiner elaborates the register framework and looks beyond its application in translations by taking into account a wide range of subdimensions as well as

Translation: influences from language mediation 

 29

concrete linguistic indicators serving as operationalisations (see section 2 of this chapter). His numerous contributions (1997, 1998a,b, 2000, 2001a, 2004a,b) demonstrate how register analysis may be used to evaluate translations, to delineate the limits of translation etc. While he stresses the importance of larger scale quantitative investigations, his studies are mainly limited to the analysis of individual texts (cf. however recent work in Steiner 2008a,c). Neumann (2003) investigates, in particular, the importance of register for translations in an explicitly empirical setting. However, the features used in this study establish a direct link between the three abstract register variables and lexico-grammatical features (see chapter 4). Moreover, register was suggested as a potential source of properties specific to translated text (Steiner 2001b). The study of specific properties is the strand of research in translation studies that is most clearly empirical in a quantitative sense. It will be introduced in the following section.

4.2 Translation properties The investigation of translation properties is based on the established perception of translated texts as being longer, simpler, more explicit, impoverished, standardised etc. (Vinay and Darbelnet 1958, Berman 2004, Blum-Kulka 1986, Toury 1995 etc.). These assumptions have led to the notion that translations represent a “third language” (Duff 1981: 116) or a “third code” (Frawley 1984: 168) or, in a less general way, that translations carry specific properties that may even be universals (Baker 1996, see also Mauranen and Kujamäki 2004, Bernardini and Zanettin 2004). It was the combination with Corpus Linguistics as proposed by Mona Baker in 1993 that brought about the necessary methodology to validate these characteristics in a quantitative way that were, until then, only discussed impressionistically. Baker subsequently developed her approach to the investigation of translation properties, concentrating on translations into English from as many different source languages as possible13 and comparing these to English originals (Baker 1995). This allows the identification of properties irrespective of the specific language pair, but rules out the possibility of tracing back conspicuous cases to the source text; this would, however, help to verify whether they really represent peculiarities of translations or whether they mirror features present in the source text. Baker identified four potential universals that should be observ13 to be investigated in the Translational English Corpus (TEC): http://www.llc.manchester.ac.uk/ctis/research/english-corpus/, last visited 12 April 2013.

30 

 State of the art

able in this corpus design, namely simplification, explicitation, normalisation and levelling-out (Baker 1996). Due to her concentration on the target language, a major candidate for translation universals, interference of the source language (Toury 1995, Teich 2003, Mauranen 2004, 2005, as well as the body of work on translations as a gateway to language contact, e.g. House 2002, Steiner 2008a etc.), has to be left out in this research design. Interference is even explicitly excluded, because it does not constitute a specificity of translation but a source language influence. Baker (1993: 243) defines “universal features of translations” as typically occurring in translated rather than original utterances “and which are not the result of interference from specific linguistic systems”. Baker’s 1993 paper instigated a whole range of individual studies into the four types of translation properties in various language pairs (Laviosa-Braithwaite 1996, Kenny 1998, Øverås 1998, Olohan and Baker 2000, Olohan 2001, Olohan 2004, Pápai 2004, Puurtinen 2004, Eskola 2004 etc.). These studies mainly use raw corpora, i.e. data not enriched with linguistic information, and investigate low-level features like lexical density, frequencies of individual or collocated lexical items, average sentence lengths etc. to detect manifestations of simplification, sanitisation and explicitation etc. A range of findings on specific characteristics of translations result from corpus projects combining the perspective of contrastive linguistics and translation studies (e.g. Johansson 2007, Fabricius-Hansen 1998, 1999, Hasselgård 1998, Aijmer 1998 and other studies in Johansson and Oksefjell 1998, Ebeling 1998). These take into account the corresponding source texts and also look at larger units like the sentence. The explanatory power of the approach to the study of translation properties using raw, monolingually comparable corpora is exemplified by Olohan and Baker (2000). The paper neatly shows how a given optional feature, the complementiser that, is used more frequently in texts translated into English than in texts originally written in English. The authors analyse concordances for the various forms of to say and to tell followed by a complement clause in the Translational English Corpus (TEC) and in comparable parts of the British National Corpus (BNC)14 and count the frequency of the complementiser introducing the complement clause. They find that the complementiser is indeed significantly more frequent in translated than in non-translated English. This finding is interpreted as an indication of explicitation in the translations. Although Olohan and Baker’s (2000) findings are compelling and illuminate this important area of translation studies, two shortcomings become apparent: Firstly, this study highlights the restrictions of the corpus-driven approach. Without additional linguistic information, the authors are limited to the cumbersome query of individual forms of 14 http://www.natcorp.ox.ac.uk/, last visited 12 April 2013.

Translation: influences from language mediation 

 31

just two verbs, although the phenomenon they investigate extends beyond to say and to tell. The benefits offered by enriched corpora to the study of translation properties has been exemplified by Hansen (2003), Teich (2003; see section 3.2 of this chapter) and more recently by Hansen-Schirra, Neumann and Steiner (2007; see also Hansen-Schirra and Teich 2009). Hansen’s (2003) work is particularly interesting in this respect as she enriches the Translational English Corpus with part-of-speech information and interprets the linguistic information in view of Biber’s dimensions. This allows her to obtain a broad and detailed linguistic overview of the specific properties of translations in terms of normalisation as well as what she calls anti-normalisation, that is a characteristic overuse of a given register’s untypical feature of Biber’s dimensions. She even goes one step further to interpret her findings with regard to the translation process. Whereas Olohan and Baker (2000) assume that their corpus findings, i.e. their insights on the product of translation, shed light on subconscious processes responsible for explicitation in the translation process, Hansen (2003) includes a small psycholinguistic case study in order to test whether the translators’ processing effort suggests a confirmation of the corpus findings. Secondly, in addition to the question of the informative potential of raw data, the evidence Olohan and Baker (2000) identify in the comparable corpora can only be seen as a preliminary clue pointing to explicitation. As Steiner (2005) and subsequently Hansen-Schirra, Neumann and Steiner (2007) point out, explicitation is a relation or process between two texts in a translation relationship. Consequently, it can only be identified in aligned text pairs (and within the text pairs in aligned translation units). Otherwise, there is no possibility of verifying whether the potential instance of explicitation in the target text may not simply be a direct translation of a similarly explicit structure in the source text. A related strand of research also investigates translation properties with the goal of identifying a particular influence of the English language as a lingua franca on other languages (e.g. House 2002, Baumgarten et al. 2004). A less general assumption about translations refers to the description of typical shifts between source and target text units both on a grammatical and semantic level as discussed by Vinay and Darbelnet (1958), Catford (1965), Newmark (1988), Leuven-Zwart (1989), Munday (1998) and most recently Cyrus (2006). These translation shifts as a corollary of contrastive language differences also pose a problem for Machine Translation and were discussed particularly in the 1990s by Barnett et al. (1991), Kinoshita et al. (1992) and Dorr (1994) etc. (see also Čulo et al. 2008 for an overview of both the translation studies and the MT approach). In the research on translation properties, registerial restriction has, so far, been included mainly as a measure to control possibly confounding variables. It would be interesting to see whether the importance of characteristic traits of

32 

 State of the art

translated texts varies from register to register. This would also shed more light on the role of register variation as a source of divergences found in translations (Steiner 2001b). Furthermore, the study of translation properties should not remain an end in itself. Rather, this line of research should be seen as a means of working towards an empirical theory of translation (cf. Toury 1995).

5 Envoi The discussion in the previous sections has identified some shortcomings of works analysing texts from a contrastive perspective (including the comparison of translations and originals). It was shown that approaches to the analysis of text are often restricted to a limited number of textual features and to the study of individual texts. Studies that aim at the classification of groups of texts are still often based on categories derived from the study of individual texts. A notable exception is Biber’s (1988, 1989, 1995) work, which is, however, limited in terms of possible generalisations due to its strictly inductive approach. This limitation is overcome in Teich’s (2003) study by situating it in a theoretical framework. Variation in her study is, nevertheless, restricted to that between languages and originals and translations. Variation in terms of registers is beyond the scope of her work. Contrastive linguistics does not contribute to this field either. In translation research, variation between originals and translations is an ongoing research topic which still requires more attention. Particularly, the area of variation resulting from contrastive register differences has not yet been fully exploited. To overcome these challenges, the design of the research should be tailored to meet the following requirements: –– It should be quantitative in nature to allow general statements based on larger quantities of data instead of individual texts. –– It should apply a comprehensive set of criteria that covers all aspects relevant to the characterisation of a register. –– The criteria should be operationalised in a way that is methodologically valid and comprehensible. –– The criteria should be comparable across languages to allow contrastive comparisons. The present study addresses these requirements with the help of a corpus-based research design that includes linguistic annotation of a range of registers in the two languages English and German in order to test three general hypotheses

Envoi 

 33

concerned with variation in the intralingual, the contrastive as well as the translational perspective (see chapter 5). It intends to adapt the existing work on register to a comprehensive quantitative analysis that narrows the gap between abstract concepts and observable features and that is comparable across languages (see chapter 4). It has to be kept in mind, however, that this can only be one small contribution towards a more empirical exploration of language. Before presenting these operationalisations for the present study, we will address some general aspects of the empirical methods used in linguistics in the following chapter.

Chapter 3 On empirical methods in linguistics As a quantitative study, this work makes use of empirical methods. Although having gained considerable ground in linguistics in the last decades, particularly in the field of language acquisition, psycholinguistics and corpus linguistics, the use of empirical methods is still not firmly rooted in linguistics. Traditional hermeneutic, rationalist and also more recent developments in critical theory continue to play an important role in the field. The aim of the present chapter is therefore to explain the background of empirical research and address some issues arising from the use of empirical methods in linguistics as well as potential problems concerning cross-linguistic empirical research. Before moving to a more detailed consideration of these issues, we will first briefly discuss some aspects of the interrelation between linguistics and the social sciences.

1 C  onceptual and methodological links between linguistics and social sciences Language as the object of research can be viewed from different perspectives. It can be analysed either with respect to the relation between thinking, comprehending and speaking from a psychological angle or from a clinical perspective, which focuses on pathological aspects of language use. It can also be interpreted as a social phenomenon, highlighting the way social structure impacts the way we speak. Or, as Hasan (1973: 253) claims: Language is primarily a social phenomenon; despite the fact that some of man’s biological attributes play a decisive part in the acquisition and subsequent use of language, it remains the property of social man rather than that of animal man. This gives rise to the possible hypothesis that aspects of social structure would be reflected in language. (…) it is rather that the structure of language is related at various levels of abstraction to social structure.

In this perspective, language lies “at the heart of making and shaping social relations themselves” (Montgomery 2008: 7). Referring to the interaction between language and society, Halliday (2007: 251) writes:

Conceptual and methodological links between linguistics and social sciences 

 35

(…) The relation of language to the social system is not simply one of expression, but a more complex natural dialectic in which language actively symbolizes the social system, thus creating as well as being created by it.

Language can then be viewed as a “social semiotic” (Halliday 1978). There is thus a conceptual link between linguistics in general and more specifically the systemic functional theory of language as chiefly developed by Michael Halliday (1961, 1963, 1966, 1967, 1968, 1973, 1976, 1978, 1985) and social science. In explaining the way registers reflect social order, Halliday (2007: 254) elaborates that “the total range of the social processes in which any member will typically engage is a function of the structure of society”. Hence, the structure of society is said to determine who will have access to what kind of social processes and consequently to which registers. This view is closely related to Bernstein’s sociology of the relationship between social class and the code used by members of the respective class (e.g. Bernstein 1971, 1973), a view embraced by systemic linguists (see Hasan 1973, Hasan 2005, chapters 7 and 8 of Halliday 2007, see also Brown 2006 and Montgomery 2008: 168ff. for a critical appraisal of Bernstein’s approach). A sociological approach to language was also taken up by the subdiscipline that came to be known as sociolinguistics (e.g. Labov 1966, Trudgill 1974).15 One concrete consequence of this perspective on language is the inclusion of social factors in the analysis of texts. In the systemic functional framework, this is typically done within the context of register analysis, which treats any given situation as a social context. A social context is a semiotic structure which we may interpret in terms of three variables: a ‘field’ of social process (what is going on), a ‘tenor’ of social relationships (who are taking part) and a ‘mode’ of symbolic interaction (how are the meanings exchanged). (Halliday 1978: 189)

Particularly in the interpersonal area of register theory, social relationships are described as contributing to the way in which participants act in a given situation (cf. for instance Hasan 1973, Poynton 1985). We have already established that sociologists, too, are interested in language. Bernstein’s work remains a major contribution to the understanding of language in a social perspective, but in more recent times Pierre Bourdieu has 15 See also Mair’s comment on the interplay between sociolinguistics and corpus linguistics in the study of language change: He explains that whereas sociolinguistics is typically concerned with phonetic changes in non-standard varieties, corpus-linguistic methods are well suited for the examination of lexical and grammatical change in written, standard language (Mair 2006b: 29).

36 

 On empirical methods in linguistics

also taken language into account in his research on the stratification of society (e.g. Bourdieu 1990). His work, in turn, has been received by linguists (in particular Hasan 2005). Systemic functional linguistics is also methodologically linked to the social sciences. One of its essential principles is the analysis of language in use, with the system assumed to exist only by way of the instances produced in concrete interactions (instead of existing somehow separately).16 The workings of the language system can then only be observed in these concrete interactions by means of empirical studies in both their qualitative and quantitative shape (see section 2.1 of this chapter). The social sciences, too, make use of the analysis of texts to answer sociological research questions, particularly those concerned with communication phenomena. The question formulated by the early political scientist Harold Lasswell “Who says what, to whom, why, to what extent and with what effect?” is still cited today in social research concerned with texts (e.g. Diekmann 1995: 485).17 The main method used for this purpose is content analysis, which works on the linguistic levels of syntax, semantics and pragmatics. Quantitative content analysis explores, for instance, the frequency of certain lexical items in order to gain insight into the topics present in a given population of texts. To this end, concordance tools developed specifically for the social sciences are used, which offer the same functionalities as similar tools in linguistics, e.g. WordSmith tools (Scott 2004), like keyword-in-context.18 Having established the link between linguistics – systemic functional linguistics in particular – and the social sciences, let us now turn our attention to the empirical methods, the set of methods preferred and elaborated in social research.

16 “The system is the underlying potential of a language: its potential as a meaning-making resource. This does not mean that it exists as an independent phenomenon: there are not two separate objects, language as system and language as a set of texts.” (Halliday and ­Matthiessen 2004: 26) 17 Interestingly, Lasswell’s formula is also used in linguistics and more notoriously in translation studies as the basis of source text analysis (Nord 1995: 40). 18 For instance TEXTPACK: http://www.gesis.org/unser-angebot/daten-analysieren/software/ textpack/, last visited 12 April 2013.

Empirical methods 

 37

2 Empirical methods In its general meaning ‘empirical’ denotes “based on observation or experience” (as opposed to “created by reasoning”). Lauth and Sareiter (2005: 13–18) distinguish four different contexts of the term ‘empirical’: empirical phenomena, i.e. observable and measurable events or processes that can be located in space and time, empirical theories for describing, explaining and predicting empirical phenomena, empirical methods (see below for a more detailed discussion) and empirical knowledge as the philosophical discussion of questions like the reliability and scope of empirical methods, or whether empirical methods are at all capable of providing reliable knowledge. In what follows, we will focus on empirical methods and point out some philosophical problems with empirical studies. Chafe (1992: 81) takes a very general view of the nature of scientific understanding, describing it as “a more focused approach to what everyone does naturally”. According to Chafe, it consists of two components: (A) The systematic, non-casual, probing observation of small portions of reality, often called data. (B) The imaginative creation of descriptions of parts of the universe that extend beyond the observed phenomena themselves, and that include casual explanations for those phenomena. We often call such descriptions theories. (Chafe 1992: 81–82)

‘Empirical method’ refers to the research method which investigates actual data. In this sense, “empirical indicates that the information, knowledge and understanding are gathered through experience and direct data collection” (Black 1999: 3). One of the main characteristics of the empirical method is that it allows systematic observations with the goal of producing replicable studies (Black 1999: 4). Empirical methods have been developed to different levels of sophistication for application in various scientific disciplines, particularly the natural sciences. The methodology adopted by psychology and social research is probably closest to the needs of linguistic research, not least because both disciplines are interested in language data, albeit to different degrees (see section 1 of this chapter). There is no single empirical method; rather many different methods exist, all of which, however, serve to investigate actual as opposed to invented or otherwise artificial data. First, we can distinguish exploratory from explanatory studies, the former being concerned with exploring a new field with the goal of generating new hypotheses, the latter starting from hypotheses and aiming at testing them. Exploratory studies may involve observation, open interviews, case studies etc., while explanatory studies often use quantitative methods.

38 

 On empirical methods in linguistics

These can be experiments, observations, interviews or surveys. In linguistics, experiments, observations and interviews are often used in branches such as psycholinguistics (see Garrod 2006), second language acquisition (see Albert and Koster 2002), and process-oriented translation studies (e.g. Krings 1986, Alves 2003, Hansen 2006). Linguistic surveys are typically conducted in the form of corpus studies. Biber et al. (1998: 4) list the following “essential characteristics of corpus-based analysis”: –– it is empirical, analyzing the actual patterns of use in natural texts; –– it utilizes a large and principled collection of natural texts, known as a “corpus,” as the basis for analysis; –– it makes extensive use of computers for analysis, using both automatic and interactive techniques; –– it depends on both quantitative and qualitative analytical techniques.

Corpus analyses are often used in areas such as lexicology and lexicography (see Sinclair’s work, e.g. Sinclair 2003 as well as his contribution in Teubert 2007), grammar writing (e.g. Biber et al. 1999), language variation (e.g. Biber 1988, 1995), language change (e.g. Mair and Hundt 1995, Mair 2006a, 2007), historical linguistics (Facchinetti and Rissanen 2006), corpus-based translation studies (e.g. Olohan 2004, Hansen 2003, Teich 2003, Neumann 2003) and natural language processing (e.g. Koehn 2005). Halliday (2005: 173) refers to corpus linguistics as an empirical approach to the description of language where the accumulation of new data and their interpretation leads to new theories. He states: after all, that’s what it did in physics, where more data and better measuring transformed the whole conception of knowledge and understanding. How much the more might we expect this to be the case in linguistics, since knowing and understanding are themselves processes of meaning.

As the above overview shows, empirical research is an occupation of linguists of all shades. There are, however, some philosophical issues associated with empirical research which should be kept in mind when evaluating the explanatory power of empirical findings. Referring to Thomas Kuhn’s work on “scientific revolutions”, Okasha (2002: 88–89) explains the “theory-ladenness of data”: As a matter of perception, different people look at data from different theoretical perspectives and thus perceive the data to be different. While this statement appears to be confirmed by day-to-day experience in scientific discourse over linguistic findings, Okasha qualifies it by pointing out that this does not rule out objectivity altogether since scientists from different paradigms may accept certain statements that are “sufficiently free of theoretical contamination”

Empirical methods 

 39

(Okasha 2002: 89). Despite their different opinions on whether there is such a thing as objective findings, most philosophers of science will accept the existence of an objective truth. Efforts aimed at ensuring objectivity in the social sciences will be addressed in section 3 of this chapter. Lauth and Sareiter (2005: 170) address another, more concrete problem of empirical research. Scientific hypotheses frequently cannot be tested by directly observation but by measurement. In their view, observation supplies direct answers to a question (e.g. observing rain as an answer to the question whether it is raining), whereas measurement is used to solve questions that cannot be answered directly by observation. Measuring, however, is inherently error-prone. Apart from systematic errors of the measuring instrument, deviating measurements might be due to chance. Lauth and Sareiter (2005) go on to explain that statistical techniques, for instance measures of central tendency (see chapter 5) help limit the effects of such deviations by compensating for the variance between the individual measurements. The objections against the empirical methods highlight some limitations that become particularly relevant when reasoning on the basis of empirical findings. These reservations do not, however, call into question the applicability and validity of empirical methods in general. In the following section, we will discuss qualitative and quantitative approaches to the analysis of language. 2.1 Qualitative and quantitative approaches Introductions to corpus linguistics typically start by discussing the difference between empiricist and rationalist approaches to the study of language.19 This has been widely discussed (see also the contributions in Svartvik 1992) and is not the major concern of the present study. No standard methodology having emerged for the analysis of actual data, this area of research still uses a wide spectrum of methods. This can be viewed as a range defined by qualitative and quantitative methods at opposite ends of the scale. What is thus of particular interest here is the relationship between analysis based on the calculation of frequencies of occurrence and those based on human interpretation. As McEnery and Wilson (2001: 76) point out, quantitative research is concerned with the 19 See for instance chapter 1 of McEnery and Wilson (2001), chapter 1 of Meyer (2002) and chapter 2 of Lemnitzer and Zinsmeister (2006). Featherston (2008) exemplifies a systematic approach to an intuition-based investigation of language that overcomes the typical critique by empirical linguists of its introspective and consequently non-systematic character. It does not, however, overcome the non-naturalness of isolated and possibly artificial sentences (see Chafe 1992: 86).

40 

 On empirical methods in linguistics

classification and counting of features and the construction of statistical models intended to explain findings obtained from observation of the data. Qualitative research, in their view, does not use frequency counts to corroborate their linguistic findings. Data are used to identify and describe aspects of language use and to provide examples. Going beyond this descriptive view on qualitative studies, their aim is to interpret instead of count, often with a different notion of the data: rather than taking the linguistic data at face value, qualitative researchers attempt to elicit some deeper, hidden meaning that underlies the surface form. In an introduction to the philosophy of social science, Rosenberg (2008: 27) explains how the natural sciences have continually increased the explanatory scope and predictive power and thus succeeded in “eliminating meaning, purpose, or significance from nature”. The record of the history of science requires every social scientist to face the question, Why should human behavior be an exception to this alleged pattern? Why should meaning, purpose, goal, and intention, which have no role elsewhere in science, have the central place they occupy in social science? The obvious answer is that people, unlike (most) animals, vegetables, and minerals, have minds, beliefs, desires, intentions, goals, and purposes. (Rosenberg 2008: 27)

The question is whether and to what extent language can or should be viewed as human behaviour, consequently also reflecting minds, beliefs, intentions etc., or as having a separate and distinct existence that can be observed independently with the help of empirical methods. The quantitative researcher will claim that meaning is not excluded from his/her studies.20 It is sometimes addressed indirectly via the formulation of hypotheses (see below). As mentioned above, it is more a matter of tendency than of disjunction whether a study is labelled qualitative or quantitative; often qualitative studies will also involve the counting of features in the exemplary text. Let us now examine the advantages and limitations of both methodological approaches in more detail. Qualitative research in linguistics typically means analysing exemplary texts. Ideally, it provides rich and detailed interpretations of these exemplary data relating both to rare phenomena and to frequent ones. Since it typically aims at a comprehensive stocktaking of all features, the distinction between rare and frequent phenomena might not be made at all. McEnery and Wilson (2001: 76) explain that “the ambiguity which is inherent in human language – not only 20 See also a thread on the CORPORA mailing list where proponents of the corpus-driven approach to linguistics take credit for eliciting meaning from the corpus: http://www.uib.no/ mailman/public/corpora/2008-August/007077.html, last visited 12 April 2013.

Empirical methods 

 41

by accident but also through the deliberate intent of language users – can be fully recognised”. In other words, abstract concepts may not necessarily be operationalised in a specified way since the analyst can use his/her intellect to interpret and assign a feature of a text to the abstract concept under investigation. The scholar using a qualitative method may not set herself/himself a hard and fast threshold as to what is interpreted as a significant finding, but rather judge on the basis of his/her knowledge and experience whether a finding requires specific attention or whether it is negligible. Whereas the use of heavily interpretative categories can be seen as an advantage in that it enables a high degree of information to be harvested from the qualitative data, these categories can also be vague and their interpretation can vary with different analysts. Furthermore, ‘interpretative’ may also signify ‘subjective’, which can make it difficult for others to follow the results thus obtained. This, of course, can and should be avoided by ensuring the transparency of the categories used and their interpretation (see section 3 of this chapter). More importantly, the impact of the qualitative findings beyond the examples described may be overrated. In order to allow general statements beyond individual instances, the data have to fulfil certain criteria in terms of quantity and balance. The findings of an example-based study “cannot be tested to discover whether they are statistically significant or more likely to be due to chance” (McEnery and Wilson 2001: 76). Qualitative research remains an important enterprise in order to explore the range of the language system and can as such be used as a basis for quantitative research. In terms of the classification of empirical methods touched on above, qualitative studies belong to the area of research concerned with the generation of hypotheses that should subsequently be tested with quantitative methods. Quantitative studies, by contrast, analyse larger amounts of texts allowing generalisations that apply to a larger population, extending beyond the individual texts contained in the corpus. Whereas qualitative studies can only discuss features that are present in the exemplary text (or absent, by chance), quantitative studies allow a distinction to be made between phenomena “likely to be genuine reflections of the behaviour of a language or variety” (McEnery and Wilson 2001: 76) and those which occur merely by chance, thus discerning between systematic features and random occurrences. This distinction requires the use of statistical techniques which are, however, typically only available for a certain minimum quantity of different cases. In comparison to invented examples and other types of manipulated linguistic data, corpora represent actual, naturally occurring language. They possess a high ecological validity. This, however, means that the data cannot be controlled to a similar degree as would be possible in an experimental environment.

42 

 On empirical methods in linguistics

Confounding factors are an issue inherent to corpus analysis and are difficult to eliminate. Furthermore, as Chafe (1992: 88) points out, corpora do not allow particular phenomena to be targeted without waiting for them to occur. He adds that this non-occurrence may in itself be an interesting finding. Corpus linguists typically try to circumvent this problem by enlarging the corpus (see section 1.1 of chapter 5). Nevertheless, this does not change the role the rarely occurring feature plays in the language system. In what follows, the discussion will be restricted to corpus-based quantitative studies (rather than experiments etc.). Rather than exploring the linguistic richness of an example text in all detail, corpus-based studies must focus on certain features and are thus typically less rich and detailed. Quantitative research that includes linguistic enrichment rather than working with raw data depends on automatic annotation since the amount of text involved (the corpus used in this study counts more than 1m words, see part 2) cannot be processed manually, particularly if the annotation is to comprise several layers. The annotation can therefore only be as accurate as the tools used (tool-related errors are systematic errors as described above and have to be taken into account in terms of reliability, see section 3 of this chapter). The more linguistic information is included in the automatic annotation, the less accurate the tool will be. There may be applications in language technology where a comparatively low level of accuracy may be acceptable (see below). This is, however, clearly not the case in linguistic analysis. It may therefore be advisable to employ less interpretative tools providing highly reliable results or even computer-assisted manual annotation, which may be more efficient than the manual correction of automatic annotation. Manual annotation is subject to the same limitations as interpretation in qualitative research with regard to subjectivity, inconsistency etc. This latter aspect can be kept under control to some extent by subjecting the corpus to consistency tests. As to the informativity of the data, quantitative studies may have to disambiguate fuzzy sets and therefore may, under certain conditions, entail what McEnery and Wilson (2001: 77) call “a certain idealisation of the data”. The present study takes a quantitative approach to the study of language. This seems particularly important in view of two features: the investigation of register variation as well as the cross-linguistic aspect of the study. As to register variation, Halliday (2005: 73) observes the central place of corpus studies in theoretical investigations of language, for instance, in the area of register variation, which he describes as “variation in the probabilities of the grammar”. The basic assumptions concerning the role of register variation in language in use presuppose an empirical and, more specifically, a corpus-based approach to the research. He stresses the important role of large-scale quantitative corpus studies in the exploration of the probabilistic character of language and claims

Empirical methods 

 43

that a register-controlled corpus enables these probabilities to be investigated and that corpora are a valuable resource for identifying quantifiable grammatical differences between registers (Halliday 2005: 70–71). This suggests a strictly quantitative approach to the analysis, particularly since any example-based approach to the study of register variation will inevitably be based on a certain amount of speculation. It involves “classifying features”, one of the tasks described by McEnery and Wilson (2001) in connection with quantitative studies. This is intrinsically a matter of interpretation, thus highlighting the division of labour between qualitative and quantitative methods: Any reasonable quantitative analysis will be preceded by a qualitative examination of the data and followed by a qualitative interpretation of the results (Gries 2008: 11). With respect to the cross-linguistic aspect of this study, preferences in usage are of particular interest to contrastive linguistics and foreign language teaching, where the fact that something can potentially be expressed is less important than whether it is actually expressed using a given structure. It is only thanks to quantitative studies that we can reveal these preferences rather than focus on the general availability of different options in the language system. The analyst using a qualitative method may not necessarily give full, or rather separate, account of the relationship between the abstract concepts s/he is interested in and the features in the text cited in support of the interpretation, because the relationship is incorporated in the interpretation. By contrast, quantitative analyses distinguish between the operationalisation of indicators and their interpretation in view of the abstract research question. Therefore, the question of how the abstract concepts under investigation are related to observable features in the data becomes more relevant.

2.2 From abstract concepts to observable features As Black (1999: 6) puts it: “The pursuit of truth is desirable, but often this constitutes trying to develop a model of reality, an explanation of events employing abstract and intangible concepts.” This means that for the most part we cannot directly observe the things we are interested in. Consequently, we are working with hypothetical links between our abstract concepts and observable parts of reality, e.g. language, most of the time. The process of deriving observable indicators from abstract concepts is called operationalisation. Taking an example from the present study, it is not possible to observe field of discourse directly in a text representing a given register. This is also true for goal orientation as one specification (i.e. subdimension) of field, or for exposition as one possible goal

44 

 On empirical methods in linguistics

(for the description of these categories see chapter 4). What can, however, be observed in texts is the frequency of nouns, nominalisations etc. These features, or indicators, are operationalisations of exposition under the assumption that expository texts pursue the goal of conveying factual information and that this goal is realised by a nominal style characterised by a high frequency of nominal elements. Only these elements are actually observable in texts.21 It is a major task of the quantitative researcher to work out the relationship between the abstract concepts and the features observed in the corpus in order to ensure the validity (see section 3 of this chapter) of the study. Systemic functional theory typically describes the relation between the different levels or strata of language as a realisation relationship. Each stratum is said to realise the respective higher-order stratum and to be realised by the lower-order stratum. Lexico-grammar, for instance, realises semantics and is realised by phonology/graphology. This perspective provides an ideal framework for qualitative research concerned with individual instances of the language system. Its application to quantitative approaches is, however, less straightforward. A more suitable approach appears to be one that links abstract concepts to observable indicators by hypotheses in the way described above. Although this distances the quantitative approach from one of the core representational categories of the theoretical framework, this view of a hypothetical linkage between abstract concepts and observable indicators allows a more flexible adoption of indicators necessary in quantitative research. Furthermore, it helps to avoid a direct equation of lexico-grammatical systems with the contextual parameters field, tenor and mode of discourse, which limits the scope and explanatory power of these parameters.

21 It has to be kept in mind, though, that categories like ‘noun’ and ‘nominalisation’ are theoretical concepts again. The linguist will assign these categories to certain units in a text, but they are not “natural” features of linguistic elements. This example illustrates the theoryladenness or the degree to which the analyst works on hypotheses like “frequent nominal elements are a symptom of an expository goal” and “linguistic units with given grammatical characteristics are nouns”. All of these limit the empirical knowledge to be gained from the study of language in use, since it means that we do not simply observe and describe “brute data” (Bishop 2007: 21), i.e. data that exist without any interpretation (see above).

Quality criteria for empirical research  

 45

3 Quality criteria for empirical research In social research, typically two quality criteria are discussed: validity and reliability (e.g. Black 1999: 272–302). Sometimes a third concept, objectivity, is included, although this is controversial due to the philosophical problems it is afflicted with (see section 2 of this chapter). Notwithstanding this objection, it will be mentioned here as well as it seems to be an important concept even if it may not be fully attainable. Reliability is concerned with the exactitude of the measuring instrument or method. If these produce accurate results, repetitions of the study under the same conditions should yield the same results. In manual analyses, this means that repeated analyses should interpret a given feature in the same way, and in the case of repeated automatic analyses of the same data, the computer tool should always produce the same results, too. Apart from systematic errors due to the limits of accuracy of automatic tools (which may indeed make a tool useless for linguistic rather than computational linguistic research, see section 2.2 of chapter 5), reliability can be one advantage of using tools, assuming that they do not change their interpretation of a given element and that – unlike the human analyst – they do not get tired. Validity has two equally important aspects: Is the choice of method appropriate to the phenomenon under investigation? Do the chosen indicators actually measure the concept under investigation (and not a confounding factor)? This is of particular importance in quantitative studies relying on hypothetical relations between the abstract concepts of interest and the linguistic indicators used to obtain information on the concepts (see above). Objectivity, by contrast, is concerned with whether the study produces the same results irrespective of the person by whom the analysis is carried out. This applies to both execution of the research and evaluation of results. The more a given study relies on human interpretation, the more important this concept becomes. Typically, in qualitative studies the concept of objectivity is replaced by intersubjective verifiability ensured by transparent documentation of the research process, the use of codified procedures (in linguistic analysis this is achieved, for instance, by adhering to clear annotation guidelines) and transparent data interpretation. These concepts have been dismissed by qualitative social research as inadequate to their research purposes (see for instance Mayring 2000: 109–111). This criticism is particularly directed at some of the techniques used in quantitative research for testing reliability and validity. Although it may not be entirely unjustified, it does not make the quality criteria useless. They should at least be

46 

 On empirical methods in linguistics

seen as guiding principles to be taken into account in designing and conducting empirical research. Apart from the criteria used in the social sciences, another evaluation method is of relevance to the automatic processing of corpora in general. A measure used to evaluate the success of natural language processing (NLP) systems, especially in information retrieval is ‘precision and recall’ and related measures (Manning and Schütze 1999: 267–271). Precision refers to the proportion of selected items retrieved correctly by the NLP system and is reduced by wrongly selected items. Recall identifies the proportion of retrieved items (correct or incorrect) in relation to the overall amount of correct elements that should be selected by the system. These two measures are often combined into a single measure of overall performance, the F score. Statistical NLP models, which typically process very large quantities of data,22 are assessed against these measures and are regarded as high quality systems with scores that may, in some cases, appear relatively low to the untrained observer. Studies intended to offer linguistic insight, however, may require a very high score. Here, the precision of automatic annotation and query tools is of crucial importance, and a trade-off in recall may have to be accepted, particularly in studies serving the generation of hypotheses.

4 Limits of cross-linguistic empirical research It is well known that direct comparisons of lexico-grammatical structures across languages can be misleading. They may be distorted due to different functions of the features in the respective languages. As Teich (2003) shows, the options of two language systems are increasingly divergent the more fine-grained the comparison is. Therefore, some kind of basis of comparison is used. König and Gast (2012: 5), for instance, employ general semantic notions such as ‘temporal relation’. In a similar vein, Teich (2003) argues in favour of comparing underlying functions. These problems on the level of contrasting language systems apply equally to quantitative studies. Comparing the frequency of occurrence of a given feature directly may be a questionable procedure since it may fail to recognise the different roles of the feature in the two language systems. A feature may be comparable in terms of the language system, but may be more, or less, common as expressed by frequencies of occurrence.

22 Koehn’s (2005) Europarl corpus, for instance, contains in version 3 approx. 407m words (see http://www.statmt.org/europarl/, last visited 12 April 2013)

Summary 

 47

Quantitative studies, however, offer the opportunity to factor out languagespecific frequencies provided that the corpus under investigation is stratified in terms of register and that a reference corpus balanced for register is included. Ideally, this reference corpus is comparable to the register-controlled corpus in terms of size, while at the same time covering a wide range of different registers that allows regarding the reference corpus as an indication of language average. In order to avoid a confounding influence from the specific characteristics of the language systems involved, an indirect method of comparison can then be adopted. The peculiarities of the language systems should be factored out by comparing only the magnitude of difference between the reference corpus serving as the baseline and the register-controlled corpus. This method will be used in the present study (see chapter 8).

5 Summary The present chapter has identified some links between linguistics, its systemic functional variety in particular, and the social sciences. These are of interest both in relation to the object of research of the present study, i.e. register variation, and the use of empirical methods in linguistics, which arguably have been adopted from the social sciences. The empirical methods were then examined in some detail, addressing some philosophical issues arising from their use in empirical research. By pinpointing the limitations of empirical research, these issues can create an awareness of the risk involved in overrating the findings of empirical studies rather than cause us to dismiss the value of empirical research completely. We then discussed aspects of qualitative versus quantitative approaches, concluding that they are not only complementary (see also Schlobinski 1996) but that both approaches actually overlap in certain respects. Whichever approach is deemed suitable for a given research question, scientific empirical research has to follow some guiding principles to ensure the quality of the study and overcome the limitations mentioned before. These quality criteria were introduced in the penultimate section before briefly addressing specific issues associated with cross-linguistic research. The following chapter will introduce the research design of the present study, which is based on the assumptions and principles discussed in this chapter.

Chapter 4 Indicators of register classification This chapter contains the description of the categories of register analysis. The findings for each feature under investigation as well as the interpretation can be found in chapters 6 to 9. The derivation of indicators is carried out in one stage for both languages with language-specific descriptions included where necessary. These are discussed in more detail in chapters 6 and 7. It is important to note that the mono- and cross-lingual comparisons as well as those between originals and translations require the individual findings to be juxtaposed. The following description of the operationalisations is biased towards the monolingual comparison of registers. Where applicable, adaptations of the data for the contrastive and translation-oriented analyses are described in chapters 8 and 9.

1 Field of discourse Field is the variable covering the description of what the register, i.e. a given situational context, is about, which experiential meanings are conveyed in the given situation to achieve what goal. Halliday and Hasan (1989) call it the nature of the social activity and the goals to be achieved by this activity. It is related to the ideational metafunction of language and is thus expected, in Martin’s (1992) words, to put the systems of transitivity, circumstantiation and agency at risk. Steiner (2004b) names the subdimensions of experiential domain, goal orientation and social activity with some operationalisations for each dimension. Social activity is a rather problematic category that is sometimes used instead of experiential domain (e.g. by Halliday and Hasan 1989), thus highlighting the social and action-oriented character of field. Sometimes it is also used to refer to a rather abstract idea of activity relevant within society like production, exchange, communication, reproduction and consumption (see Steiner 2004b). There seems to be a rather wide gap between the textual instances of a given register and very general types of activity in this sense. It remains unclear how these types of activity can be related to observable indicators in a methodologically sound way. While it would be interesting to investigate the types of activity represented by various registers, this aspect of register analysis is left aside here due to ­methodological concerns.

Field of discourse 

 49

1.1 Experiential domain This category’s task is to elicit the subject matter of the register. As mentioned above, it is related to what Halliday and Hasan (1989) call the nature of the social activity. However, they do not expand on how to elicit this activity, neither does Martin (1992: 540–542), who proposes a taxonomy of fields. While this is a very interesting approach, he concedes that he can only provide a very sketchy overview. We argue that the taxonomical approach to experiential domain is too abstract to be applied to the analysis of concrete texts, particularly because the field covered by a text attributed to a given register will be rather specific. So describing the subject matter of a register from below seems preferable even if this description is more specific and consequently less generalisable. Steiner (2004b) proposes the following operationalisations: lexical fields, terminology, lexical chains, transitivity, headings, paragraphing, expressions of time, perspective and Aktionsart. Offering the widest range of insights, the first four of these seem to be the most effective operationalisations. The present study therefore focuses on these and further condenses lexical field and terminology to a general survey of the vocabulary preferred in a given register. Vocabulary The most frequent items of vocabulary, or more specifically, the most frequent lexical items, are indicators of the experiential domain since they reflect the subject matter of a specific text and, more generally, of the respective register. The vocabulary used should give an indication of the lexical field covered by the text. Broadly speaking, “lexical field” refers to vocabulary belonging to the same semantic area. Obvious candidates for specifying the “same semantic area” are repetitions of frequent lexical items as well as all sense relations as described by Halliday and Hasan (1976) in relation to lexical cohesion. Since an annotation of sense relations is presently not available for the CroCo corpus, we will not analyse lexical fields here. Neumann (2010) concentrates on an analysis of repetitions of lexical items as a first indication of the lexical field for the registers under investigation there. The manual analysis of sense relations, even of only the most frequent lexical item per text, is beyond the scope of the present study, which focuses on a broad coverage of different indicators. A study aiming at the quantification of variation cannot determine the subject matter of each individual text in the corpus. The obvious approach, adopted by the present study, is to count the frequency of lexical items and process it statistically. Rather than determining the subject matter by reading and assigning an individual characterisation to each text, the subject matter is

50 

 Indicators of register classification

inferred from highly frequent lexical items. The following operationalisations reflect the balancing act of quantifying concrete vocabulary. The analysis consists of four types of comparisons exploiting both the individual texts and the complete registers in the form of cumulated frequencies. We start with a broad overview of the cumulated findings and gradually narrow down the view to the individual texts. First, we discuss the 10 most frequent lexical items per register, i.e. cumulated frequencies of the lexical items. This gives us a first idea of the variation in experiential domain in the different registers. Still in the realm of the cumulated texts in the registers, the second step consists in analysing the range of lexical variation by comparing the number of different types per register. We then zoom in on a closer analysis of the most frequent item found in the respective registers and evaluate which role this item plays in the individual texts and compare this role in the different registers. This gives an impression of the spread of the cumulated most frequent item in each register. Finally, we compare the range of frequency of the most frequent lexical item per text within one register in order to examine the level of variation within that register. A register exhibiting a high degree of lexical variability should be characterised by a low mean frequency of the most frequent lexical item in the individual texts. Lexical verbs and transitivity Transitivity is the grammatical system under which the clause represents experience as a flow of events (cf. Halliday and Matthiessen 2004: 170). This analysis affords insights into the typical lexico-grammatical patterns of referential information realised in a given register, whereas the analysis of the vocabulary discussed above only reflects the lexical choices in terms of frequency of occurrence. Transitivity chunks “the world of experience into a manageable set of process types” (2004: 170). We are thus mainly interested in the processes expressed by a clause. Halliday and Matthiessen (2004: 168–259) distinguish six process types. Material processes reflect concrete and hence material actions of “doing” and “happening”. This also includes metaphorical actions. These types of actions unfold in the here and now thus tending to select the present continuous (I am building a house). Relational processes, by contrast, link, or relate, two entities with each other in order to characterise or identify some element in terms of another (London is the capital of the UK). Characterising entities reflects a static phenomenon for which temporal dynamics do not matter; therefore the typical tense selection for these processes is the simple present. While material processes can be intransitive – to borrow a term from traditional grammar – relational processes require the two participant roles which are linked by the relational process. The default verbs used in relational processes are copular verbs, which are largely devoid of lexical meaning.

Field of discourse 

 51

Mental processes are concerned with experiences of our consciousness by expressing cognition, perception, emotion etc. They do not represent a material act. The participant role experiencing this process, the Senser, is (or is construed as) a being endowed with consciousness. These processes have a habitual flavour and therefore typically occur in the simple present (I don’t like my new boss). Finally, verbal processes represent all kinds of actual “saying” (she told me to leave), i.e. of “symbolic exchange of meaning” (Halliday and Matthiessen 2004: 253). The participant role of the Sayer is typically a human(-like) speaker but can also be another symbolic source (Martin et al. 1997: 108). Verbal processes also tend to be realised in a simple tense form such as the simple present. Martin et al. (1997: 120) further distinguish verbal from mental processes in semantic terms pointing out that verbal processes can combine with the participant role of a Receiver. Halliday and Matthiessen (2004) also describe behavioural processes, which share similarities with both material and mental processes, as well as existential processes, which are closely related to relational processes but consist of only one participant role, the Existent, and a placeholder element (e.g. the English there). These latter two process types are not analysed in chapters 6 to 9 because they are too marginal in frequency.23 Process types can be indicative of different kinds of interactions. For instance, a text which, compared to a reference corpus, contains significantly more relational processes will probably deal with the description of entities rather than material or verbal action. The qualitative investigator will manually assign values to process types based on the analysis of the participant roles and applying a number of classification tests (see Martin et al. 1997), affording a detailed view of experiential aspects of the texts under investigation. A study focussing on the quantification of register characteristics in a larger corpus cannot achieve a complete analysis of process types – interesting though it may be (see section 3 of chapter 1). The present study approximates the interpretation of process types in a three-step approach. First, the cumulated most frequent lexical verbs per register are examined in the same way as the most frequent lexical items described above. To this end, we use a simple query to retrieve items provided with part-of-speech tags identifying them as full lexical verbs. This query – or rather the part-of-speech tagger – is not capable of discerning between relational uses of to be or to have or their 23 Neumann (2003: 144, 164) reports the following distribution of process types in two small register-neutral corpora in English/German: material 49.15/50.87 %, relational 24.49/24.66 %, mental 13.67/12.07 %, verbal 10.06/10.53 %, existential 1.37/1.67 % and behavioural 1.26/0.19 %.

52 

 Indicators of register classification

German equivalents sein and haben and auxiliary uses of these verbs.24 In order to at least approximate the frequency of relational processes, we include a search for the third person singular present form of to be and sein (is and ist) in the corpora and count their relational uses. Of course, this only offers very limited information on the occurrence of to be and sein functioning as full verbs. The next step is again comparable to the investigation of the vocabulary in that we also examine the range of variation in lexical verbs by analysing the amount of verb types and tokens per register. The focus placed on the most frequent lexical verbs may blur the overall picture: it does not provide information on the distribution of process types in the given register. It is therefore complemented by a very simple analysis of process types conducted on the basis of the lexical meaning of the verbs concerned. It must be stressed that this analysis is extremely limited in its significance, the main advantage offered by a fully-fledged transitivity analysis, i.e. consideration of the context a process occurs in, not being included.25 Moreover, this analysis inherits the exclusion of to be/sein and to have/haben, a constraint which further limits the significance of the analysis. This limitation forbids a detailed interpretation of the results and means that the results can only give us a rough idea of the distribution of process types. This interpretation is, however, manageable on a large scale and is capable of shedding some light on the representation of the flow of events, i.e. the typical process types, pertaining to one register – and possibly provide information on distinctive differences between two registers or a comparable register in two languages. However, only a thorough transitivity analysis can detect general patterns in the distribution of process types. Lexical chains Lexical chains are closely related to lexical fields. While the latter are mainly defined semantically, the former are identified by the patterns formed by certain lexical items in a text. The most frequent items of vocabulary give us an important indication of the experiential domain covered by the register. Beyond the mere frequency of the respective lemma in a word list, lexical chains, i.e. sequences of related words (Morris and Hirst 1991: 23), provide information on

24 Another verb that is lost in the query is do in full verbal uses. It would probably have to be interpreted as a material process in these cases. 25 This is made feasible by more focused research designs which extract processes as concordance lines and then annotate them manually using a tool for semi-automatic annotation (e.g. Mick O’Donnell’s UAM CorpusTool, http://www.wagsoft.com/CorpusTool/index.html, last visited 12 April 2013). Holtz (2007) exemplifies this for certain verb-noun combinations in a specific domain.

Field of discourse 

 53

whether a frequent lemma forms a topical thread throughout the text or whether it is repeated only locally. In the former case the chain underpins the lemma’s relevance to the determination of the referential meaning of the whole text. In the latter case the lemma forming a chain merely represents a local strand of referential meaning. For a comprehensive analysis of lexical chains in a 1 million word corpus, the corpus has to be annotated with sense relations since semantically related items like synonyms, hyponyms etc. must be interpreted as contributing to a lexical chain (e.g. with the help of a WordNet and GermaNet annotation). Teich and Fankhauser (2004) describe how chains of lexically related words can be processed and analysed automatically on the basis of WordNet. The present account concentrates on chains created by repetitions of the same lemma. In Halliday and Hasan’s terms (1989: 84), chains consisting of repetitions represent similarity chains made up of items that refer to members of non-identical but related classes: The items in a similarity chain belong to the same general field of meaning, referring to (related/similar) actions, events, and objects and their attributes. (Halliday and Hasan 1989: 85)

This view is corroborated by Baumann (2006: 95), who argues from the point of view of intonation that implying the same referential entity is not necessary for identifying a repetition of a given lexical item as an anaphor, since the repetition is likely to be de-accented. Of course, “how much of such a [semantic] grouping will appear in the shape of similarity chains in a particular text” (Halliday and Hasan 1989: 85) is open to variation. Our query looks for each lemma and the sentence ID in which the given lemma appears. The sentence IDs are interpreted as representing a sentence’s linear position in a text, thus the IDs of two consecutive occurrences represent the distance between the two links in the chain. Morris and Hirst (1991: 23) also mention the span from the first to the last occurrence of the lemma within the text. This span is an additional cue to the relevance of the chain for the overall referential meaning of the text. We interpret the repetition of the lemma as a link in a continuous chain if the distance between the occurrences is less than four sentences (see Morris and Hirst 1991: 32). If the distance is longer, the new occurrence of the lemma is interpreted as a resumption of an existing chain. Chain lengths thus result from the addition of the number of occurrences no more than four sentences apart.

54 

 Indicators of register classification

1.2 Goal orientation In a very general sense, we can assume that linguistic texts are “a special case of products of human goal-directed actions” (Steiner 1991: 87). Consequently each text can be analysed with respect to the goal pursued by the speaker.26 The types of goals pursued by a text are often used as the sole criterion to classify texts and are then labelled text types (e. g. Werlich 1976, Brinker 2005). This classification is often based on the functions of texts. Brinker (2005: 100), for instance, defines textual function as the communicative purpose (“Kommunikationsabsicht”) pursued by the speaker. Engel (1991: 118) does not speak of text types but rather of text goals (“Textziele”). In this sense, our discussion of goal types also draws on these approaches. Nevertheless, we prefer the term ‘goal type’ over ‘text type’ since the latter term suggests a general validity that would supersede all other indicators – including the three register variables. Such an overarching classification based on very few types runs the risk of creating a wide methodological gap between the highly abstract concept of types and the concrete texts assigned to the respective types. Goal types, by contrast, are regarded as one indicator of the referential space covered by a register. This does not negate the possibility of classifying texts into text types, but this classification would have to be based on a broader derivation. While all accounts of register theory consider the goal of the (linguistic) action, different authors handle this category in different ways. Like Hasan in Halliday and Hasan (1989), Martin (1992) is more concerned with tenor of discourse and does not offer a detailed description. Hasan (1999: 234–237) explicates goal as an inherent aspect of human social action and thus as an important component of a text’s relevant context. She points out that the concept is “riddled with problems” mentioning among others the potential invisibility of goals. Arguably, an interview with authors will afford insight into those goals not explicitly marked in the text but more often than not we will be forced to content ourselves with the analysis of the product of interaction, i.e. the text. Lavid (1993: 5–6) describes the “communicative goals” of a text, i.e. the purposive role or social function of the text. Her use of the word “communicative” highlights an important difference to the above mentioned view of goal orientation. While the former attempts to frame the “material” intention in the mind of the interactants, the latter takes a step away from the material setting (Hasan 1995: 219, 1999: 230–231) and concentrates on the linguistic realisation of the intentions in the interaction. This communicative goal can be observed in 26 In dialogic texts “speaker” may refer to each interactant active in the interaction. As far as the role of the addressee in monologic texts is concerned, see section 2 of this chapter.

Field of discourse 

 55

texts. Lavid names interchange, exposition, persuasion or argumentation and instruction as purposes “that may ‘dominate’ in any given text” (Lavid 1993: 6). These are related to Werlich’s (1976) account of text types. Some of these types (“explaining, defining, generalising, reporting, recounting, narrating, chronicling etc”, Hasan 1999: 282) are also discussed in Hasan (1999: 281–282) under the heading of rhetorical modes (see also section 3.1 of this chapter). In a similar vein and with reference to Engel (1991), Steiner (2004b) lists exposition, instruction, argumentation, narration and persuasion. As Lavid points out, there is no strict categorisation since several types may be at work in the same text and different types may be prevalent in different text segments. Hatim and Mason (1990) discuss text types in relation to translation. Their typology closely follows Werlich’s typology and consists of the three main types exposition, argumentation and instruction and their subtypes conceptual exposition, narration, description (exposition), through-argumentation, counter-argumentation (argumentation) and with option, without option (instruction). However, there seem to exist clear differences between narrative and expository texts as discussed by de Beaugrande and Dressler (1981: 19027), who characterise descriptive texts as displaying conceptional relations for attributes, states, instances and specifications reflected in density of modifiers. Narrative texts, in contrast, are expected to have a high frequency of conceptual relations expressing cause, reason, purpose, enablement, and time proximity, these relations being reflected in a high density of subordination. While the limitation to modifiers and subordination as indicators for the respective text types seems too narrow, the overall distinction adopted by de Beaugrande and Dressler (1981) seems plausible. Other indicators such as tense (present tense for descriptive/expository texts and past tense for narrative texts), a marked difference in the usage of pronouns etc. may underpin the distinction. Although these goal types seem to represent a rather static classification, they provide a sound and comprehensive basis for deriving operationalisations of invisible, implicit goals. It should be possible, for instance, to identify features of argumentation in the text. Possible operationalisations could be the appropriate thematic structure of the text (cf. Lavid 1994 for an extensive discussion of thematic progression in different text types). Other observable indicators include modality, mood, voice and pronominalisation. The same derivation of indicators is applicable to the other goal types. Each goal type should thus be associated 27 The English citations of de Beaugrande and Dressler (1981) are taken from the on-line version of the book at http://www.beaugrande.com/introduction_to_text_linguistics.htm, last visited 12 April 2013, whereas the pages refer to the German version of the book from the same year.

56 

 Indicators of register classification

with a characteristic constellation of lexico-grammatical features. In the following section, we will discuss operationalisations for four goal types which will be analysed in chapters 6 to 9. The four goal types investigated in this study are argumentation, exposition, instruction and narration. By no means can we expect these four types to cover the whole range of possible goals exhaustively, to invariably apply to whole texts or to allow clear assignments of all registers in all cases. The assignment of texts to goal types requires the interpretation of the indicators that are assumed to represent linguistic manifestations of the given goal types. Only registers that have statistically significant results for several indicators are interpreted as pursuing the respective goal type. Since each register is analysed for all goal types, this also includes assessing which goal type scores the highest amount of significant results. Steiner (2004b: 16) lists “clause-mood, presence or absence of ellipsis, tense selection, transitivity, agency patterns in particular, patterns of identifiability, types, frequency and realization of conjunctive relations, patterns of thematic progression, typical patterns of topic construction, and paragraphing in written texts” as operationalisations of goal types. Examples from individual texts serve to illustrate how these features can be analysed. A problem with the following operationalisations is that, particularly in German, the authors discussing goal or text types do not elaborate on the concrete linguistic features indicating the respective types. Where they do, it is mainly in the form of examples without specifying which linguistic feature is at work so as to allow the assignment of the given goal type. This means that we mainly draw on the features described in the relevant literature on the English language. One aim of the present study is therefore to gain a better, evidence-based understanding of goal types in German.

1.2.1 Argumentation Argumentative texts are concerned with relations between concepts of phenomena, that is they are related to expository texts. Our focus will be especially on the “explicit or implicit opposition to deviant or alternative propositions” (Werlich 1976: 40), or, in de Beaugrande and Dressler’s words, to “the acceptance or evaluation of certain beliefs or ideas as true vs. false, or positive vs. negative” (1981: 190). We are thus again concerned with processes describing relations between entities, i.e. relational ones, namely “quality-attributing sentences” (Werlich 1976: 40). Werlich further subdivides argumentation into comment and scientific argumentation, describing the former as passing judgement by “relating concepts of events, objects, and ideas to his private systems of thought, values

Field of discourse 

 57

and beliefs” (1976: 107) and the latter “by relating concepts of events, objects, and ideas to systems of thought and knowledge so that the resultant propositions can be verified as valid or as valid hypotheses” (1976: 113). He assumes a preference for the use of the present tense in both subclasses. Biber’s (1995) inductive findings relating to the English dimension termed “overt expression of argumentation” additionally lists modals (subdivided into prediction, necessity and possibility), infinitives and conditional subordination. Modality should be complemented by modal lexis, i.e. downtoners, amplifiers, etc. reflecting the evaluative meaning contained in argumentative texts. In German, the subjunctive verb mood is an additional carrier of modality. As to mood selections, we assume that the declarative mood is preferred since argumentative texts merely expound their arguments while persuasive texts may also use imperatives to change the addressee’s opinion and employ interrogatives as a rhetorical means. Rhetorical devices may also be used in argumentative texts in the form of “cohesive devices for emphasis and insistence, e.g. recurrence, parallelism, and paraphrase” (de Beaugrande and Dressler 1981: 191). To this end, conjunction must be expected to be used more frequently. A varied thematic structure should reflect the elaborated rhetorical style as well. Many studies concerned with argumentation focus on the thematic progression as a realisation of argumentative logic (cf. Brinker 2005, Eggs 2000). The argumentation indicators introduced here should all be more frequent in argumentative texts than in language average. In Biber’s (1995) terms, they represent positive features. 1.2.2 Exposition Werlich (1976) distinguishes between description and exposition: the former is concerned with factual phenomena in space while the latter deals with the analysis or synthesis of constituent elements. Hatim and Mason (1990: 158), on the basis of Werlich’s text types, use exposition as a superordinate for conceptual exposition, narration and description. The sole difference between narration and description is that they deal with phenomena in time and space respectively. We would argue that this difference is not distinct enough, particularly since it can hardly provide a decisive criterion for the classification of entire texts. We will therefore include a description of phenomena in space as a part of narration and assign descriptive texts to exposition on the basis of the following description by de Beaugrande and Dressler (1981: 190). Descriptive texts would be those utilized to enrich knowledge spaces whose control centres are objects or situations. Often, there will be a frequency of conceptual relations for attributes, states, instances, and specifications. (Original emphasis)

58 

 Indicators of register classification

In line with this characterisation of conceptual relations, Werlich (1976) identifies “phenomenon-identifying” and “phenomenon-linking” sentences as the key criteria for classification. In the systemic terminology, these are relational processes. We would thus assume to find a high frequency of relational processes in expository texts. Werlich also presumes a preference of the present tense group “with the timeless Present Tense as the dominant tense” (1976: 73, see also Heinemann 2000c). He proposes a distinction between what he calls “expository essays” described as adopting a subjective point of view and variants offering an objective point of view. Since this distinction blurs the difference between argumentative and expository texts, we prefer to assign texts that convey a personal stance to argumentation. The main distinctive feature in this context is the absence of personal pronouns that enables the focus to be placed on factual information in order to create an objective perspective. The only pronoun that can be assumed to occur frequently in fact-oriented texts is the third person singular it in English and es in German, both of which can be used to refer to objects. A non-personal point of view can be expected to be supported by a high frequency of the third person singular pronoun it/es, which makes “phenomena appear as the source of actions and processes” (Werlich 1976: 137). The present study adopts the negative features of Biber’s (1995) dimension “involved vs. informational production” for exposition. Biber identifies, among other features, a high frequency of nouns and attributive adjectives as well as a high type-token ratio, all of which represent indicators of nominal style. De Beaugrande and Dressler (1981: 190) mention an increased “density of modifiers”, which corresponds to Biber’s attributive adjectives. A high frequency of nominalisations is an additional indicator of nominal style. Nominalisations can function as a grammatical metaphor referring to goings-on that would otherwise have to be expressed in the form of a clause employing the congruent verb form of the nominalisation. The grammatical metaphor can then function as an entity and fill a participant role in relational processes. Halliday and Martin (1993) identify this as a characteristic of scientific writing, a domain mainly concerned with expository information. Biber (1995) mentions a high frequency of agentless passives and also, with respect to the dimension “abstract style”, by-passives. Heinemann (2000c) views this indicator as a characteristic of German descriptive texts along with static forms, verbs and impersonal constructions. The fact-oriented language of expository texts suggests that there is no need for mood variation: Since exposition is concerned with the exchange of information rather than goods and services, we can exclude imperative mood in the present context. Since the goal of these texts is to make statements rather than enquiries, we can expect the declarative mood to be the prevalent one.

Field of discourse 

 59

1.2.3 Instruction The instructional text type is concerned with the planning of future behaviour, the speaker “tells himself (in sender-directed instruction) or others (in receiverdirected instruction) what to do” (Werlich 1976: 40). Werlich further divides instruction into the subjective and the objective point of view. The former is concerned with “demands for a certain behaviour” put forward by the speaker, implying a certain degree of personal authority over the reader. The objective point of view subsumes “the text form variants of directions, rules, regulations and statutes” (1976: 121), which are further divided into “practical” and “statutory” instruction (1976: 127–128). As will be discussed below, we only include Werlich’s objective type under the heading “instruction”, with the other belonging to a possible goal type “persuasion” (see below). Werlich identifies “action-demanding sentences” as the constitutive characteristics of instructional texts. In systemic terms, this would mean sentences realising a proposal, the speech function of exchanging goods and services, typically realised by imperative mood. This frequent use of imperatives will also prompt a frequent use of the present tense (1976: 129). We will also expect a bias towards material processes as compared to language average reflecting the speech function of exchange of goods and services as well as the expression of procedural information. Werlich continues to describe objective instruction as taking either a non-personal third-person or a second-person point of view (1976: 128). This means that we will find few pronouns or, if at all, second person pronouns required to address the reader. This is particularly true for German where the imperative is used in the polite form (see Teich 2003: 105). The procedural steps explained in instructional texts can be expected to be realised mainly in the active voice, thus expressing the agent responsible for the material action. Furthermore, we can expect an experiential iconicity (Enkvist 1991: 10) in the sequence of the actions to be carried out in order to influence the reader’s behaviour in the way and order intended by the instructional text. This should be reflected in the thematic progression.

1.2.4 Narration Werlich (1976: 39) describes narration as the type of communication concerned with factual and/or conceptual phenomena in time. Following Werlich (1976), Hatim and Mason (1990) subsume narration as one of two subtypes of exposition, the other one being description. However, it seems plausible to classify description as a type of expository text distinct from narrative texts. These are characterised by the “action-recording sentence” (Werlich 1976: 39). Gülich and Hausendorf (2000: 373) name three distinguishing features of narration, i.e.

60 

 Indicators of register classification

verbal reconstruction, discourse unit and action/event. This latter feature is also mentioned by de Beaugrande and Dressler (1981: 190) stating that narrative texts are “those utilized to arrange actions and events in a particular sequential order”. Linguistic studies of narrative texts typically view narration from a very broad perspective covering various mediums, contexts and genres (Gülich and Hausendorf 2000: 370). This is an important aspect for the general analysis of goal types. Verbal reconstruction suggests the frequent use of past tense, since narrative texts report on things in the past (see also Werlich 1976: 57). In addition to the complementary sparsity of present tense verbs, Biber (1995: 152) also identifies the relative absence of attributive adjectives as a negative feature (in his terminology) of narrative discourse. Verbal reconstruction also involves recounting dialogue. We can expect an increase in verbal processes, i.e. the symbolic exchange of meaning (Halliday and Matthiessen 2004: 253), and more fragmented structures using full clauses rather than condensed nominal groups and phrases. The preference for more fragmented structures will be reflected in an increase of verbs and a decrease of nouns and nominalisations. Furthermore, dialogic passages of narrative texts should require more mood variation, particularly in terms of interrogatives. The subjective point of view frequently adopted in narrative texts (Werlich 1976: 55)28 leads us to expect a high frequency of mental processes, i.e. processes of perception, affection and cognition (Halliday and Matthiessen 2004: 198–199). This may also go hand in hand with a high frequency of personal pronouns. According to Biber et al. (1999: 70) pronouns are used in situations where the entities referred to are identifiable through the situational or textual context or where the reference is unknown or unspecified. They observe a particularly high frequency of pronouns in conversation and – to a lesser degree – in fiction (Biber et al. 1999: 92). Personal pronouns may serve the purpose of referring to the persons or entities involved in the actions and events being related. Particularly in accounts of dialogues, personal pronouns may also be used to address the participants in a dialogue. Referring to a higher frequency “of conceptual relations for cause, reason, purpose, enablement, and time proximity”, de Beaugrande and Dressler (1981: 190) posit a high density of subordination in narrative texts. Argumentation, exposition, instruction and narration are the goal types that will be discussed in the present study (see chapters 6 to 9). The restriction on only four types inevitably gives a coarse overview of types of goals and leaves many types undescribed. One possible additional type, which is related to 28 Thus narrowing the scope of Werlich’s (1976: 55) account which also includes the report as narration from an objective point of view.

Tenor of discourse 

 61

argumentation­, is the persuasive type. Not only does it aim to inform the addressee about an evaluative judgement as is the case of argumentative texts, but it is intended to convince the addressee of the speaker’s own judgement (cf. Brinker 2005: 120). Hatim and Mason (1990) point out that Reiß (1976) treats argumentative and instructional text types as one category under the heading of “operative text type”. Similarly to Reiß’ operative text type, Werlich’s (1976: 121) notion of instruction also includes advertisements; these are both regarded as belonging to the category of ‘subjective point of view’ (see above). We would rather suggest that there may be a cline from instructional goals to argumentative goals, which can be represented by a fifth goal of ‘persuasion’ that is not investigated further in the present study. Other possible goals often included in accounts of these types are description (see above), explication, interchange etc. It would be interesting to investigate the distinctiveness of the various types in future work. The comprehensive range of the indicators characterising the different goal types suggests that the variable ‘goal orientation’ has a more general character than the other variables. However, there are two counter-arguments: First, in the framework of register analysis, it seems of great interest to gain some insight into the potential goals pursued by the speakers, even if the information available from the text as the final product is limited. These goals need to be described as comprehensively as possible. Secondly, if this analysis relied solely on this classification, we would lose valuable information, mainly on interpersonal, but also on textual features of the registers.

2 Tenor of discourse Tenor of discourse comprises criteria that capture the nature of the relationship between the interactants. The relationship is analysed with respect to agentive as well as social roles borne by the interactants and their social distance. Sometimes appraisal or affect (Poynton 1985) are also included and will therefore be discussed below as well. The interpretation of these subdimensions is subject to the following restrictions. The analysis carried out in this study focuses on the speaker since we expect that, in most cases, one of the interactants has a more active role and that there is a (group of) interactant(s) re-acting to what the active interactant produces. The speaker is the active part while the addressee is the re-active part. Interactants may swap roles in the course of the interaction, particularly in certain spatio-temporal constellations. In interactions that contain contributions by all interactants, the analyst can interpret both the speaker’s and the addressee’s part in the interaction.

62 

 Indicators of register classification

In written texts, however, the roles are fixed, with the author of the text being the speaker and the reader being the addressee.29 This has consequences both for the interactants and the analyst of written texts. As Hasan (1999: 230) explains, the addressee is unable to influence the process of text production. In the case of, say, published texts written without any knowledge of the concrete readers, the speaker can only imagine a prototypical reader, or as Hasan (1999: 229) puts it, “the intended addressee of this text has an imaginary being”. And finally the analyst can only look at the active part’s output, not at the anonymous addressee’s reaction to it.30 The analyst may infer from the interpretation of the findings about the speaker to the audience. In Hasan’s words (1999: 238): where the addressee is virtual, all aspects of the interactant relation – their respective status, their social distance, the specific attributes of the addressee – are logically entirely created by the language of the text, none having a basis in reality for obvious reasons.

This means that for the analysis of texts written for anonymous readers, as is the case with the texts under investigation in the present study, all statements regarding tenor of discourse and its subdimensions will relate to the speaker. The conclusions drawn with respect to the addressee are only indirect inferences obtained from interpreting what we assume to be the speaker’s projections of his/her addressee(s). The parameters analysed under the heading of tenor of discourse are to some degree indebted to sociological findings. Interactants are analysed as individuals who are part of society. The way they interact linguistically can be assumed to be influenced by their social positions and relationships. This becomes particularly obvious when we look at the titles of some of the subdimensions analysed under this register variable, which include social role relationship or rather social hierarchy, terms used in sociological research (see, for instance, Mead 1934, Linton 1936, Dahrendorf 1977). Hasan (e.g. 1973) draws on Basil Bernstein’s work (e.g. 1971), which relates social class to language use. Even the title Language as a social semiotic (Halliday 1978) reflects the importance attached by Halliday to society. Hasan (1999: 220) points out that language is an inherently variable meaning potential (Halliday 1978) which varies with variation in its speakers’ material and social conditions of living.

29 He or she may react to the speaker, but this would be considered a different text (cf. Hasan 1999 for a discussion of such related texts). 30 An additional problem for the analyst may arise from the fact that there may be little or no information available on the speaker aside from his/her output, the text under investigation.

Tenor of discourse 

 63

Social aspects become a pervasive influence if we posit that the description of the nature of human language requires language to be considered within the context of its social environment (Hasan 1999: 224). Although these social aspects have an impact on all facets of context, i.e. all register variables, they are most directly observable in the relationship between the participants in an interaction in a given situational context, i.e. in the tenor of discourse. 2.1 Agentive roles Referring to agent roles, Halliday and Hasan (1989: 56) elaborate that [t]his social activity [i.e. buying food-stuffs] is institutionalised. And so the nature of the activity predicates the set of roles relevant to the unfolding of the activity.

They continue to identify vendor and customer for the text in question. The roles here seem to be related to the participant roles in terms of lexico-grammar. In Hasan (1999: 247), the focus shifts slightly to a more abstract view that helps shed additional light on the interactants’ roles in communication. Referring to a mother and child interaction, Hasan explains that “it is difficult to identify one single relation of the agentive kind which would apply constantly to the entire dialogue”. Particularly in a context where both the speaker and the addressee are present in some form, the question of who is the agentive part in the interaction becomes relevant. Agentivity, i.e. the active control of the interaction, can vary independently of the interactants’ social role or their social distance, thus requiring a separate analysis. Indicators could be the proportion of turns per interactant and distribution of different mood and modality options among the interactants. In written texts where the addressee remains virtual in the sense discussed above, the analysis of agentive roles is pointless since the interaction is completely sustained by the speaker. Agentive roles are therefore not included in the present study. 2.2 Social role relationship Social role relationship, sometimes also referred to as social hierarchy, is concerned with the degree of control (or power) one interactant has over the other (Halliday and Hasan 1989: 57). This may be, as Halliday and Hasan write, “almost by virtue of their agent role relationship”, but other agentivity constellations are conceivable as well. Therefore, it is worthwhile to conduct separate analyses of the two subdimensions. The assessment of social role relationship should enable us to determine whether speaker and addressee hold equal social roles or

64 

 Indicators of register classification

whether they are in a hierarchical relationship. These roles should be reflected in the interactants’ linguistic choices. The social role taken up by an individual in a given situation31 depends on the person’s levels of authority, expertise and education. Other aspects contributing to an individual’s position in the social role relationship are religion, gender, sexual orientation etc.32 An analysis of level of expertise, for instance, should show whether there is a difference in expert knowledge between the interactants.33 It should be stressed, however, that an analysis of the product of the interaction in the form of a written text can only enable us to establish the author’s level of expertise as well as his/her expectations regarding the reader’s expertise as shown by the presence or absence of, say, explanations of technical terms. It is thus difficult to determine on the basis of a monologic product of communication whether the social relationship between the interactants is hierarchical or not. This also includes the interpretation of the speaker’s projection of the addressee’s social role: Indicators for, say, a high level of expertise can be read as either pointing to an equal relationship if the addressee is assumed to share the author’s high level of expertise in the case of expert-to-expert discourse or as pointing to an unequal relationship if the addressee is expected to possess less expertise in communication between expert and lay person. More often than not, there is no indication in the text as to which level of expertise the speaker expects on the part of the addressee. Dreitzel (1980: 103–104) explains from the sociological point of view that the various social roles played by the individual depend on the situational context. It is not up to the individual to decide which role to play, rather, the options are determined by the situation. The present study analyses social roles in terms of level of authority, level of expertise and education. The other aspects noted above as contributing to the role in society borne by the individual in the interaction appear to have a somewhat weaker impact on the concrete make-up of the linguistic output produced by this individual. They are therefore not further pursued here. This does, however, not mean that their sociological impact is negated. 31 Note that social roles are not fixed with respect to a certain person but will change in different contexts. 32 “It will be noted that age, gender, and race, familiar sources of power unequality, are not mentioned. This is because all of these are seen as matters of ideology. As we will see in the case of gender, all three dimensions of social relations are affected by it, so that gender must be outside this network and not part of it.” (Poynton 1985: 77) 33 This part of the analysis is often subsumed under the heading “agentive roles”. However, expert knowledge does not necessarily govern who acts as the agent in the context of situation: rather, it is concerned with the interactants’ roles in society. It is therefore analysed as part of the subdimension social role relationship.

Tenor of discourse 

 65

2.2.1 Level of authority Max Weber’s classical definition of authority refers to the power which is recognised as legitimate and justified by both the powerful and the powerless. Arendt distinguishes authority from power as well as from persuasion: Since authority always demands obedience, it is commonly mistaken for some form of power or violence. Yet authority precludes the use of external means of coercion; where force is used, authority itself has failed. Authority, on the other hand, is incompatible with persuasion, which presupposes equality and works through a process of argumentation. Where arguments are used, authority is left in abeyance. Against the egalitarian order of persuasion stands the authoritarian order, which is always hierarchical. If authority is to be defined at all, then, it must be in contradistinction to both coercion by force and persuasion through arguments. (The authoritarian relation between the one who commands and the one who obeys rests neither on common reason nor on the power of the one who commands; what they have in common is the hierarchy itself, whose rightness and legitimacy both recognize and where both have their predetermined stable place.) (Arendt 1961: 92f.)

We discussed persuasion as one possible goal type above. In line with Arendt’s view, Poynton (1985: 76) differentiates ‘force’ involving physical superiority and authority. Investigating inequality in social relations, she describes authority as “a function of socially-legitimated inherently unequal role relationships such as parent-child, teacher-child, employer-employee, or ruler-ruled” (1985: 76). Additional intermediate categories are ‘status’ as a ranking in relation to some uneven, but desirable standing or achievement as well as ‘expertise’. While expertise is included in the present study (see below), force and status are not, force not being expressed verbally and status being conflated with authority. The focus of the present study on written language requires us to restrict our attention to identifying features of authority in the speaker’s linguistic output, and, as a result of concentrating on this aspect, (in-)equality or (a-)symmetry in the interaction are not investigated in any further depth in the current study. The speaker’s authority may be observable in a text in the form of mood, modality, reciprocity of vocation (Poynton 1985: 81). In connection with mood, Poynton refers to the “extent of congruence in relation to speech function” (1985: 81). This must be seen in the light of interactions where an interactant in a subordinate position makes use of interpersonal metaphor (Halliday and Matthiessen 2004) in the form of an interrogative rather than an imperative as the congruent option. This kind of metaphorical meaning is, however, difficult to quantify. Furthermore, we would expect it to occur as a feature of dialogic discourse rather than in the monologic written texts contained in the CroCo corpus. Reciprocity of vocation is another feature that does not play an important role in written texts. Modality, however, is seen as an important feature, since “the use of modalised

66 

 Indicators of register classification

or modulated clauses rather than ones with definite positive or negative polarity […] is commonly dependent on whether one is subordinate or superior to one’s addressee” (Poynton 1985: 79). The interpretation of the features is complicated by the fact that an unequal social relationship is not the only reason for the use of mood or modality. Academic hedging in peer interaction, for instance, can be another factor causing the use of modality.

2.2.2 Level of expertise According to Poynton (1985: 77), “expertise is a matter of the extent to which an individual possesses knowledge or skill”. It thus refers to specialised knowledge in a particular area. Unlike authority, it is not directly linked to the exercise of power, even though the expert may be in a position of power by virtue of his/her high level of expertise. Observable indicators for level of expertise are features of language for specific purposes (LSP, see for instance Sager et al. 1980, Halliday and Martin 1993, Martin and Veel 1998, Fluck 1996) like LSP terminology and grammatical density. Over time, every specialised subject field creates its own terminology, i.e. lexical items with a meaning specific to its use by this subject field. The frequency of occurrence of field-internal terms is indicative of the level of expertise in a given subject field. Grammatical structures typical of LSP texts have been described as packing more information into (nominal) groups and phrases and at the same time reducing the complexity of the clause structure (cf. Halliday and Martin 1993, Ventola 1996). This does not necessarily mean that the language is more complex on all levels (Halliday and Martin 1993: 67). Steiner (2005: 22) refers to this phenomenon as grammatical density, assuming that informationally dense texts include a high proportion of “‘intermediate phrase types’ (groups, phrases, rather than words or clauses) per clause”. A register is therefore interpreted as being grammatically dense if it contains a higher ratio of tokens per chunk and a lower ratio of chunks per clause and sentence as well as clauses per sentence. Less specialised registers are expected to contain a relatively higher proportion of chunks and clauses per sentence and relatively fewer tokens per chunk. The number of tokens per sentence should be generally lower. However, it could also be equal to the proportion found in more specialised registers since the tokens may either be squeezed into intermediate phrase types in specialised registers or distributed across several clauses in the case of more general registers. Another way of condensing information into smaller units is nominalisation. This is a potential (observable) output of grammatical metaphor, a phenomenon characteristic of specialised texts as explicated by Halliday (in Halliday and Martin 1993).

Tenor of discourse 

 67

2.2.3 Level of education Education may have an impact on the individual’s position in society and thus affect the way the individual interacts in a given situational context. Although it is related to the category of expertise to a certain degree, level of education can still be thought of as a distinct category. An interactant may be a specialist in his/her field and consequently make use of specialist language, he or she may, at the same time, have a low level of education and vice versa. Elaborate vocabulary, complex and intricate grammatical structures, lexical density etc. represent potential observable indicators of a high level of education, while spelling or grammatical errors, simple grammatical structures etc. could serve as indicators of a low level of education. Given that the corpus used for the present study consists of texts that were translated and published in some form, there is likely to be little variation in terms of the speakers’ level of education. This subdimension will therefore not be included in the analysis in chapters 6 to 9.

2.3 Social distance The subdimension social distance encodes the interactants’ relationship based on their mutual interactive history in terms of density as well as formality of context. Martin (1992) uses Poynton’s (1985) term “contact” for this subdimension. As Halliday and Hasan (1989: 57) write, this subdimension represents a continuum whose end-points are maximal and minimal. The continuum is specified by House (1997: 41–42) on the basis of Joos’ (1962) categorisation of levels of formality for spoken English, namely frozen, formal, consultative, casual and intimate. The present study draws on House’s description, particularly with respect to observable indicators used to identify the degree of social distance reflected in the corpus. The categorisation has, however, been adapted to written language, particularly to cover texts showing no distinct indication of formality. The main reason for this modification is the fact that the relationship between speaker and addressee may not be realised explicitly by any linguistic means, resulting in a style that can be described as neutral. The intermediate categories analysed here are ‘neutral’, representing factual information and thus actually not being part of the continuum, ‘casual’, covering all shades of social “closeness”, and finally ‘consultative’, representing a higher social distance.34 A different approach to the subcategorisation of this dimension is taken by Poynton (1985: 77–78), who is interested in the frequency of interaction, its extent 34 This coarse categorisation is owed to the constitution of the corpus which does not reflect the fine-grained distinction in, say, Joos’s categorisation.

68 

 Indicators of register classification

in time, the extent of role-diversification and the orientation towards persons or tasks. These criteria appear to narrow down the characterisation of the relative distance between the interactants very plausibly. Nevertheless, they require very specific extra-linguistic knowledge of the concrete interactive history of the participants. As discussed above, this kind of information is not available for the texts analysed in this study. From the point of view of research method, this characterisation would require interviews with the interactants whose discourse is analysed. The present research design requires operationalisations retrievable from the product of discourse, i.e. the texts. Each of the three styles introduced above have to be operationalised in terms of observable indicators. Without assigning them to a concrete style, Martin (1992) mentions tone, accent, ellipsis, vocation and terminology as operationalisations. Steiner (2004b) lists tagging, forms of address (i.e. vocation), modality, accents, dialects and sociolects. Operationalisations mentioned by Poynton (1985: 81) are vocation, truncation, suffixation, slang and elision (i.e. ellipsis). House’s (1997: 84–86) distinction between orientation towards addressee and towards content (or, in Poynton’s (1985: 78) terms, “towards tasks”) provides a good starting point for identifying observable indicators for social distance. Indicators for orientation towards content can be used to identify a neutral style. Indicators for orientation towards addressee require additional qualification in terms of the two styles representing the limits of the range defined by minimal and maximal social distance, with consultative also carrying indicators for orientation towards content. Relevant features can also be adopted from Biber’s (1995) dimension “involved versus informational production”. Features interpreted by Biber as representing involved production are used here as indicators of orientation towards addressee, and features considered by Biber as representing informational production (i.e. the negative features of his dimension) are adopted as indicators of orientation towards content. Another account of observable indicators concentrating on features for orientation towards addressee is given in Hyland (2001). The indicators discussed below can be organised into three groups, each of which covers different characteristics relevant to the analysis of social distance. In addition to orientation towards content and towards addressee mentioned above, the third is the expression of, or reference to, shared communicative common ground including background knowledge.

Tenor of discourse 

 69

2.3.1 Neutral style Neutral style is marked by the absence of any signs of interaction between speaker and addressee. As discussed above, registers representing neutral style are therefore expected to be characterised by the relative absence of markers of orientation towards addressee. More concretely, the frequency of indicators of a shared common ground with respect to the situational context and/or meanings referred to should be below average. These indicators comprise contractions, colloquialisms, deictic elements represented by place and time adverbs and ellipses. Personal pronouns also represent a relevant indicator as they serve to express the relation between speaker and addressee. Other indicators for orientation towards addressee include forms of address, titles, salutations, interrogatives and imperatives. There should be a significantly lower frequency of occurrence of these features in neutral texts than in the reference corpora. Markers of orientation towards content, by contrast, should be significantly more frequent than in the reference corpora (see also Oesterreicher 2001: 1568). These comprise nouns, attributive adjectives, prepositions and lexical density (see section 3.3 of this chapter for a closer description of this indicator). This latter indicator represents a deviation from Biber (1995). He includes type-token ratio and word length as indices of the variation of vocabulary. Lexical density is not among his indicators. Word length is excluded from this study because it does not appear to constitute a sufficiently significant indicator for register variation.35 There are two reasons why type-token ratio is not included either: First, the differing lengths of the texts in the CroCo corpus do not allow comparisons between the registers because type-token ratio decreases with increasing text length. Secondly, lexical density appears to be more appropriate to the investigation of orientation towards content since it focuses on content words. Biber et al. (1999: 62) discuss the “differences in information load” reflected by lexical density explaining a high lexical density found in news reportage by its purpose “to convey information, and preferably as concise as possible”. A high type-token ratio may also reflect a high frequency of function words, thus weakening the explanatory power of this indicator.

35 Word length would be interesting in connection with an expected higher frequency of compounds in fact-oriented texts. The contrastive comparison of this indicator appears highly problematic because of the differing spelling conventions in the two languages.

70 

 Indicators of register classification

2.3.2 Casual style Casual style covers all linguistic realisations of some degree of social “closeness” and thus combines House’s (1997) (and Joos’s 1962) intimate and casual styles. According to House, casual style is characterised by degrees of implicitness due to the knowledge shared by speaker and addressee, making background information superfluous. Intimate style is an extreme case of casual style used between individuals on very close terms, who therefore share a high level of background information (see House 1997: 41). Discourse characterised by a casual style primarily uses linguistic means expressing orientation towards addressee. In comparison to a reference corpus, interactions using casual language should contain more contractions, colloquialisms, ellipses, forms of address, personal pronouns, deictic elements, interrogatives and imperatives.36 Casual style is characterised by the relative absence of markers of orientation towards content in comparison to the reference corpora. Nouns, attributive adjectives and prepositions are expected to occur less frequently than in the reference corpora. Casual discourse is expected to be less lexically dense than the reference corpora. Although titles are viewed as realisations of orientation towards addressee, they are associated with a formal style and are, therefore, likely to be infrequent in casual style.

2.3.3 Consultative style According to House (1997: 41), consultative texts supply fairly elaborate background information, i.e. full lexical instead of pronominal reference, and a general absence of both formal and informal style markers. Consultative style can be distinguished from neutral style by the direct or indirect participation of the addressee, which is characteristic of the consultative style. The present study merges consultative, formal and frozen styles under the heading of consultative style. In monologic written texts, the participation of the addressee may be observable in the way s/he is addressed. By addressing the reader, the speaker involves the addressee in the interaction. This can be realised by imperative or interrogative mood or a form of address, the use of titles (in a more formal setting) as well as the use of salutations. Apart from these, markers of shared background knowledge should not be frequent in a consultative style of language. In particular, there should be below average frequencies of contractions, colloquialisms, ellipses etc. since these require a certain amount of communicative common ground. 36 Eggins and Slade (1997) discuss more fine-grained features that also combine some of these indicators.

Tenor of discourse 

 71

Representing a certain degree of social distance, consultative style is also expected to show some characteristics of orientation towards content. More specifically, nouns, attributive adjectives and prepositions are assumed to be more frequent than in the reference corpora. The lexical density of these texts should exceed the reference values. 2.4 Appraisal A final aspect that may characterise the relationship between speaker and addressee are evaluative or emotional aspects of the situational context. Poynton (1985: 76) describes it as “an attitudinal dimension concerned with attitude or emotion towards addressee (or towards the field of discourse)” and calls it “affect”. The notion has since developed into what is called appraisal theory.37 Under the heading of ‘attitude’, this view distinguishes between emotional (‘affect’), ethical (‘judgement’) or aesthetic (‘appreciation’) meaning. Furthermore, the interpersonal force the speaker attaches to an utterance (‘graduation’) and linguistic means to vary the speaker’s engagement in what is expressed (‘engagement’) are included as well. Recent years have seen various studies of this highly important aspect of social interaction (cf. for instance Thompson and Hunston 2000, Martin and White 2005). However, it remains unclear how indicators can be identified that are capable of being adequately operationalised and suitable for use in connection with large amounts of data. As to lexical selections, the interesting and challenging part is to tease out not only explicitly evaluative lexis but also implicit meaning in a systematic and quantitative way.38 While computational models may appear too shallow, linguistic approaches in this field are prone to using blurred and indistinct descriptive categories of classification. As a result, the assignment of instances in texts to the descriptive categories appears subjective, failing to produce intersubjectively verifiable and systematic results. Moreover, this subdimension particularly highlights issues in contrastive comparison. Little work has been done to compare the options of evaluation in English and German (one exemption being Bublitz 1978).39 The study of the language of evaluation in German (under the heading of “Sprechereinstellung”) is mainly concerned with modality and is not embedded in the systemic framework. Since it is not within the scope of the present study to provide a more detailed 37 http://grammatics.com/appraisal/index.html, last visited 12 April 2013. 38 See also the work in computational linguistics on sentiment analysis and subjective language, e.g. Wiebe et al. (2004) or Taboada et al. (2006) 39 cf. Ochi (2005) on the comparison of English and Japanese

72 

 Indicators of register classification

comparison, we will only name some possible indicators without presenting a fully-fledged analysis. Martin (1992) names tone, attitude, comment, intensification, repetition, mental affection, manner degree and attitudinal lexis as indicators. Steiner (1998a) supplies a very general list comprising lexical selections, grammatical choices and rhetorical devices such as repetitions, parallelisms etc. Mental processes and evaluative patterns represent further possible – and potentially comparable – indicators (cf. Bednarek 2007). The latter indicator works on the basis of pattern grammar (Francis and Hunston 2000) and aims at extracting word order patterns characteristic of evaluative contexts. Although it is possible to replicate Bednarek’s (2007) evaluative patterns using the CroCo annotation for the German language, these structures are, however, rare in German. This suggests that either evaluative meaning does not play the same role in German texts as in English ones or – more plausibly – that patterns do not adequately reflect evaluative meaning in German. It is presumably due to language typological differences between English and German that this feature works better for English than German with its more flexible word order, which allows more variation in terms of positional patterns. Due to the various restrictions applying to a sound quantitative analysis of appraisal, the present study does not include this subdimension.

3 Mode of discourse The final of the three register parameters is concerned with how language is organised in texts to reflect the social action between speaker and addressee. This parameter is based on the assumption that the means of message transmission have an impact on the text’s language. We are therefore interested in how much language contributes to accomplishing an intended social action (language role), how the text is transmitted (channel) and finally medium, i.e. whether the interaction takes place in written or spoken mode (since this influences the organisation of the text).

3.1 Language role Speakers respond to different situations by relying to a greater or lesser degree on language to achieve their goal. Extreme examples would be cases where a nod may be sufficient to convey the intended meaning or a comic strip where the story may be realised almost completely in pictures with language only coming into play at some focal point like the punch line. In cases like these, language

Mode of discourse 

 73

serves what is called an ancillary role, merely supporting the information provided mainly through other semiotic means. At the other end of the continuum, social action may be realised entirely by means of language without relying on any kind of material action to support the verbal action. This role of language is constitutive of the verbal action. Hasan (1999: 281–282) argues conclusively that this subdimension belongs to field of discourse. However, her main point is that “the so-called rhetorical modes such as explaining, defining, generalising, reporting, recounting, narrating, chronicling etc. are best viewed as constitutive verbal actions” and thus as specifying the nature of the social activity, which is in turn analysed under field of discourse. In our account these modes are – in line with Hasan’s line of argument – treated as goals of verbal action and analysed within the context of the subdimension goal orientation (see section 1.2 of this chapter). The present study takes as its central focus written text – any material action outside the printed or otherwise displayed text is therefore not taken into account. Language role as analysed under mode of discourse covers a different aspect of verbal action, one that can be categorised more clearly as belonging to the study of the textual make-up of the register. In this perspective, the non-verbal parts of the text (and they are clearly a part of the text) do not represent a different field; rather, they present the subject matter of the text in a different mode. We are concerned with the interaction between verbal, i.e. linguistically encoded, parts of a text and other semiotic modes like photos, figures, graphs etc. The presence or ab­ sence of these modes is likely to impact upon register-specific language use. If, for instance, instruction manuals (our INSTR register) are found to make extensive use of non-verbal presentations of the object described to convey meaning, the verbal parts of the text may not explicitly describe the object but rather use endophoric reference in the form of deictic elements, pronouns etc. to complement the figures (cf. Bartsch’s (2007) discussion of cohesion between the different modalities in texts from the field of mechanical engineering). Figure 1 displays a case in point: Here, the verbal part only makes reference to the figure by means of the local adverb here (and the colon). The meaning is conveyed solely by the drawing. This case represents an ancillary use of language. By contrast, a literary text may describe the same object in quite a different way, completely depending on verbal means, thus using, for instance, lexical reference as well as material and relational processes to describe the object. Here, language is the sole means by which the intended meaning is conveyed. Its role is therefore constitutive. The operationalisations are thus concerned with the two roles of language, namely ancillary and constitutive (Halliday and Hasan 1989: 57). It seems plausible to view the options as two ends of a cline, but this requires an interpretation of the findings (see chapters 6 to 9).

74 

 Indicators of register classification

Figure 1. Relation between different modalities in an example from EO_INSTR

3.1.1 Ancillary language use Observable indicators of the auxiliary role fulfilled by language are ellipsis since elliptical sentences may omit information the addressee can retrieve either from other modalities in the text or from the situational context and imperative mood because it commands or requests a material action. The broad field of reference can be further specified with respect to reference to items not further described by the use of language (the study concentrates on verbal means, thus not quantifying other modalities). These could be either exophoric, i.e. only accessible from within the material setting of the concrete text, or endophoric, that is they are linked to other modalities present in the text as exemplified above. These endophoric uses may be identified with references such as see figure X. Although the presence of other modalities in the form of pictures, figures, diagrams etc. must be seen as a strong indication of a more ancillary use of language, there may also be cases where the pictures are not explicitly referred to in the text and must, therefore, be seen as ancillary to the verbal action.

Mode of discourse 

 75

3.1.2 Constitutive language use Constitutive language use is expected to manifest itself in the use of full, i.e. non-elliptical sentences and of the indicative mood, particularly in declarative sentences, rather than the imperative mood. And finally, lexical reference should be preferred over pronominal or deictic reference. This can be observed in a quantitative way in terms of a high lexical density. 3.2 Channel This subdimension is concerned with the physical conditions of the communication. It is relevant to the study of registers in that different channels both enable and constrain choices in meaning and their realisation in different ways (cf. Steiner 2004b). The phonic channel, i.e. transmission via sound waves, requires different linguistic expressions than a graphic setting. Halliday and Hasan (1989: 58) elaborate that process sharing has an impact on the interaction. In oral communication, the interactants can hear the same thing at the same time, for instance a yawn (Halliday and Hasan 1989: 58). If they are within sight of each other, they may also see facial expressions, body postures etc. that may have an impact on the speaker’s choice of linguistic means. Material actions of this kind are of no consequence if the interaction is transmitted via paper, i.e. in the graphic channel. In most of the cases, electronic environments share the characteristics of the graphic channel with graphic texts being provided in an electronic form. There are, however, interactions like chat room conversations with their real time transmission of written turns that are probably unique to the electronic channel, thus making a third option ‘electronic’ plausible. While we can name criteria for assigning texts to one of the three options, in most cases the assignment should be possible by mere inspection of the texts without needing to go into a detailed analysis of the text. The registers chosen for the present study (see section 1.1 of chapter 5) can all be assigned to the graphic channel, even if collected from the internet. 3.3 Medium As Hasan in Halliday and Hasan (1989: 58–59) convincingly points out, the subdimension channel does not cover the spoken-written distinction. This aspect has to be investigated separately under the heading ‘medium’ as the two cover different – though related – aspects. This becomes obvious when we look at cross-classifications: A text may be transmitted through the phonic channel but still bear more characteristics of the written medium. Although transmitted by

76 

 Indicators of register classification

sound waves, the text may thus be produced as if written. SPEECH, the register in the CroCo corpus representing prepared speeches, is a case in point. The speeches belonging to this register were written to be spoken. They were originally transmitted via the phonic channel but were prepared in writing and thus bear characteristics of the written medium. This is, for instance, reflected in a relatively high lexical density (cf. Vela et al. 2007). Ure (1971) also points out that the time available to prepare the text is an additional aspect possibly impacting on the medium. Investigating the differences between speech and writing in an empirical inductive study, Biber (1988) comes to the conclusion that registers cannot be described solely on the basis of their relative position on one continuum. Therefore, Biber introduces a set of dimensions to describe register variation in more general terms in Biber (1995, see also Biber et al. 1999 as an exemplary approach of including spoken language in a grammar). The merits and limitations of this approach are discussed in chapter 2. It can reasonably be argued that a corpus consisting exclusively of written texts does not require an extensive investigation of medium. However, we can still expect some variation between the registers in the CroCo corpus and therefore include this parameter in our analysis. As with the cline of language roles, medium should also be viewed as a continuum, with different registers situated at different points along the spoken-written continuum. Lexical density Ure (1971: 445) introduces the concept of lexical density as a measure to quantify the difference between spoken and written medium.40 This ratio expresses the relation between lexical words and the total number of running words in a text. She calls the spoken end of the continuum “language-in-action” and shows that this kind of language is characterised by a lower lexical density than written language. Halliday (1994a, 2001) takes up Ure’s concept and argues that her calculation is too prone to language-specific peculiarities of function words since, together with lexical words, they account for all the running words. Halliday therefore proposes a ratio of lexical words in relation to the number of clauses (2001: 183, see also Ventola 1996). While this calculation is certainly better suited to the contrastive comparison attempted here, the less accurate computation is used here because it is available for all CroCo registers. In order to counteract any inaccuracies arising from the calculation of lexical density, its counterpart grammatical intricacy is included as well and is described in what follows. 40 Ure’s recommendation to take into account other factors that have an impact on the “overall language patterning” (1971: 446) is followed in the present study by the wide range of subdimensions and indicators investigated.

Mode of discourse 

 77

Grammatical intricacy In Halliday’s view (e.g. 1994a, 2001), the use of lexical density as an indicator of written language has to be complemented by grammatical intricacy as an indicator of spoken language. He argues that written language as characterised by high lexical density is organized around the nominal group; and this – since the nominal group construes reality as entities (objects, including institutional and abstract objects, and their quantities, qualities and types) – creates a world of things and structures, discontinuous, rigid, and determinate. (Halliday 2001: 186)

Spoken language, on the other hand, is organised around the clause thus creating “a world of movement and flux, or rather a world that is moving and flowing, continuous, elastic, and indeterminate” (2001: 186). Consequently, spoken language can be expected to spread lexical information over clauses instead of packaging it into groups and phrases, or, in the more theory-neutral term used in the CroCo project, into chunks. Registers located near the spoken end of the continuum should therefore exhibit a higher number of clauses per sentence – and, more generally, a lower number of tokens per clause.41 This latter generalisation accommodates Halliday’s (1994a: 65) statement that “the AVERAGE number of clauses per clause complex will [not] be greater in spoken language, because there may also be a tendency towards very short ones, especially in dialogue”. Written registers, by contrast, are expected to contain a higher ratio of tokens per chunk as a reflection of the more densely packed nominal groups/ phrases. These densely packed chunks are then simply linked resulting in a less intricate clause structure. We adopt Steiner’s term ‘grammatical density’ as discussed in section 2 of chapter 2 for this opposite pole of the continuum of grammatical intricacy. The term ‘density’ refers to the contents of the chunks, but, as pointed out by Halliday and Martin (1993), it goes hand in hand with a simpler grammatical structure at sentence level and is observable as a low number of intermediate units per sentence. Thematic structure Very generally speaking, theme structure is the primary grammatical system serving the realisation of textual characteristics. As Halliday (2001: 186) writes, theme is backgrounded “as the point of departure for what is coming next”. 41 In contrast to Biber (1988: 229), we thus do not break down Halliday’s argumentation to subordinate structures (i.e. hypotactic ones in Halliday’s terminology), but cover all types of clauses regardless of whether they are in subordinate/hypotactic or coordinate/paratactic relationships.

78 

 Indicators of register classification

From clause to clause, the meaning conveyed by the text progresses from the theme as the local point of departure to the rheme etc. Besides the question, whether and how the preceding rheme is resumed and becomes the theme of the following clause, we are interested in the function of the thematic element as this may further our understanding of the focus of the text on one of the metafunctions (as manifested in the assignment of the respective grammatical function in terms of ideational, interpersonal or textual metafunction). As regards the spoken-written continuum, we would expect registers tending towards the spoken end of the range to contain more verbal elements (both finite and non-finite) in theme position, suggesting a more interpersonal orientation that is regarded as typical of spoken discourse. Registers more prone to the written medium are expected to make more use of participant roles, i.e. subject, and the various types of objects as well as adverbials in thematic position, reflecting an orientation towards experiential meaning. Reference Reference is of particular interest to the analysis of the subdimension medium since spoken registers are generally seen as relying more on pronominal reference than written registers do. Biber (1988: 225–226) offers an overview of studies discussing the role of various pronouns in spoken and written registers. In particular, the distinction between exophoric and endophoric reference (Halliday and Hasan 1976) should shed light on the spoken-written distinction. We can also assume that – beyond this – syn-semantic, i.e. pronominal, versus full lexical reference is an indicator of this distinction with pronominal reference characterising spoken, situation-dependent registers and lexical reference being more likely to feature in factual, written registers (cf. Hansen-Schirra, Neumann and Steiner 2007). The overall number of pronouns should therefore be higher in spoken registers.

4 Summary This chapter has given an overview of subdimensions of the three register variables. Only those subdimensions were included in the derivation of indicators that are sufficiently well described and enable the quantification of indicators. The indicators were then derived, again bearing in mind their suitability for quantitative analyses. The range of analyses and interpretations necessary for achieving a comprehensive overview of register characteristics present in a 1 million word corpus demands a confinement in terms of depth of ­analysis.

Summary 

 79

Several of the indicators can be investigated in much more detail. This is, however, not feasible in a study of this scope. The analyses of the indicators thus derived (see chapters 6 to 9) will identify potential areas for further in-depth research. Moreover, the discussion in this chapter showed that several indicators can be “reused” for the analysis of different subdimensions. It does not follow from this that the indicators are necessarily interpreted in the same way. Since they provide insights into different subdimensions, these recurring indicators will be interpreted in different ways. With the indicators derived, the theoretical foundations of the study are in place. We can now proceed to describe in all due detail the methodology used to investigate register variation in the two languages as well as in originals and translations.

Part 2: Empirical analysis

Chapter 5 Research design This chapter presents the methodology used to investigate cross-linguistic register variation. The centre of the methodology is the corpus and its linguistic analysis. Its design and the linguistic information included in the electronic resource are described in the first section. The second section is concerned with how to retrieve information from the corpus as well as with the statistical techniques used to interpret the obtained quantitative information. The third section explains the constellations of the corpus contrasted to gain insight into various aspects of register variation. These sections in combination with the previous two chapters lay the foundations for the formulation of three hypotheses in the last section of this chapter that will then be tested in the empirical analysis.

1 The corpus The research presented here was carried out in the framework of the CroCo Project,42 which was concerned with the investigation of typical linguistic properties of translations as compared to originals in the language pair English-German. The CroCo corpus was created for this purpose, meaning that decisions relating to its design and linguistic enrichment were made considering this original research question. While there is a good deal of overlap between the requirements of the research question underlying the CroCo Project and the research question of the present study, there are also aspects of the corpus design that are not relevant to the investigation of register variation or that may even impair it. The following two sections will introduce the general design and enrichment of the CroCo corpus, including aspects that are of lesser importance to this study. 1.1 Design The corpus design was guided by a number of criteria to ensure a corpus size sufficient to allow generalisable statements, a balanced composition as well as comparability across languages (see Neumann and Hansen-Schirra 2005). Since 42 http://fr46.uni-saarland.de/croco/index_en.html, last visited 12 April 2013.

84 

 Research design

register variation is assumed to be one of the factors having an impact on the specific characteristics of translations, a range of registers was included in the corpus. Furthermore, texts were included to cover both translation directions in order to distinguish phenomena that are related to the translation direction from phenomena that apply irrespective of the translation direction and may thus point to more general characteristics of translated texts. These design criteria had a strong impact on the selection of registers for the corpus. Only those registers were considered for inclusion in the corpus that are translated into the two languages at all, that is where translations into both directions could be collected. This entailed, for instance, the exclusion of the register of court decisions, an analysis of which is likely to have revealed interesting cross-linguistic register differences (see Maksymski 2008). However, we were only able to identify translations into English belonging to this register. In order to satisfy the design criterion of balance, the registers were selected with a view to covering a certain variation in terms of context of situation. The corpus meets these requirements by including texts from eight different registers that are not only relevant to translation, but that are in fact translated into both English and German. Additionally, these registers were assumed to foreground different registerial characteristics in a preliminary analysis (cf. Neumann and Hansen-Schirra 2005). The registers can be briefly characterised as follows (a more detailed characterisation can be found in chapter 2 of Hansen-Schirra, Neumann and Steiner 2012).43 ESSAY consists of political essays on a range of political topics. The official author of these texts is typically a well known politician, although they may have been written by members of the politician’s staff. The intended audience is the general public, and, in some cases, the texts are written for foreign public audiences. FICTION contains samples from contemporary novels of which translations into the respective other language exist. One might argue that the character of literary texts of creating an imaginary world and situating the (linguistic) action within this world prohibits analysing these texts as a register in its own right. However, like any other register, literary texts also reflect a given context of situation including authors/writers and readers. As Halliday and Hasan (1989: 40) argue, “we are never selecting with complete freedom from all the resources of our linguistic system”. This should also apply to literary texts, which despite their originality, reflect a situation and can be said to be related to other literary texts in terms of intertextuality. Since high-quality literary texts are of particular 43 This characterisation represents a first overview of the registers. Chapters 6–9 will give a detailed analysis of all the features characterising these registers.

The corpus  

 85

interest for linguistic analysis because they frequently display a wider range of lexico-grammatical means to express the intended meaning, they are included in this study as well. INSTR consists of instruction manuals for a variety of appliances and products. Since both the CroCo Project and the present study are mainly interested in verbal aspects of texts rather than other modalities, only texts which mainly rely on verbal means to introduce the given product were selected. Consequently, the texts in the corpus represent a specific subset of instruction manuals in terms of mode of discourse. POPSCI contains texts introducing scientific topics to a wider public (socalled popularisations). These are mainly articles from magazines, to which have been added two extracts from books translated from English to German.44 As German popular scientific articles do not seem to be translated into English at all, this part of the corpus consists of extracts from German book-length popular scientific writings and their matching translations into English. The different publication modes may cause some differences between the comparable registers in both languages. However, it can be assumed that the general characteristics of the register are similar. SHARE covers letters from or in the name of the CEO of various companies to their shareholders. These texts inform the shareholders on the performance of the company in the reporting period. SPEECH contains manuscripts of political speeches drafted in advance and delivered by senior government officials. The texts in the register TOU are promotional texts published by the respective tourism organisations of a city or region. The texts advertise a certain place of interest and provide a range of information. WEB contains a collection of websites. The decision to include this register was originally guided by the idea to cover youth language as a specific sociolect potentially influenced by the electronic mode of presentation. However, it turned out to be very difficult to find websites that are actually translated instead of having a localised (different) version in the other language. Therefore, the selection process had to be extended to include all kinds of websites that contain translations. As a consequence, this subcorpus displays a wide range of variation covering different web genres (Santini 2006). Each of the registers is sampled in four versions resulting in four subcorpora (see Figure 2): English originals (EO), German translations (GTrans), German originals (GO) and English translations (ETrans). 44 The starting point for collecting this subcorpus was a corpus originally collected by Monika Doherty (see Doherty 1998, 2006), however, only three of the texts from this corpus are part of the CroCo corpus.

86 

 Research design

As to targeted size, the project generally followed Biber’s (1990, 1993) calculations, which show that a well-balanced corpus consisting of at least 10 texts and 2,000 words per text45 in a given register sufficiently reflects the actual distribution of linguistic features in this register. Thus, each register in each of the CroCo subcorpora contains at least 10 texts. The initially planned text length of 3,125 words was not reached in every register because the texts in some registers are typically shorter. In these cases, more texts were collected until the overall register size of 31,250 words was reached. Depending on the text lengths typical of the different registers, full texts or samples were collected. Where samples were used, these were selected by drawing random numbers to identify the start page of the sample. The actual size of the overall corpus is 1,181,435 tokens. Table 1 gives the exact size, both in terms of texts and tokens of the subcorpora and the registers. The fact that the actual corpus size is larger than intended is mainly due to the inclusion of matching translations. These cannot be abridged just to reach the targeted size. Interestingly, the English translations contribute most to the increased number of tokens. Table 1. Overall size of the CroCo corpus EO

ESSAY FICTION INSTR POPSCI SHARE SPEECH TOU WEB Total

ETrans

GTrans

  GO

texts

tokens

texts

tokens

texts

tokens

texts

tokens

29 10 10 11 13 14 11 12

34,998 36,996 36,167 35,148 35,824 35,062 35,907 36,119

23 10 14 10 11 18 22 13

42,036 40,037 39,663 37,878 39,511 39,766 43,677 39,657

29 10 10 11 13 14 11 12

35,345 37,652 35,820 33,603 36,375 35,909 34,152 35,705

23 10 14 10 11 18 22 13

35,668 36,778 36,880 36,177 35,235 35,337 36,574 35,779

110

286,221

121

322,225

110

284,561

121

288,428

An interesting point pertaining to corpus size requires our attention: The probabilities to which Halliday (2005) refers (see section 2.1 of chapter 3) can only be yielded on the basis of very large corpora, larger than the one available for the present study. The precise meaning of very large is a matter of dispute in corpus linguistics (see, for instance, Tognini-Bonelli 2001: 59). From the point of view 45 Biber uses the split-half method to show that 1,000-word samples already contain enough spread, consequently proving that 2,000-word samples are large enough.

The corpus  

 87

of statistics, the statistical power of a test may be increased by using a larger sample size, making the acceptance of the alternative hypothesis more likely (Pospeschill 2006: 177), an effect which may lead to undesirable distortions of the analysis (see also Kilgarriff 2001). In Pospeschill’s view, an optimal sample size has been reached once an unambiguous decision on the hypothesis is possible. It follows that a larger sample is not required. From the point of view of corpus linguists working within the corpus-driven paradigm, our corpus size will appear too small. It is certainly too small to allow statements to be made about the probabilities of grammatical phenomena as advocated by Halliday (2005). It does, however, contribute to the study of variation, register variation in particular, which is only accessible to quantitative corpus-based research. Any statements within this area of research not based on the examination of an acceptable amount of textual instances must remain a matter of speculation. In compiling a corpus for the study of registers, researchers are faced with a dilemma that is a matter of continuing debate: The selection of the texts in the CroCo corpus is based on the assumption that they belong to the same register (see the introduction to the registers above), yet only an in-depth analysis of their defining features can determine whether they actually do. CroCo addresses this aspect by including a cursory register characterisation in the metadata (see section 1.2 of this chapter). In addition to this, statistical processing of the corpus findings may help identify outliers, it does not, however, resolve the problem of circularity inherent in this type of study. The core corpus is complemented by two small reference corpora in both languages (ER and GR), each containing 2,000-word samples of original texts from 17 different registers. The reference corpora are designed to level out register-specific peculiarities by representing a wide range of registerial spread. Although 17 different registers cannot reflect the complete range of register variation in highly differentiated languages such as English or German, they can at least serve as a basis of comparison for determining specific characteristics of individual registers. The reference corpora are enlarged versions of the ones used in Neumann (2003): Size was doubled and two registers (court decisions and travel guide books) were added. The reference corpora consist of between two and ten extracts from different texts in the 17 registers. The structure and the major part of the texts were originally taken from the FLOB Corpus of British English (Hundt et al. 1998).46 It was modified to reduce the bias towards fictional texts. This structure was also used as a template for the German reference corpus (Neumann 2003: 91–94). Some of the registers in the reference corpora are also 46 The Freiburg-LOB Corpus is a replication of the original LOB Corpus containing texts published in 1961 (Johansson et al. 1978) with texts from the year 1991.

88 

 Research design

part of the core CroCo corpus. This seems only natural, the CroCo registers being part of language use in the respective language. In total, the English reference corpus contains 42,332 tokens in 107 texts, whereas the German reference corpus consists of 42,122 tokens in 110 texts. Figure 2 gives a schematic overview of the corpus design. English texts Registercontrolled Corpus EO

Reference Corpus ER

Translation Corpus GTrans

Translation Corpus ETrans

Registercontrolled Corpus GO

Reference Corpus GR German texts Figure 2. Overview of the CroCo corpus design

1.2 Enrichment The CroCo corpus contains several annotation and alignment layers on word, chunk,47 clause and sentence level in addition to metadata about each text in the corpus. The annotation of metadata is based on the TEI standard48 and includes a specification of the publication and translation as well as a shallow register analysis (Klinger et al. 2006). The annotation on word level comprises tokenisation, part-of-speech tagging and morphology. Tokenisation and part-of-speech annotation were performed automatically using Thorsten Brants’ statistical TnT tagger (Brants 2000). The 47 ‘Chunk’ is the cover term used in the CroCo Project for intermediate grammatical units. It covers both the formal interpretation in terms of groups/phrases as well as the functional use in terms of subject, object, predicate etc. 48 http://www.tei-c.org/, last visited 12 April 2013.

The corpus  

 89

tag sets used with the tool are Susanne (Sampson 1995) for English and STTS (­Schiller et al. 1999) for German. Morphology including lemmatisation was annotated automatically with MPRO (Maas 1998), a rule-based tool operating on both languages. On chunk level, the annotation was carried out manually with the help of MMAX2 (Müller and Strube 2006), a tool for manually creating linguistic annotation. The annotation comprises the formal classification in terms of phrase types as well as the assignment of grammatical functions to units on the highest level within the sentence.49 The guidelines for the annotation use the same labels for both languages but language-specific descriptions. They are based on Quirk et al. (1985) for the English language and Helbig and Buscha (2001), the Duden grammar (Eisenberg et al. 1998) and the Wahrig grammar of German (Götze and Hess-Lüttich 1999). Wherever possible, the guidelines for the German part are based on the annotation guidelines for the German treebank project TIGER.50 Due to limited resources, the manual annotation was only carried out once per text, thus reducing the reliability of the data (see section 3 of chapter 3). In addition to the annotation, the texts in all eight registers are also aligned on word, phrase, clause and sentence levels. The present study does not make use of the (whole) range of alignment layers available since its focus is on register variation and thus on the examination of whole texts/text samples. All individual annotations and alignments are converted to the same XML format (‘CroCoXML’) and stored in individual files, making it possible to retrieve either each type of linguistic information separately or in combination with other files (Hansen-Schirra, Neumann and Vela 2006). Chapter 3 of Hansen-Schirra, Neumann and Steiner (2012) provides a detailed introduction to all matters concerning enrichment, representation and exploitation of the CroCo corpus. The central principle guiding the compilation of the CroCo corpus was to provide a resource suitable for a wide range of research applications (see also Hansen-Schirra, Neumann and Steiner 2007). In order to meet this goal, linguistic information has been added to the texts in the corpus in such a way as to minimise any bias towards a particular linguistic theory. However, the derivation of indicators by queries into the corpus and their interpretation in terms of the theory-guided research question draws on the theoretical framework provided by systemic functional linguistics. For the present study this means that theoryinternal concepts like the thematic structure of texts are queried on the basis of 49 At the time of writing this volume the annotation was not yet available for all registers and the reference corpora. 50 http://www.ims.uni-stuttgart.de/forschung/ressourcen/korpora/TIGERCorpus/annotation/index.html, last visited 12 April 2013.

90 

 Research design

the available information, i.e. grammatical functions in combination with word order. An additional benefit of this approach is that it facilitates the quantitative investigation of the corpus in two ways: First, automatic tools providing SFL annotation are simply not available, therefore tools that provide a theory-neutral annotation were selected. Secondly, linking abstract concepts and observable indicators by means of operational hypotheses appears particularly suited for quantitative research (see section 2.2 of chapter 3). This link can be established more flexibly when not dependent on fixed categories such as, for instance, the options in systemic networks. The theory-neutral approach to annotation adopted by the CroCo Project makes the CroCo corpus a valuable resource for other areas of research within different theoretical frameworks. The next section will explain in more detail how the linguistic information incorporated into the corpus that may combine different annotations depending on the indicators is extracted from the annotated corpus. It will also present the statistical techniques used to assess the linguistic information extracted from the corpus and describe the main principles underlying the interpretation of the data.

2 Exploiting the corpus 2.1 Query tools The enrichment of a collection of texts by a wide range of linguistic information, as was done in the CroCo Project, is quite costly. Nevertheless, retrieving this information from the corpus in a meaningful way is not a trivial exercise, especially if the task includes the combination of several annotation layers. The present study employs various querying methods including simple concordancing, advanced queries using the IMS Corpus Workbench (Christ 1994) and various PERL scripts. One reason for the use of such a wide range of query techniques is the fact that the CroCo resource was not finalised by the time of writing this volume, necessitating a number of workarounds. Processing concordances is often the first step in exploring a corpus. It does not require annotation and can provide valuable information, particularly about lexical characteristics of a text collection. This approach is adopted by the present study: Mike Scott’s WordSmith tools (Scott 2004) are used to extract frequency lists from each subcorpus. However, since concordances obtained on the basis of raw data contain all the morphological variants of given tokens, concordancing is conducted on the basis of the lemmatised corpus. Furthermore, a major area of interest is the field of lexical or content words as well as lexical

Exploiting the corpus 

 91

verbs. Therefore, a script selecting all lexical words and verbs respectively on the basis of the part-of-speech tags51 is run prior to concordancing. The keywordin-context option of WordSmith is used to identify relational uses of the third person singular present forms of the verbs to be/sein (see the sections on experiential domain in chapters 6 to 9). Most of the queries were run on the part-of-speech annotation as this type of annotation allows retrieving a considerably range of grammatical structures. The IMS Corpus Workbench (CWB, Christ 1994) is an efficient tool for exploiting the part-of-speech annotation. The following example of querying mood illustrates the way systemic functional categories can be identified without a dedicated annotation. The system network displayed in Figure 3 contains the basic mood options in both English (Halliday and Matthiessen 2004: 135) and German (Teich 2003: 102).

sentence

mood

indicative imperative

indicative type

declarative interrogative

interrogative type

wh yes–no

Figure 3. System network of the basic options for mood

Imperative and interrogative mood can be queried quite reliably in both languages with the help of a combination of positional and part-of-speech information. The number of declaratives can then be obtained by subtracting the imperatives and interrogatives from the total number of sentences. Figure 4 shows some query results obtained from searches for imperatives, which look for base forms of verbs following a punctuation mark.52 Only one hit is incorrect (corpus position 26001 is an interrogative). Mapping frequencies of linguistic features onto the text ID in which they occur is a prerequisite to the use of statistical techniques which operate on the mean (see section 2.2 of this chapter). This in turn helps us to enhance our understanding of registers. Where necessary, the CWB queries are replaced by queries using scripts that map the linguistic information onto the respective text ID. Scripts are also used to implement more complex queries that combine several annotation layers, for instance to identify grammatical functions in thematic position, where again the use of punctuation in combination with the layer of grammatical functions serves as an indicator (see the sections on medium in chapters 6–9). 51 Manual post-editing is carried out to correct systematic errors. 52 The first line in the figure contains the actual query.

92 

 Research design

Figure 4. Example of a query for imperative mood

After determining the frequencies of the various indicators, a processing step is necessary to prepare the frequency counts for statistical analysis. Baroni and Evert (2009: 796) claim that the unit of measurement (e.g. a word token) typically does not correspond to the unit of sampling (e.g. a given number of tokens in a text). Even if the texts in the corpus represent a random sample of the basic population, a degree of non-randomness of the frequency counts will ensue from this discrepancy, which may lead to misinterpretation of the observed data. The authors explain that this problem can be solved by adjusting the unit of measurement, i.e. by calculating the proportion of the instances of a given feature in the text to the number of all units in the text. Baroni and Evert (2009) give the example of the ratio of passive sentences to the total number of sentences in the text. Using proportions rather than absolute frequencies has a convenient side effect: Frequencies merely represent nominal yes/no decisions stating whether a given feature obtains for a given unit or not (in Baroni and Evert’s passive example this would be ‘passive’ versus ‘non-passive’). Decisions of this type are assigned to the nominal level of measurement and therefore only permit non-parametric statistical techniques. Proportions, however, represent what Baroni and Evert (2009: 798) call “real-numbered values that can in principle assume any value between 0 and 1”. Values of this type fulfil the requirements of the interval level of measurement and thus allow the use of parametric statistical techniques based

Exploiting the corpus 

 93

on the mean (see below). Therefore, all frequencies of occurrence of the various features in this study are converted to proportions. The following section will provide a more detailed overview of the statistical techniques used in this study.

2.2 Statistics This study uses statistical techniques to summarise the quantitative findings with the help of descriptive statistics and to determine whether they reflect a systematic property of the linguistic data with the help of significance testing as part of inferential statistics. The present section does not claim to provide an exhaustive overview of statistics in (corpus) linguistics; it only addresses aspects that are relevant to the present study. Furthermore, all the calculations and formulae of the statistical techniques presented are omitted. The reader is referred to reference books on statistics such as Black (1999), Pospeschill (2006), and books specifically aimed at linguists by Oakes (1998), Butler (1985), Gries (2008) etc.

2.2.1 Descriptive statistics A major reason for approaching linguistic analysis with quantitative methods is because the researcher is interested in findings about groups of texts. This is particularly the case for studies of (register) variation. Statistical methods and tools facilitate the identification of quantifiable characteristics of and differences between data sets (Black 1999: 329). Descriptive statistics help provide an overview of the frequency distributions of features in the corpus by organising, summarising and displaying them. The organisation of data refers to the representation of the results of a corpus analysis in the form of matrices. The present study uses the SPSS software53 to process the quantitative data and evaluate it statistically. Summarisation of the data consists of measures of central tendency and measures of statistical dispersion. The three measures of central tendency, the mode, median and mean (or arithmetic mean) are used to identify group characteristics and describe the group of values as a whole. The mode, used for nominal data, refers to the value which occurs most frequently in a distribution. The median refers to the middle of a distribution, it is the number dividing the group of all values in two. The advantage of the median (and the mode) in comparison to the mean is that it is 53 http://www-01.ibm.com/software/analytics/spss/, last visited 12 April 2013.

94 

 Research design

less sensitive to outliers (which may distort the mean). Since the median does not include information on the distances between the values, it is suited for ordinal scales. The (arithmetic) mean is the addition of all individual values divided by the number of values. It requires numerical data at the interval level because it takes into account the distance between the different values. Additionally, the comparison of data sets that contain different numbers of individual cases requires weighted means. These measures of central tendency help to balance out any measurement errors (see section 2 of chapter 3), they summarise the individual values in one group and thus allow statistical tests to be carried out to compare these group values. Measures of central tendency are complemented by measures of statistical dispersion. These include (interquartile) range, variance and standard deviation. Range denotes the interval containing all values, i.e. the difference between the largest and the smallest value. It is a rather weak measure since it only includes these two values and does not give any indication of the variation between all values. Variance is a measure of spread that captures the distance of every value from the arithmetic mean (for the exact calculation see e.g. Pospeschill 2006: 80–85). The standard deviation standardises the difference from the arithmetic mean to the same units applying to the data. A high standard deviation indicates that the observations vary widely. In linguistic analyses, the standard deviation is an important measure of the variability in a corpus. Furthermore, a well balanced data set displaying a high standard deviation for a given feature implies that the corpus contains much variation. This can be of interest when comparing originals and translations under the assumption that translations tend to level out (Baker 1996) the variation typical of non-translated texts in the same language. The results are displayed in two ways. Box plots (see Figure 5) illustrate five descriptive key figures: the smallest observation, the lower quartile (Q1), the median (see below), the upper quartile (Q3) and the largest observation. Quartiles refer to the three values that divide a distribution into four equal parts, i.e. 25 percent of the values (the lower quartile, or Q1), 50 percent of the values (Q2) and 75 percent of the values (the upper quartile, Q3). The difference between the lower and the upper quartile is called the interquartile range. Any outliers, i.e. values outside the main cluster of values in the data set, are marked as such in the box plot. The box plot may be complemented by a mean chart giving an overview of the means (see below) in different data sets. It should be noted that the lines linking the individual values do not have any meaning and merely serve to illustrate the distance between the means of each data set. Descriptive statistics give important information about the characteristics of a data set or a corpus. They do not, however, help us interpret and determine, for

Exploiting the corpus 

 95

instance, whether two data sets represent different populations. This interpretation process, which is at the core of research into linguistic variation, requires the use of significance testing.

outlier

⎧ ⎨ interquartile range ⎩

largest observation upper quartile median lower quartile smallest observation register

Figure 5. Example of a box plot for personal pronouns per all tokens in English originals (EO)

2.2.2 Significance testing Significance testing concerns the verification of assumptions relating to correlations, differences and changes of given features in certain data sets. Beyond examining these interrelations between features, hypothesis testing research attempts to explain and, ultimately, predict effects. The statistical techniques used in this branch of research are therefore subsumed under the heading of inferential statistics. The present study concentrates on testing and explaining the differences between the different data sets. In empirical research this is done with the help of significance tests (for the logic of significance testing see section 6.3 of Butler 1985, Baroni and Evert 2009). Since the present study intends to test existing hypotheses rather than generate new hypotheses (see section 2 of chapter 3), it does not employ factor analysis as used by Biber (1988, 1989, 1995) or other inductive methods but rather techniques developed for hypothesis testing such as the t-test and analysis of variance. Compared to non-parametric techniques, i.e. techniques operating on nominal or ordinal variables like the

96 

 Research design

chi-square test, these two parametric tests provide more information as will become clear from the following description. Both techniques are used in connection with interval data and may only be used under certain defined conditions. They require a normal distribution of the data and a quality that permits the use of the mean and standard deviation as appropriate measures of central tendency and dispersion (see Oakes 1998: 11; on the issue of random sampling see below). Normal distribution is assumed to apply to many naturally occurring events, with most occurrences centring around the mean and fewer scoring extreme values. This distribution of values can be represented as a bell curve with the mean as the middle value. In a perfectly normal distribution, mean, median and mode (see above) are equal. The normal distribution is tested by default for all data sets in this study with the help of the Kolmogorov-Smirnov test (Bühl 2006: 332–333). The t-test is a technique for comparing two data sets, more precisely, it tests the difference between two means. Two types of t-tests are available for two different research designs, the between-groups design and the repeated-measures design. In the latter case the same data set (or group of subjects) is tested repeatedly. Although it is an interesting idea to conceptualise translations as a different state of the originals and thus use the t-test for repeated measures, this is rather far-fetched and will not be pursued any further in this study. The t-test for a between-groups design, by contrast, seems appropriate for the present study to test whether the differences between the individual register data sets and the reference corpora were produced at random or can be interpreted as systematic. This test highlights the importance of the reference corpora: Only by comparing the register-controlled and the register-neutral subcorpora can register features be reliably identified. The t-test is restricted to the comparison of two groups. The comparison of all the registers represented in the corpus requires a different approach. Although we could repeat the t-test for all registers in the corpus, this would lead to an accumulation of the Type I error (see Black 1999: 380ff.) and to a decrease of the test power (Rasch et al. 2006: 5). The analysis of variance, abbreviated ANOVA, is based on the t-test, but compares several groups or data sets. It tests whether the individual cases come from the same basic population or whether there are significant differences between them. The independent variable of the one-way ANOVA must be nominal (e.g. texts in different registers), while the dependent variable must be interval and normally distributed. Finally, the variance of the dependent variable of all samples must be equal.54 The ANOVA computes two 54 This is taken into account by SPSS. The software offers a default calculation of the test for homoscedasticity that includes balancing of unequal variances.

Exploiting the corpus 

 97

indices: the within-groups and the between-groups estimate of variance. If both indices are similar, the null hypothesis must be adopted, stating that all groups correspond to the same mean value, i.e. come from the same group. If the between-groups value is significantly higher, this indicates that the mean values of the groups differ more markedly than would be expected from a random distribution. The ANOVA does not give any indication of the direction of the tendency of central variance. In order to learn more about the relationship between the individual results in the ANOVA beyond the information that the items in the data set come from different groups, it may be supplemented by post-hoc tests to identify patterns in subgroups of the sample. In the case of the present study this could mean identifying one or several registers as forming one subgroup significantly distinct from the other registers, i.e. other subgroups. Depending on the similarity of the values in the data set, there may also be an overlap between subgroups. Since the analysis of variance is based on strong assumptions regarding normal distribution etc., it is important not to further burden the interpretation by choosing post-hoc tests that require additional strong assumptions. A test that operates on rather conservative calculations is the Tukey HSD test. It compares all possible pairs of means and calculates the smallest significant difference between the means taking into account the cumulated Type I error level, thus maintaining the test power of the analysis of variance (Rasch et al. 2006: 46–48). Where an analysis of variance is employed in the present study, this test is used for post-hoc tests. In general, statistical techniques are essential for quantitative studies. Nevertheless, the previous discussion already suggested that a number of problems may arise, for instance, from the strong assumptions made by some techniques. The following section will therefore address some constraints of statistical techniques.

2.2.3 Limitations of statistical techniques The limitations of working with statistical techniques broadly fall into two categories. First, there are issues relating to statistics in general. Secondly, there are particular issues concerned with the application of statistics to linguistic studies. The two areas will be addressed in turn in the present section. A general issue with statistical tests is related to the preconditions required in terms of level of measurement and assumptions made by the respective technique. Although it helps to structure the data and select an appropriate statistical test, the assignment of a given data set to a level of measurement is not always straightforward and must be seen as a matter of interpretation in many

98 

 Research design

cases (see the brief discussion in section 3.3 of chapter 2). As to the assumptions underlying many tests such as, for instance, particularly the requirement of using random sampled data poses a problem in working with corpus data. It is doubtful whether a flawless random sample can be achieved in social research with many representatives of the basic population being outside the reach of social researchers for interviews, e.g. patients in hospitals or senior managers. This is even more challenging in linguistic studies, where determining the basic population is a problem in itself. It is possible to draw a random sample of all books in print on the basis of the publications listed, for instance, in the British National Bibliography (cf. McEnery and Wilson 2001: 79, Neumann 2003: 75). There are, however, many registers which are not listed in this way, making it impossible to determine their population – not to mention the fact that they are, as a consequence, not accessible for linguistic research (McEnery and Wilson 2001: 79). At any rate, the requirements for the corpus design used in this study (see section 1.1 of this chapter) make it impossible to draw a proper random sample. Running the significance tests described above on the CroCo corpus is therefore problematic. The findings discussed in chapters 6 to 9 must be seen in this light. Another issue in linguistic research is the definition of the unit of research. In social sciences, the unit of research is typically the human being. In linguistics, the identification of the unit of research and consequently the identification of what counts as a case in statistics is less straightforward. Do we analyse words, phrases, clauses, sentences, texts or even whole corpora? The choice of unit has an impact on the statistical techniques we can employ (Baroni and Evert 2009) irrespective of the linguistic claims we make on the basis of this unit. The next problem is known as the scarce data phenomenon. A range of linguistic features are scattered widely across large quantities of texts, making them difficult to identify. This may lead to a distortion when statistical techniques are used which are not suited to the analysis of very small quantities of data. Typically, this is handled by enlarging the corpus,55 however enlarging the corpus is not an option for the present corpus; this is mainly due to the amount of manual annotation involved in creating the resource. An alternative approach to handling scarce data would be the adaptation of statistical techniques to this specific problem. To our knowledge, techniques specifically adapted to the linguistic requirements are not currently available.56 Therefore, the frequencies of 55 See also Oesterreicher (2001: 1569) for a critical view on the usefulness of enlarging corpora. 56 Some research has been done in this direction (see Frötscher and Özçetin 2007) in the framework of the Collaborative Research Centre 538 ‘Multilingualism’ at Hamburg University

Exploiting the corpus 

 99

some types of features used in the present study (see chapters 6 to 9) are tested using the tests described above. The results of the tests should be seen with caution. Needless to say that statistical tests are but a means to facilitate the interpretation of the data, be it to a very comprehensive extent as in Biber’s (1988, 1989, 1995) hypothesis-generating approach, be it to a very limited extent as in the present study where the tests merely help determine a threshold for the interpretation of the frequencies found in the corpus. In any case, the appropriate interpretation of the tests and the data remains the crucial step. The sheer volume of the features to be interpreted means that it is not possible to interpret every feature analysed in this study in every detail. The various discussions of individual indicators must therefore be seen as exemplifications of viable interpretations. This holds for instance for the more complex interpretations relating to goal orientation and social distance where a whole set of indicators is interpreted in terms of the respective subdimensions. It also applies to the specific method of comparing only the intralingual magnitude of difference between a register-controlled corpus and the reference corpus across languages in the contrastive comparison (see section 4 of chapter 3). When interpreting any variation within a given register, we must keep in mind that the findings depend heavily on the choice of texts included in the corpus. If the texts are chosen from a very closely related field (e.g. comments made by senior politicians or government members as in ESSAY and SPEECH), it is likely that there will be less variation, while texts selected from less similar sources (as is the case for the websites collected in WEB) will display more variation within the given register. More generally, any study that contains deductive elements runs a certain risk of circularity in the interpretation of the data. This is certainly true for the present study. One might claim that a certain degree of circularity is inherent in most corpus-based studies. Although the intrinsic danger of circularity does not make corpus research a futile endeavour, it does limit the explanatory power of the respective study. The present study is therefore limited to describing data, helping to confirming or refute assumptions about registers that have not yet been tested on a quantitative scale. Biber’s work on text types (1989) circumvents this complication by defining categories exclusively on the basis of processing co-occurrences of linguistic features without any prior assumptions. This is a very interesting approach in terms of evaluating deductive findings. However, it does not produce a helpful text classification since it creates artificial groupings of texts that do not necessarily reflect the context of situation. (http://www.uni-hamburg.de/sfb538/, last visited 12 April 2013).

100 

 Research design

On the basis of the information on the corpus design provided in section 1 of this chapter and its exploitation in this section, the next section will summarise which parts of the corpus are actually contrasted in this study before finally explicating the hypotheses that will be tested in chapters 6 to 9.

3 Corpus contrasts 3.1 Overview of the contrasts The design of the present study with its multitude of features and comparisons requires a condensed presentation of the findings wherever possible. The different comparisons require contrasting a number of constellations of various corpus parts. The bulk of the analyses (chapters 6 to 8) is concerned only with the subcorpora of original texts. First the focus will be on the intralingual comparison between the eight registers separately for English in chapter 6 and German in chapter 7. Chapter 8 then concentrates on cross-linguistic differences and commonalities within register pairs in the two languages. This prepares the ground for an assessment of the specific properties of translations as compared to the originals in both languages. Translations are consequently taken into consideration in chapter 9, which addresses the impact of translation including both the intralingual perspective by examining originals and translations in the same register and the cross-linguistic perspective by comparing originals and their matching translations in both translation directions. All combinations are summarised in Table 2. The intralingual comparisons of originals cover all eight registers in the corpus.57 The contrastive and translation-related comparisons are exemplified for the two registers FICTION and SHARE. In Steiner’s terms (2008: 3, see also chapter 5 of Hansen-Schirra, Neumann and Steiner 2012), the present study focuses on the following types of contrasts: the intralingual comparison of registers in both languages represents Steiner’s contrast C2.2, and the interlingual comparison within a given register between the two languages relates to contrast C2.1. Finally, the comparison of originals and translations concerns C3 where originals and translations within one language are compared as well as C4.1 where registers are compared by contrasting aligned texts. 57 Occasionally, the analysis of a given indicator may be still exploratory and therefore only carried out for two registers.

Corpus contrasts 

 101

Table 2. Summary of the corpus comparisons Intralingual comparison

Cross-linguistic comparison

Originals only

All registers in English (chapter 6) All registers in German (chapter 7)

Two registers in English and German (chapter 8)

Translations included

English comparable originals and translations in two registers (chapter 9)

Originals and matching translations in two registers, English-German (chapter 9)

German comparable originals and translations in two registers (chapter 9)

Originals and matching translations in two registers, German-English (chapter 9)

3.2 Relative register values The issue of cross-linguistic comparison was already raised in chapter 3 and will be elaborated here. Differences between language systems often bar direct comparisons of the findings for a given linguistic feature. In order to avoid any confounding influences arising from this fact, we use an indirect method to compare the contrastive results in chapters 8 and 9. Wherever possible, only the magnitude of difference between the frequency of a given indicator in the reference corpus and the respective value per register is compared yielding what one might call relative register values. This method – which is also relevant to the examination of translations in comparison to their source texts in chapter 9 – thus factors out systematic frequency differences in the language pair and highlights register-specific deviations from the respective baseline, i.e. the value for the reference corpus. Nevertheless, the comparison to the reference corpus value is not always applicable because values may be incomparable due to differences in corpus structure or because they are altogether not available. In these cases the value for the register is compared to the register mean, i.e. the mean value of the eight registers included in the study (see chapter 8). Although this value is less balanced than the one for the reference corpus, it seems to be a reasonable replacement. Finally, in those cases where the analysis of an indicator is preliminary due to its exploratory character the absolute results are considered. In the previous sections, the research design as well as the methods of analysis including their limitations were presented. Based on this as well as the discussion in chapters 2 and 3 we can now formulate some general hypotheses that will be tested in the empirical study.

102 

 Research design

4 Hypotheses The three main variables kept under control in this study are language, register and translation status.58 They form the independent variables resulting in three main hypotheses underlying the present study. All similarities and differences discussed below in connection with the hypotheses refer to relative frequencies of given features. It is important to stress that the following hypotheses are very general in character and, therefore, represent some kind of macro-assumptions in terms of the empirical analysis. Since the study aims at providing a comprehensive overview of variation in a wide range of individual features in the chosen combination of data sets, establishing individual hypotheses for each feature is not feasible. Chapters 6 to 9 will show that these are complemented by local expectations concerning the individual indicators under investigation. 4.1 Language-internal variation in originals In the first set of comparisons, the variables language and translation status are kept constant, with only originals being included in the investigation. Due to the variable language being kept constant, two identical sets of analyses have to be conducted for English and German. The hypothesis is the same for both languages. We generally expect differences to occur between the groupings of texts in the corpus. These differences should reflect differing contexts of situation and the resulting registers and should in turn allow the texts to be grouped according to their similarities in view of the features describing a common context of situation. Consequently, there should be a systematic variation between groups of texts that can be attributed to the register, which in turn reflects the context of situation in which the interaction is embedded. Variation is described on the basis of differences between individual features of the groups of texts. There may be an overlap between some groups with respect to some features and at the same time variation with respect to others. Consequently, only complete profiles taking into account all individual indicators will reveal the variation between the groups of texts identified as registers in the corpus. Hypothesis 1 is therefore formulated in the following way: H1 Distinctive profiles are identifiable for each register in the corpus in each language. 58 This is the superordinate concept of the distinction between non-translated and translated texts.

Hypotheses 

 103

4.2 Cross-linguistic variation in originals When investigating contrastive variation, the two variables to be kept constant are translation status (here again only originals are included) and register, thus allowing comparisons within a given register expected to be comparable in both languages. The Anglophone and German-speaking cultures are assumed to be sufficiently similar in terms of social processes (see section 1 of chapter 3) and the resulting registers to suggest that there are a larger number of comparable registers. This leads to the hypothesis that the contrastive registers in the corpus carry more similarities than differences, which, in turn, enables pairs of comparable registers to be identified which contrast with other pairs. Differences between the comparable registers are assumed to arise mainly from contrastive differences between the language systems. These contrastive differences should result in some degree of variation be­­ tween originals belonging to a pair of comparable registers in the two languages. Moreover, possible differences in the context of situation (induced by different contexts of culture) should be reflected in differences between texts representing a comparable register in both languages. It is assumed that the registers in the CroCo corpus are essentially similar, which is probably at least partly attributable to a sampling effect. The necessity of including translations in both directions appears to result in similar texts being sampled from the respective registers in the two languages. Given the degree of overall similarity, it is expected that the comparable registers exhibit gradual differences, which means that a pair of registers can still be regarded as being comparable. Categorical differences, i.e. deviations in the overall interpretation of subdimensions, are expected to be rare. If the differences between the interpretations of subdimensions are determined as being categorical, the interpretation of the contrastive registers as comparable must be abandoned. The resulting hypothesis 2 reads: H2 T  he differences in the comparable registers are gradual rather than categorical.

4.3 Variation between originals and translations The comparison of translations and originals involves two aspects. First, translated and non-translated texts are compared within the same language, namely the target language of the translations (Steiner’s 2008b type of contrast C3, see section 1.1 of this chapter). In this case, not only the register (as in the contrastive

104 

 Research design

comparison) but also the language is kept constant. In the second comparison, the matching source and target texts are compared (Steiner’s type of contrast C4). In this case, only the variable register is kept constant, whereas both language and translation status vary. The translations can be expected to differ from the originals in three respects (Steiner 2001). First, contrastive differences between the languages involved may force the translator to vary a source text feature. This type of variation is sometimes called ‘obligatory’ in translation studies. Secondly, differences between the comparable registers may demand changes in the translations as well.59 Finally, the translations may vary from non-translated texts as a consequence of some conscious or unconscious decision as to the interpretation of the source text. This type of variation is sometimes called ‘optional’ and is often the main interest of research into typical properties.60 These three factors, language contrast, register differences and translation decisions, result in a complex situation with respect to variation between originals and translations. Let us therefore first describe a default situation, or, in statistical terms, the null hypothesis. Wherever the contrastive comparison yields cross-linguistic commonalities, the translations should be similar to both their source texts and the comparable originals in the target language. In cases of contrastive or registerial differences, the results for the translations by default should lie between the source texts and the comparable originals in the target language. This default situation also includes tendencies towards either the source or the target originals. The null hypothesis assumes that in cases of deviation between source and target language originals both shining through, i.e. source language and register interference, and normalisation in terms of assimilation to the target language and register are at work in the translations to varying degrees (see Teich 2003). The alternative hypothesis reflecting the conspicuous cases applies to situations where the translations diverge from both the source and the target texts, i.e. the expected interference and normalisation effects are not present. This is true for cases where –– source and target originals are similar but the translations diverge significantly, –– there is a difference between source and target originals, and the translations display a significantly higher value than the higher value in the originals, 59 This would be the case if hypothesis H2 is refuted because the differences within a pair of registers are interpreted as categorical. 60 The present study does not follow this distinction between obligatory or optional factors because both kinds of factors result in specific properties of translations. Therefore, the three factors should be taken into account as potential sources of explanation.

Summary 

 105

–– there is a difference between source and target originals, and the translations display a significantly lower value than the lower value in the originals. These cases reflect independent variation in the translations that arguably cannot be explained by influences of language contrasts or register differences with the exception of cases where language comparison and register specification point in different directions. Cases of independent variation should be detected with the present research design where the comparison of originals and translations follows the investigation of intralingual as well as contrastive variation. It can be assumed that independent variation can be observed in the translations, thus creating translation-specific hybrid registers. This leads to hypothesis 3: H3 Translations vary to some degree from non-translated texts independently of the influence from language contrasts and register differences.

5 Summary The present chapter has laid the ground for the empirical analysis. The corpus used in the study was described in terms of its design as well as the linguistic information added to it. We then continued to explain the procedure of extracting the required information about observable indicators as well as the statistical processing carried out to evaluate the relevance of the respective indicators. It is important to stress again the limitations of statistics mentioned before in chapter 3 and addressed again in this chapter. In a methodological approach like the one developed here, statistics only play an auxiliary role that helps establish cut-off points for the interpretation. This interpretation has to be qualified further with respect to the strong assumptions made by a number of techniques that cannot be met by linguistic data. Finally, three general hypotheses were established that refer to the three basic comparisons, or independent variables, of the study: intralingual variation between the original registers in the corpus, cross-linguistic variation between original registers and, finally, variation between originals and translations. These hypotheses will be tested in the empirical analysis discussed in the following chapters, which will assess whether these hypotheses have to be confirmed or refuted on the basis of the findings obtained from the analysis.

Chapter 6 English intralingual register variation This chapter is concerned with the detailed investigation of the eight English registers in the corpus concentrating on originals only (see section 3.1 of chapter 5). As discussed in chapter 4, the three register variables field, tenor and mode of discourse are broken down into six subdimensions, which are in turn operationalised in terms of observable indicators. Where the nature of the data produced by the analysis allows it, the quantitative results will be tested statistically using the techniques described in section 2.2 of chapter 5.

1 Field of discourse Field of discourse is the register variable which refers to the experiential, i.e. referential meaning activated in the social action conducted by the interactants. In Halliday and Hasan’s (1989: 12) words, it is concerned with “the nature of the social action that is taking place”. The present study narrows this aspect down to the two subdimensions experiential domain and goal orientation as conceptualised in chapter 4. We will first turn to experiential domain, the subdimension exploring, broadly speaking, the topic of the interaction in a given context of situation.

1.1 Experiential domain Experiential domain describes “the nature of the social activity” (Halliday and Hasan 1989: 56) carried out by the interactants. It is thus concerned with features of the subject matter typically negotiated in a given register. Since the experiential meaning can be expected to vary substantially, the eight registers under investigation should reflect this variation, which is expected to concern all indicators derived in section 1.1 of chapter 4: vocabulary, lexical verbs and lexical chains. These three indicators will be explored in the following sections.

Field of discourse 

 107

Vocabulary Variation in vocabulary usage in English registers is approached by comparing the cumulative frequencies of lexical items in the eight registers. This will give us a rough idea of the subject matter characterising the respective register. The most frequent items of vocabulary are retrieved in their lemmatised form using the lemmatisation as part of the morphology annotation in CroCo. The script used for this query only retrieves those items annotated with open-class part-ofspeech tags, i.e. content words. It thus identifies the most frequent lexical items in the respective corpus. Table 3 displays the 10 most frequent lexical items in the English registers, their frequency (“Freq.”) and their proportion in percent (“%”). Table 3. The 10 most frequent lexical items in EO ESSAY

FICTION

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

TRADE COUNTRY UNITED STATES WORLD U S MARKET NEED EUROPE

299 201 168 162 141 139 137 108 106 104

1.51 1.01 0.85 0.82 0.71 0.7 0.69 0.54 0.53 0.52

SAY GO MAN COME DAY KNOW SEE LOOK BACK TIME

147 103  98  86  82  80  78  68  67  65

0.89 0.63 0.59 0.52 0.5 0.49 0.47 0.41 0.41 0.39

INSTR

POPSCI

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

PAGE USE SELECT PRINT CLICK BATTERY MENU POWER DOCUMENT WEB

312 298 157 151 146 144 136 127 124 122

1.58 1.51 0.79 0.76 0.74 0.73 0.69 0.64 0.63 0.62

CELL INTERFERON PROTEIN WATER GENE HYDROGEN MAKE ACID DRUG BOND

330  93  87  87  84  82  73  70  70  68

1.74 0.49 0.46 0.46 0.44 0.43 0.39 0.37 0.37 0.36

→

108 

 English intralingual register variation

SHARE

SPEECH

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

YEAR BUSINESS COMPANY SERVICE CUSTOMER NEW PERCENT PRODUCT MARKET BAKER

218 215 213 167 135 115 108 107 104 100

1.04 1.02 1.01 0.79 0.64 0.55 0.51 0.51 0.49 0.48

NEW S WORK U COUNTRY UNITED EUROPE STATES PRESIDENT SECURITY

111 104 103 100  93  92  89  88  85  81

0.58 0.54 0.54 0.52 0.48 0.48 0.46 0.46 0.44 0.42

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

CITY TOWN CASTLE MILE CENTRE YEAR WALE CARDIFF COUNTRY SCOTLAND

120 110 107  93  89  81  80  78  75  72

0.59 0.54 0.52 0.45 0.44 0.4 0.39 0.38 0.37 0.35

SERVICE USE UPS LION RIGHT CALL SCHOOL CHILD CLUB INFORMATION

125 108 104 102  97  91  88  86  84  83

0.62 0.53 0.51 0.5 0.48 0.45 0.43 0.42 0.41 0.41

TOU

WEB

This overview reveals clear differences between these registers both with respect to the lexical items most frequently used and the proportion of these items in relation to the total number of lexical items in the respective register (to be gathered from the column “%”). The first four registers, ESSAY, FICTION, INSTR and POPSCI, do not have one single lexical item in common and thus seem to be clearly distinct in terms of vocabulary. There is, however, a certain degree of similarity between other registers in this respect. Particularly, ESSAY and SPEECH appear to use a relatively similar vocabulary: The items United, States, U, S, country and Europe can be found among the ten most frequent lexical items in both registers, suggesting a similar experiential domain covered by both registers. In fact, the senders of both registers are senior government officials. The difference between the two registers rather concerns the indicators describing the mode of discourse since SPEECH is written to be spoken and is therefore likely to display some traits of the spoken medium. The lexical item market occurs both in ESSAY and SHARE. As the most frequent lexical item in ESSAY is trade, we can conjecture that (international) trade politics form the subject matter of the ESSAY texts. Some minor

Field of discourse 

 109

similarities can be found between tourism texts and two other registers. Year, the most frequent lexical item in SHARE, also figures among the top ten in TOU. While it refers to the concrete period of the report to the shareholder in SHARE (e.g. for the year [of the report] or during the year [of the report]; see (1)), it refers to a time relevant to the place of interest described in the respective tourism brochure in TOU. Reference tends to be general and typically relates to seasons of the year as in throughout the year or all year round (see (2)). (1) During the year, we continued our focus on organic growth, added some highly targeted acquisitions, and divested some noncore businesses. (EO_SHARE) (2) Throughout the year, major festivals blow the trumpet for our region’s vitality, flair and sense of FUN! (EO_TOU) Finally, WEB contains two items that also figure among the most frequent lexical items in two other registers: Service is also among the top ten in SHARE, whereas use is also part of the INSTR top ten. These similarities can be explained by the heterogeneous character of the individual texts in the register WEB. Some of the texts are similar to SHARE in that they inform readers about a company (additional evidence for this is the company name UPS among the top ten in WEB), other texts are similar to INSTR in that they contain information on how to use a certain object, hence the item use among the top ten in the register of websites. It must be stressed, though, that it is not so much the individual lexical items that distinguish the registers, but the character of these items or the lexical field that they represent. The top ten items in ESSAY and SPEECH represent the field of international politics, the items in FICTION are very general in character, with the number of verbs being markedly higher than in the other registers. This latter aspect indicates that fictional texts contain many passages describing the actions of the characters. INSTR is also characterised by the use of verbs that have a particularly procedural character. Almost all ten most frequent lexical items in POPSCI are technical terms. SHARE contains various lexical items from the business domain, TOU contains items referring to geographical places. The only register that does not give a clear impression in terms of character of the vocabulary is WEB – a fact that again characterises this register. Table 4 shows the number of lexical types per register. FICTION displays the widest variation in lexical types with 4,378 different lexical types followed by TOU with 4,157 different types. We can assume that this is due to different subject matters covered by the literary texts and the wide range of places of interest described in the tourism brochures. The lowest diversity in lexical types is represented by INSTR with only 2,014 types and the second lowest in ESSAY (2,969). The low value for INSTR can be explained by the restricted character

110 

 English intralingual register variation

of the register. Procedural texts may use a standardised vocabulary or even a controlled language (see Hartley and Paris 2001) to avoid ambiguities in the procedural descriptions. The ESSAY texts were produced by senior government officials in a certain period of time and therefore probably address similar topics, thus limiting the type of vocabulary used. The design of the corpus may partly account for this, but government officials from Anglophone countries may also tend to use a similar range of vocabulary. Table 4. Number of types per register in EO Types ESSAY FICTION INSTR POPSCI SHARE SPEECH TOU WEB

2,969 4,378 2,014 3,598 3,058 3,208 4,157 3,612

The cumulative frequencies may be biased towards some “average” lexical items that appear in all – or at least – most of the texts but are not necessarily the most frequent items in the individual texts. Therefore, we check whether the most frequent item in the cumulative top 10 list is the most frequent word per text as well. For reasons of space, we focus on the registers FICTION and SHARE, which are analysed in further detail. As to say in FICTION, it is the most frequent item in two texts, among the top five in four texts, among the 25 most frequent items in two texts and among the top fifty in one text. In one text, it does not figure among the fifty most frequent words at all. It must be concluded that the most frequent item of the cumulative list for the overall register does not play an important role in all of the individual FICTION texts. What is obviously lost in the cumulative list, are proper names. In the fictional texts these are the names of the characters in the respective story. Being relevant only in the context of the individual text, they do not appear in the cumulative list. This phenomenon is even more apparent in SHARE, where, in six out of 13 texts, the most frequent lexical item is the name of the company reporting to the shareholder about the financial period under review. One of the company names (Baker) is even included in the top ten of the cumulative frequencies.61 Year, the 61 This is due to an imbalance in this subcorpus: There are two texts from the same company in the sample.

Field of discourse 

 111

most frequent lexical item according to the cumulative table, is the most frequent lexical item in only one out of 13 texts. It is among the top five in four texts, among the top ten in five texts and among the top 25 in two texts. One text lists year among the fifty most frequent items. Compared to FICTION, the most frequent lexical item in SHARE occurs more frequently in the individual texts, suggesting a less varied vocabulary. This discussion shows that, while the most frequent lexical items in the cumulative list do play an important role in all the texts, vocabulary specific to the individual texts permits a more accurate characterisation of these texts. The next point to be discussed is whether the texts within a given register are similar to each other in terms of the frequency of the most frequent lexical item of each text and whether they reveal statistically significant differences between the eight registers. Before testing the significance of the variance between the registers, we will first explore the data in the form of the percentage of the most frequent lexical item per all lexical items in a text, the lexical item frequency ratio. The values in EO are normally distributed. Figure 6 displays a box plot of the English registers, showing one outlier62 each in the FICTION and POPSCI registers. The figure also shows that there is some difference in range of the registers with the lowest range in SPEECH (2.49) and the highest range in POPSCI (7.17). As to measures of central tendency, SPEECH also displays the lowest mean and median (M=2.3419, median 2.1968, SD=0.82067). The highest mean and median can be found in instructional texts (M=5.4538, median 5.2926, SD=1.57539, range 4.67). Interestingly, FICTION has the second highest mean and median (M=5.2613, median 5.2122, SD=1.11458, range 4.05), suggesting that there is little variation in terms of highly frequent lexical items in this register. A wide range in a given register points to large differences between the individual values for the individual texts. Low values in central tendency indicate a high degree of lexical variation because they imply that even the most frequent lexical item is not very frequent. In general, SPEECH texts persistently seem to rely on lexical variation, hence the small range of variation in the descriptive statistics. The opposite applies to POPSCI. Here, some of the texts tend to repeat the same items of vocabulary, thus displaying high percentages of the most frequent lexical item, while other texts contain a rather varied vocabulary, resulting in low percentages of the most frequent item. The heterogeneity of the 62 Outliers have a clear impact on the mean value because already few extreme values can distort the mean, and are therefore sometimes factored out for computing significance tests. We keep it in the data in an attempt not to manipulate the data. A text should only be excluded from the analysis if it is identified as an outlier in all (or the majority of the) features. The box plot displayed for those features that are tested using the mean value also contains the median (see section 2.2 of chapter 5).

112 

 English intralingual register variation

10.00

lexical item frequency ratio

8.00

6.00

4.00

2.00

0.00 ESSAY

FICTION

INSTR

POPSCI

SHARE

SPEECH TOURISM

WEB

register Figure 6. Box plot of the lexical item frequency ratio across registers in EO

register WEB explains why this register has the second widest range (5.63). This register may contain both highly repetitive texts and texts which use a more varied vocabulary. Figure 7 63 concentrates on the mean values of the different registers. An analysis of variance tests whether the individual means represent statistically significant differences (for a brief description, see section 2.2.2 of chapter 5). The ANOVA produces a significant difference between the between-groups and the within-groups indices (F(7,102) = 8.863; p < .001). We can thus assume that the registers under investigation realise different populations with respect to the percentage of the most frequent lexical item per all lexical items in a text. In order to establish whether all registers belong to different groups (or whether they form subgroups), we perform the Tukey HSD post-hoc test (see section 63 Note that the lines joining the points are provided merely for the sake of visual clarity. The points refer to the individual registers in this and the following diagrams displaying mean frequencies.

Field of discourse 

 113

means of lexical item frequency ratio

6.00

5.00

4.00

3.00

2.00 Essay

Fiction

Instr

Popsci

Share

Speech Tourism

Web

register Figure 7. Mean values of the lexical item frequency ratio in EO

2.2.2 of chapter 5). It computes three subgroups with some overlaps between the groups (see Table 5). Table 5. Subgroups computed by the Tukey HSD test Register

N

subgroup for alpha=.05 1

SPEECH TOU SHARE ESSAY POPSCI WEB FICTION INSTR Significance

14 11 13 29 11 12 10 10

2.3419 2.7275 3.0198

.930

2 2.7275 3.0198 4.2820 4.4070

.069

3

4.2820 4.4070 5.0079 5.2613 5.4538 .438

114 

 English intralingual register variation

As was to be expected from the mean values in Figure 7, INSTR represents the highest mean value, being part of subgroup 3 with the highest values. The other registers in this subgroup with highly repetitive vocabulary are, in order of descending mean value, FICTION, WEB, POPSCI and ESSAY. The latter two, i.e. POPSCI and ESSAY, also belong to subgroup 2, a middle ranking group in terms of the mean values for the lexical item ratio per text. This group also contains SHARE and TOU, which are also part of subgroup 1 with the lowest mean values. This group is completed by SPEECH and represents the highest degree in lexical variation. The registers in this subgroup can be regarded as being characterised by their use of diverse vocabulary. The subject matter of the individual texts therefore either covers quite broad topics or is expressed in a wider range of sense relations. This may be particularly true for prepared speeches, which may often employ a range of rhetorical devices. The former explanation may apply to tourism brochures; usually these describe various places of interest, appearing to result in relatively low percentages of the most frequent lexical items. By contrast, INSTR clearly builds on a restricted vocabulary with frequent repetitions of the most frequent lexical items. The second highest mean values in FICTION can be explained by the fact that the most frequent lexical items in the individual texts are often the names of the main characters. It is obvious that these proper names are frequently repeated throughout the individual texts. Once again, it becomes apparent why it does not suffice to interpret the cumulative frequency lists for whole registers since, of course, the individual names are only frequent in the individual texts (except for the theoretical case in which two literary texts use the same imaginary names). The names, thus, top the individual frequency lists, but are not visible in the cumulative list, which, as a result, displays lower frequencies compared to other registers where the individual texts may share a common vocabulary. The registers can be expected to differ not only in the choice of vocabulary as was shown by the most frequent lemmas per register. There is also some statistically visible difference with respect to the percentage of the most frequent lexical item in a text. Lexical verbs As an approximation to a transitivity analysis (see the discussion in section 1.1 of chapter 4), we take a closer look at lexical verbs and focus on potential process types resulting from an examination of the lexical meaning of the verbs used in the corpus. Table 5 displays the ten most frequent lexical verbs in all English originals. It gives figures for the absolute frequency of the respective verb lemma (“Freq.”) and the lemma’s proportion of all occurrences of lexical verb lemmas (“%”).

Field of discourse 

 115

Table 6. The 10 most frequent lexical verbs in EO ESSAY

FICTION

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

NEED TRADE WORK MAKE DEVELOP CREATE TAKE INCREASE HELP MEET

 73  61  57  54  47  42  40  36  34  34

1.28 1.07 1 0.94 0.82 0.73 0.7 0.63 0.59 0.59

SAY GO COME SEE KNOW TAKE GET MAKE LOOK THINK

137 102  85  76  63  59  56  55  52  52

1.89 1.41 1.17 1.05 0.87 0.81 0.77 0.76 0.72 0.72

INSTR

POPSCI

N

Word

Freq.

%

Word

Freq.

%

1 2 3 4 5 6 7 8 9 10

USE SELECT CHOOSE CLICK PRINT SET SEE PAGE CHANGE OPEN

158 114  91  85  76  61  59  57  51  49

2.97 2.14 1.71 1.6 1.43 1.15 1.11 1.07 0.96 0.92

MAKE PRODUCE KNOW SHOW USE FORM NEED FIND TAKE BECOME

 61  54  47  38  37  36  34  33  33  31

1.02 0.9 0.79 0.64 0.62 0.6 0.57 0.55 0.55 0.52

Freq.

%

SHARE N

Word

1 2 3 4 5 6 7 8 9 10

CONTINUE MAKE INCREASE PROVIDE HELP WORK SERVICE CREATE TAKE FOCUS

SPEECH Freq.  66  65  55  48  39  39  38  37  36  34

%

Word

1.25 1.23 1.04 0.91 0.74 0.74 0.72 0.7 0.68 0.64

WORK MAKE CONTINUE SAY NEED PROVIDE COME STATE TAKE SUPPORT

84 60 46 43 40 34 30 30 30 29

1.38 0.98 0.75 0.7 0.66 0.56 0.49 0.49 0.49 0.48

→

116 

 English intralingual register variation

TOU N

Word

1 2 3 4 5 6 7 8 9 10

SEE MAKE INCLUDE VISIT BUILD OFFER TAKE COME ENJOY PROVIDE

WEB Freq. 54 50 42 42 40 36 35 33 26 24

%

Word

1.15 1.06 0.89 0.89 0.85 0.76 0.74 0.7 0.55 0.51

USE MAKE WORK CALL LEARN GET INCLUDE TAKE PROVIDE OFFER

Freq. 78 70 48 44 42 36 36 36 35 33

% 1.33 1.19 0.82 0.75 0.71 0.61 0.61 0.61 0.6 0.56

The table does not include the full verbs be and have and is therefore skewed (see section 1.1 of chapter 4). Notwithstanding register-specific differences, it is to be expected that the full verb use, particularly of be, figures among the highest frequencies. To obtain some idea of the role of be, we counted its third person singular form in present tense (is) in relational uses in the two registers analysed more closely (see Figure 8). It is expected that this form occurs frequently in the relational uses of to be. The most frequent lexical verb in FICTION is say (see also Biber et al. 1999: 375 for an overview of the most frequent lexical verbs in a selection of domains), followed by the relational use of is. The remaining verbs among the ten most frequent lexical verbs in fictional texts are indicative of mental (see, know, look, think) and material processes (go, come, take, get, make). In SHARE, the relational use of is already tops the list with 202 occurrences, which is clearly suggestive of an important role of relational processes in this register. Figure 8 shows an extract from the concordance lines for relational uses of is. Continue, the next most frequent verb, often expresses aspect (see example (3)) and is particularly typical of corporate communication, informing shareholders that certain strategies will be maintained or assuring them that the company’s success will last. (3) To become the most respected global financial services company, we must continue to advance our strategic goals-to expand our international franchise, to continue to grow our consumer business, and to ensure that our corporate and investment banking business is best in class. (EO_SHARE) Each register contains at least two verbs that do not appear in any other of the registers. In contrast, the two verbs make and take appear in all registers – with

Field of discourse 

N 50 51 52 53 54 55 56 57 58 59 60

 117

Concordance

at P&G. This has been true for us in the past and is just as true today. P&Gis unique when it comes and business data moving across the Web. This is essential for the growth of e-commerce. We

Diversity. “THE FUTURE Our underlying business is strong, our brand is powerful, our reach is

they do, and perform their jobs with integrity. This is what we value in our people and what we look for shareholders, and franchise so consistently is due, in no small part, to our talented and diverse

after an outstanding record of achievement. Arthur is a wise and courageous leader and our company is strong, our brand is powerful, our reach is unparalleled. We feel tremendous pride in our

Our underlying business is strong, our brand is powerful, our reach is unparalleled. We feel

Second, some two-thirds of the global economy is outside the United States, and no company has

is important to remember that, as big as Citigroup is, we are still small relative to the international

significant growth in the years to come. It is important to remember that, as big as Citigroup

Figure 8. Concordance lines for relational uses of is

the interesting exception of INSTR. This register contains the most verbs specific to this register. This is probably due to the fact that instructional texts require highly explicit verbs that represent the action to be done by the user of the object, hence the lack of light verbs like make and take and hence the specific verbs. Apart from these very general statements, the overview of the ten most frequent lexical verbs does not provide any more insights. All other interpretations require the inclusion of the complete set of verbs – not only the most frequent ones. The number and variation of lexical verbs in the different registers represent further noteworthy aspects. Given a comparable overall corpus size in words (approx. 31,250 per register), the null hypothesis would be an equal number of lexical verbs in all registers. The prominence of verbs among the most frequent lexical items in the register FICTION discussed above suggests register-specific differences in the frequency of lexical verbs. Table 7 gives an overview of the types and tokens in lexical verbs per register. There is a considerable range of diversity of the verb vocabulary from a minimum of 603 lemmas in INSTR to 1,302 lemmas in FICTION. This reflects the limited vocabulary used by instructional texts, which moves in the direction of controlled languages (cf. e.g. Power et al. 2003). Furthermore, the relative proportion of the most frequent verb in INSTR (use, see also the percentage in Table 6) in relation to all verbs in that register is considerably higher than that in all other registers. Consequently, the type-token ratio among lexical verbs (15.28 %) is markedly lower in INSTR than in the other registers reflecting the frequent use of relatively few lexical verbs. Fictional texts, on the other hand, contain a very varied verbal vocabulary. They contain by far the highest number of verb types as well as the highest frequency of verb tokens. This results in the second highest verb type-token ratio. Its value is 4.68 percentage points above the mean value for all eight English original registers (24.51).

118 

 English intralingual register variation

Table 7. Verb statistics for EO

ESSAY FICTION INSTR POPSCI SHARE SPEECH TOU WEB

Types

Tokens

TTR

824 1,302 603 979 790 889 808 939

3,600 4,460 3,947 3,668 3,483 3,641 2,718 3,722

22.89 29.19 15.28 26.69 22.68 24.42 29.73 25.23

The final step in the interpretation of any potential register-specific distribution of lexical verbs is a preliminary examination of process types based on a classification of the lexical meaning of the verbs used in the different registers. As discussed in section 1.1 of chapter 4 this kind of examination is of very limited significance and, therefore, is not tested statistically. 14.0

diff. to ER in percentage points

12.0 10.0

material

mental

relational

verbal

8.0 6.0 4.0 2.0 0.0 –2.0 –4.0 –6.0 –8.0

Essay

Fiction

Instr

Popsci

Share

Speech

Tou

Web

Figure 9. Process type differences between ER and EO

Figure 9 shows the register-specific differences in process types between the English reference corpus and the register-controlled English originals. Keeping in mind the methodological shortcomings of this overview, the results give a first indi­ cation of register differences with respect to process type distribution. All registers

Field of discourse 

 119

show more or less specific distributions as compared to the reference corpus, with FICTION and INSTR showing the most specific pattern. FICTION deviates from the other registers in that it even contains slightly fewer material processes than the reference corpus, while all other registers contain distinctly more material processes. The interpretation of relational processes is difficult because the results are not representative due to the absence of be and have. With respect to mental processes, FICTION again stands out displaying clearly more mental processes than the reference corpus. Arguably, this explains the decrease in material processes in this register. The authors seem to focus on the inner goings-on of the characters rather than their material actions. Finally, all registers include less verbal processes than the reference corpus, albeit to different degrees. The register that stands out most clearly in this comparison is INSTR: It contains 12.51 percentage points more material processes at the expense of all other process types. Instructional texts thus seem to rely to a very high degree on material processes to inform the user what to do with the object being described. Lexical chains The analysis of lexical chains indicates how pervasive the most frequent lexical item in a text is. As discussed in section 1.1 of chapter 4, the analysis is confined to repetitions and does not include sense relations like synonymy. If the most frequent lexical item is not only highly frequent but also occurs continuously in short distances in a text, we can assume that the text focuses very closely on the topic expressed by this lexical item. If, on the other hand, the most frequent lexical item is not repeated very often and appears either only locally in one part of the text or at long intervals, the focus of the text can be assumed to be less restricted. Table 8 gives an overview of all calculations in connection with lexical chains in the English originals. Since this analysis is based on the manual annotation of the corpus, it is only available for FICTION and SHARE at the time of writing. As explained in section 2.2.3 of chapter 5, these results are not processed statistically. The following key figures are computed in connection with lexical chains: –– The average number of sentences per text as the baseline. –– The average frequency of the most frequent lexical item per text (normalised for text length in tokens) as the unit for which lexical chains are calculated. –– The average span between the first and the last occurrence of the most frequent lexical item per text. –– The average length of a lexical chain in number of sentences per text (calculated as the number of consecutive occurrences within a maximum distance of three sentences).

120 

 English intralingual register variation

–– The average distance between occurrences of the most frequent lexical item per text in sentences. Table 8. Key figures for lexical chains in two EO registers

av. no. of sentences per text av. frequency of most freq. item av. span av. chain length av. distance between occurrences

FICTION

SHARE

181.40 18.46 157.90 1.90 7.49

114.54 44.08 106.69 3.80 3.15

The overview in Table 8 suggests register-specific differences between the registers FICTION and SHARE: First, the average number of sentences per text is of interest since it gives us the baseline for the span of lexical chains and for the distance between occurrences. The same stretch of text contains more sentences in FICTION, at the same time, the most frequent lexical item is less frequent on average. Even though less frequent, the most frequent lexical item in FICTION spans a longer stretch of text than in SHARE. Consequentially, the distance between the occurrences is clearly higher in the FICTION texts than in the SHARE texts. The figures in Table 8 imply that there may be a correlation between text length in terms of sentences and span of a lexical item. Nevertheless, the same number of repetitions of a lexical item could also occur within a short span rather than being dispersed loosely over a longer stretch of text and would point to the presence of a local strand of meaning. The resulting chain lengths supply information on the importance of the given lexical item. A longer average length of the lexical chain signals more importance.64 This is, of course, particularly interesting if not only the most frequent lexical item is taken into account but also concurring items. The average chain per text is clearly shorter in FICTION than in SHARE and the distance between the occurrences is longer. We can thus assume that shareholder communication is characterised by a rather frequent recurrence of the most frequent lexical item in the text (at least in comparison to fictional texts). This item is closely connected with the topic of the text: as we have seen above, it frequently is the company name.

64 In terms of cohesion, the subject matter of the text is probably maintained by other cohesive devices as well, particularly reference.

Field of discourse 

 121

It seems plausible that full lexical repetition – irrespective of its being part of a lexical chain – can be a feature of a given register, whereas another register may rely more on pronominalisation. As will be discussed below in section 1.2 of this chapter, fictional texts contain more pronouns than SHARE texts. We can therefore tentatively interpret the lexical chains in economic texts as pointing towards more lexicalised reference. Example (4) shows how the most frequent lexical item in this text, the company’s brand name, is repeated in two consecutive sentences. (4) This environment of dynamic change poses many challenges to Citigroup, but also creates great opportunities to serve our customers, to provide exciting careers for our employees, and to do well for our shareholders. Citigroup has a long and storied history in its constituent parts, including pioneering in international banking at Schroders and Citibank, leading the way in many areas of trading, asset management, and investment banking at Salomon Brothers, bringing modern banking to entire nations through Banamex and Bank Handlowy, and offering thoughtful advice in wealth management to clients through Smith Barney and Citigroup’s Private Bank. (EO_SHARE; our emphasis) This interpretation must be seen as a first indication only. A more fine-grained examination of lexical chains should at least also include synonyms and should result in a more precise picture of lexical chains, moving beyond the focus on mere repetitions of a given lemma. In summary, the two registers which represent the extremes within the range of registers in the CroCo corpus seem to be fictional and instructional texts, with the former being characterised by the highest number of lexical types, a high number and frequency of lexical verbs as well as short lexical chains. The latter, in contrast, show the lowest number of lexical types, a low number of different lexical verbs and, at the same time, frequent repetitions of these, and, moreover, they show the least variation in process types. Unfortunately, figures for lexical chains are not available for INSTR at the time of writing this study. It is to be expected that they confirm the results for vocabulary and lexical verbs and potentially form longer chains than the ones computed for SHARE. FICTION and INSTR have in common a high frequency of lexical items. The discussion of indicators for experiential domain shows that more detailed analyses are available that are particularly geared to the identification of quantitative differences between the registers, which go beyond merely determining the subject matter of a given register. Moreover, it appears that registers make different use of lexico-grammatical means to establish their respective experiential domain. Registers clearly differ in the role highly frequent lexical

122 

 English intralingual register variation

items play, in the way the subject matter is pursued in terms of lexical chains and in the realisation of processes by means of verbs.

1.2 Goal orientation The subdimension goal orientation is concerned with the objectives behind the social action realised by the interactants since these are expected to providing further information on the ideational meaning conveyed in a given register. It will be scrutinised by discussing the four goal types derived in section 1.2 of chapter 4 in turn and determining which of the registers retrieves significant results for the highest number of indicators assumed to point to the respective goal type. However, since results are not available for all relevant features in all registers this discussion is a proof of concept rather than a conclusive analysis of goal orientation. For the study of goal types, it is important to assess the relation between the frequency of the indicators for each of the registers and the respective frequency of occurrence in the reference corpora. It is not sufficient to compare the registers as was done in connection with experiential domain. The register-specific value obtained for a given feature assumed to be an indicator of a given goal type is identified as pointing to this goal type if the value diverges in a statistically significant manner from the reference value in the direction expected to indicate the respective goal type. In some cases, the value is expected only to differ to a lesser degree, or not at all, from the reference corpus to be identified as indicating the given goal type. Additionally, an indicator may also point to a given goal type by its relative, i.e. significant, absence in comparison to the reference value. These indicators are related to Biber’s (1995) “negative” features. A register is identified as representing a given goal type if it complies to all – or at least most of the – expected values for the indicators. However, these are not weighted although it would, perhaps make sense to rate a high frequency of imperative mood in instructional texts higher than, say, the use of the active voice. This weighting is not included at this stage because it requires more research to establish which indicators are more salient than others, which goes beyond the scope of the present study. It should be stressed that the assignment of a register to a goal type can only be carried out on the basis of the combined set of indicators assumed to be representative of the respective goal type. Although each indicator contributes valuable information, its individual interpretation in terms of goal orientation is barred because it does not allow unambiguous assignment to one goal type. This is particularly important for those indicators that are part of the feature matrix of more than one goal type.

Field of discourse 

 123

The interpretation proceeds in two steps: First, the register most representative of a given goal type is identified. Second, the eight CroCo registers are assigned – where applicable – to one of the four goal types. Since at least WEB contains a mixture of texts in terms of goal orientation, it is impossible to make clear statements for all registers.

1.2.1 Argumentation The features identified as indicators relevant to the goal type argumentation in section 1.2 of chapter 4 are relational processes, present tense, modal verbs, modal lexis, infinitives, conditional subordinators, declarative mood, conjunctions. Aspects of theme are not analysed as this would require a functional annotation of the reference corpora, which is not as yet available. With respect to relational processes, we expect a significant increase in relation to the average represented by the reference corpus. Figure 10 suggests that ESSAY, POPSCI and SHARE diverge sufficiently from the baseline to make these registers candidates for the goal type of argumentation. For the reasons discussed in section 1.1 of this chapter, we will not test the significance of this feature. 2.5

Essay Fiction

diff. to ER in percentage points

2.0

Instr

1.5

Popsci Share

1.0

Speech

0.5

Tou Web

0.0 –0.5 –1.0 –1.5

relational

Figure 10. Relational processes in EO compared to ER

In texts that argue in favour of or against a given position, we will expect to find a clear predominance of present tense compared to the language average. The

124 

 English intralingual register variation

following analysis of present tense is very general since it is blind to aspect, distinguishing only between present tense and past tense in the finite verb. Werlich (1976: 144) argues that the axis of orientation in the continuum of time, to which the sender relates phenomena, is either signalled by the present tense group, i.e. present tense, present perfect, future I and future II, or by the past tense group consisting of past tense, past perfect, conditional I and conditional II. In line with this view we assume that the distinction between present and past tense groups is the most important one and therefore interpret our findings on this basis. The eight individual t-tests performed against the English reference corpus produce the following results. The four registers ESSAY, INSTR, POPSCI, SPEECH and TOU contain significantly more present tense forms in relation to all finites than the reference corpus. FICTION contains significantly fewer present tense forms. The two registers SHARE and WEB do not differ significantly from the reference corpus. The most plausible candidates for argumentation in terms of the indicator present tense are thus ESSAY, INSTR, POPSCI, SPEECH and TOU. The next indicator that is assumed to point to argumentation is modality in the form of modal verbs and modal lexis. It is expected that argumentative texts contain significantly more modality than texts pursuing other goals. This expectation applies both to the speaker’s assessment (i.e. epistemic modality) of his/her proposition and to the degree of obligation or inclination (i.e. deontic modality) accorded to a proposition by the sender. A distinction between both forms of modality that does not take into account the context of the utterance is extremely difficult. The frequencies used here are therefore the total frequencies of modal verbs and modal lexis obtained by searching for the lemmas of the core modal verbs and typical lexis. The t-tests conducted against the English reference corpus retrieve two significant results for the frequency of modal verbs per all sentences. POPSCI contains significantly more modal verbs compared to the reference value. TOU has significantly fewer modals than the reference corpus. This result, of course, signifies that tourism brochures do not contain this feature of argumentative texts. The differences between all other registers and the reference value can be attributed to chance. POPSCI is thus the only register which suggests an interpretation in terms of argumentation with respect to this indicator. The t-tests for modal lexis against the reference corpus do not retrieve any significant increase in one of the registers. The only significant result concerns INSTR in the form of a significantly lower mean value in this register. The indicator modal lexis therefore does not yield a candidate for argumentation. We assume that argumentative texts are mainly realised in declarative mood since the sender intends to elaborate on his/her arguments. A register more directly aimed at persuading and thus influencing the addressee of the sender’s

Field of discourse 

 125

view might also contain a significant number of imperative sentences and possibly even interrogatives in order to actively engage the addressee. This might constitute the only difference between argumentative and persuasive texts, and this, in turn, is the reason why we do not include persuasion as a goal type in its own right in this study. Other mood types are expected to be indicative of other goals. High frequencies of imperatives, for instance, are attributed to instruction (see below). Finally, a high frequency of interrogatives is expected to be related to narrative texts (see below). The mean value for declarative mood in the English reference corpus is 95.55 %. The t-tests do not provide any significant results for the registers FICTION, POPSCI, TOU and WEB. SPEECH, ESSAY and SHARE contain significantly more declaratives than the reference corpus. Finally, INSTR is the only register with significantly fewer declaratives than the reference corpus. As regards this indicator, SPEECH, ESSAY and SHARE, all of which tend to use declarative mood exclusively, are probably most representative of argumentation. In a similar vein, we assume that argumentative texts put considerable effort into elaborating a text to achieve coherence and creating explicit connections between the arguments by using more conjunctions than language average. The eight t-tests comparing the individual registers with the reference corpus identify only one significant result for SPEECH, suggesting this register to be representative of argumentation on the basis of the indicator conjunction. Conditional subordination is among the features interpreted by Biber (1995) as pointing to overt expression of argumentation. The only register in our corpus that has a significantly higher relative frequency of the conditional subordinator if than the reference corpus is INSTR. SHARE and TOU contain significantly fewer conditional subordinators than the reference corpus. The sole potential representative of argumentation in terms of conditional subordination is thus INSTR. Another indicator for argumentation identified by Biber (1995) is the use of infinitives. Here, our analysis only retrieves a significant difference between the registers and the reference corpus in TOU. There are two possible explanations why the indicators associated with the argumentative goal by Biber are not equally relevant to the present study, both are concerned with the corpus design of the two studies. Biber’s corpus only contains a small number of instructional texts in the “skills” section of the corpus, whereas this register obviously has a strong influence on the findings of the present study. Had Biber included instructional texts as a whole section of his corpus, his findings might have been different with regard to conditional subordination, possibly resulting in exclusion of this indicator of argumentation (or a different interpretation of the dimension). The same issue applies to the corpus design of the present study. Here, too, the

126 

 English intralingual register variation

findings are strongly connected to the registers chosen and, in particular, to the make-up of the reference corpus. Due to the small size of the corpus, the selection of texts and registers is reflected in the results. Used on a reference corpus consisting of different texts and registers, the significance tests might produce different results. Let us now summarise the findings for the argumentation indicators.

FICTION INSTR POPSCI SHARE SPEECH TOU WEB

×

×

×

×

Infinitives

Conditional subordinator

×

Conjunctions

×

Declarative mood

×

Modal lexis

×

Modal verbs

Present tense

ESSAY

Relational processes

Table 9. Significant results for argumentation indicators in EO

× ×

× ×

×

×

×

Table 9 lists the significant results obtained for the various indicators in the eight English registers. Judging from the number of indicators providing significant results for the respective registers, the three registers ESSAY, POPSCI and SPEECH point to argumentation (printed in bold in the table). The intermediate summary at the end of this section will discuss whether all three registers are best represented by argumentation.

1.2.2 Exposition The indicators used to characterise the goal type exposition overlap in part with those for argumentation (relational processes, present tense, declarative mood). However, in this connection, other indicators come into play that are not marked for argumentation, namely a high frequency of the passive voice, nouns, attributive adjectives and nominalisations and a low frequency of personal pronouns with the exception of the third person singular neuter it. Type-token ratio will not be analysed here because the diverging text lengths of the individual texts do not allow comparisons of the mean values.

Field of discourse 

 127

As we have seen above, the registers ESSAY, POPSCI and SHARE are characterised by a higher frequency of relational processes as compared to the reference corpus. Present tense is used significantly more frequently in ESSAY, INSTR, POPSCI, SPEECH and TOU than in the reference corpus. And finally, declarative mood is particularly characteristic of ESSAY, SHARE and SPEECH. Let us now turn to those features not yet described in greater detail. As to passive voice, a text serving the description of an object or a situation, i.e. an expository text, will focus on the object rather than on the agents affecting the object. The passive voice allows the focus of an utterance to be placed on facts by enabling the agent of a process to remain unspecified. The range to which the passive voice is used in the 17 registers of the reference corpus varies considerably with a minimum value of 3.59 per total number of sentences in cookbooks and a maximum of 49.06 in calls (for tender) (see Figure 11). 50.00

passives per no. of sentences

40.00

30.00

20.00

10.00

0.00 ESSAY

FICTION

INSTR

POPSCI

SHARE

SPEECH TOURISM

WEB

REFERENCE

register Figure 11. Box plot of passives per sentences in EO and ER

Given the fact that all of the registers which differ significantly from the reference corpus have significantly fewer passives than the reference corpus, we can interpret the three registers not deviating significantly from the reference

128 

 English intralingual register variation

corpus, i.e. POPSCI, TOU and WEB, as pointing to exposition. Of these three registers, POPSCI is the only one with a mean value higher than the reference corpus mean value. The registers ESSAY, FICTION, INSTR, SHARE and SPEECH contain distinctly fewer occurrences of the passive voice. The nominal style expected to be characteristic of expository texts should be observable in terms of a high frequency of nouns as well as a high frequency of attributive adjectives since they reflect more complex nominal phrases. The t-tests retrieve significant increases of nouns compared to the reference corpus in ESSAY, INSTR, SHARE, TOU and WEB. A significant drop in nouns was found in FICTION. Judging by the frequency of nouns, the five registers ESSAY, INSTR, SHARE, TOU and WEB qualify as candidates for exposition. As to the relative frequency of attributive adjectives, the t-tests identified significantly higher frequencies in the four registers ESSAY, POPSCI, SPEECH and TOU than in the reference corpus. These registers may consequently represent the expository goal type with respect to the frequency of attributive adjectives. The remaining registers do not differ significantly from the reference corpus. Expository texts are expected to present information in a highly condensed fashion. One result of this compression of information could be a high frequency of nominalisations since they package information which would otherwise be realised in a complete clausal structure in nominal structures. Nominalisations are identified by searching for nominal parts-of-speech with suffixes signalling nominalisation. Table 10 exemplifies this for nominalisations on -ism in the English reference corpus. Table 10. Nominalisations with the suffix -ism in ER Text

Token

PoS tag

ER_GENFICTION_001 ER_LETT_001 ER_LETT_002 ER_POP_005

optimism Ritualism optimism realism anti-Semitism Fascism Communism Judaism Criticism Judaism rationalism Deism favouritism favouritism feudalism

NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1 NN1

ER_PREPSPEECH_001 ER_RELIG_001

ER_RELIG_002 ER_REP_001 ER_REP_002 ER_REP_007

Field of discourse 

 129

Three out of eight registers contain significantly more nominalisations than the reference texts. These registers are ESSAY, SHARE and SPEECH suggesting an interpretation as representatives of exposition with respect to a high frequency of nominalisations. FICTION is characterised by significantly fewer nominalisations as compared to the reference corpus. The focus on factual information which was already shown to be expressed by means of many nominal lexical items suggests that expository texts use relatively few personal pronouns. The only register scoring a significant result for a decrease in personal pronouns compared to the reference corpus is TOU. This register is thus identified as a potential representative for exposition as far as this indicator is concerned. The only personal pronoun that is expected to occur frequently in exposition is the third person singular neuter personal pronoun it. While expository texts are expected not to refer frequently to persons with the help of personal pronouns, they should frequently refer to the object that is being described or the relation between entities (and, in this sense, the pronoun it should co-occur with relational processes). Four of the eight registers have a significantly lower proportion of it per all personal pronouns (FICTION, SHARE, SPEECH, WEB). The other four registers (ESSAY, INSTR, POPSCI, TOU) do not differ significantly from the reference value. We can therefore not identify a candidate for exposition with respect to this indicator.

INSTR POPSCI SHARE SPEECH TOU WEB

× ×

× × × ×

× ×

× × ×

×

×

× × × ×

× × ×

× ×

×

“it”

×

Absence of personal pronouns

Nominalisations

×

Attributive adjectives

×

Nouns

×

Passive voice

Declarative mood

FICTION

Present tense

ESSAY

Relational processes

Table 11. Significant results for exposition indicators in EO

130 

 English intralingual register variation

Out of the nine indicators associated with exposition six apply to ESSAY (printed in bold in Table 11), suggesting that this register pursues the goal of exposition. The fact that ESSAY was identified as pursuing argumentative goals on the basis of only three indicators suggests that it may be better interpreted as pursuing expository goals. Table 11 also illustrates that a considerable share of the registers must be at least interpreted as displaying expository traits (the assignment of these registers to a goal type will be discussed at the end of this section). Particularly POPSCI, which was also identified as a good representative of argumentation, carries more indicators related to exposition, thus overruling the preliminary interpretation.

1.2.3 Instruction While argumentative and expository goals are represented by many overlapping features in texts, instructional goals should be realised quite differently in texts. It is no coincidence that one of the CroCo registers is called ‘instruction’ since this goal seems to represent the overriding characteristic of texts explaining users how to work with a given object. However, the goal type instruction is not limited to instruction manuals as represented by the relevant CroCo register, but it concerns all kinds of texts intended to influence their respective addressees’ behaviour. The indicators deemed characteristic of instruction are material processes, present tense, frequent use of the imperative mood and the active voice, few personal pronouns and, finally, a high proportion of second person pronouns. The expected iconicity in terms of thematic structure discussed in section 1.2 of chapter 4 will not be analysed here because results are not available for all registers. Since instructional texts are assumed to focus on actions, for example, interaction with a given object, we expect the material process type to be the most frequent one, at the expense of all other process types. Figure 12 shows that, while all the English registers except for FICTION seem to contain more material processes than the reference corpus, the difference is by far most marked for INSTR followed by ESSAY, TOU and SHARE. In the light of this observation, these registers appear to be representative of the instructional goal type with respect to the indicator material process frequency. The instructional goal should be firmly set in the present tense. Instructional texts are not likely to be about what happened in the past or will happen in the future; but rather we expect them to concentrate on what the addressee does presently. We have seen above that ESSAY, INSTR, POPSCI, SPEECH and TOU contain significantly more present tense forms than the reference corpus. All of these are therefore indicative of instruction in terms of present tense.

Field of discourse 

14.0

Essay Fiction

12.0 diff. to ER in percentage points

 131

Instr

10.0

Popsci Share

8.0

Speech

6.0

Tou Web

4.0 2.0 0.0 –2.0

material

Figure 12. Material processes in EO compared to ER

Texts pursuing an instructional goal are expected to make more extensive use of the imperative mood than language in general since the imperative provides a direct means of asking the addressee to change his/her behaviour, e.g. by handling an object. The t-tests detect a significant increase in INSTR and WEB in relation to the reference corpus, with none of the other registers deviating significantly from the reference value. With respect to imperative mood, INSTR and WEB are therefore candidates for the instructional goal type. Instructional texts can be expected to frequently use the active voice in order to explicitly specify the agent of the action expressed by material processes. The registers found to contain significantly fewer passives per number of sentences in the t-tests against the reference corpus are ESSAY, FICTION, INSTR, SHARE and SPEECH. Since these registers contain significantly more active sentences than the reference corpus, they are the most likely representatives of instruction with respect to this feature. Owing to an assumed focus on the object on which the addressee is instructed, we would expect instructional texts to use fewer personal pronouns than the reference corpus. The t-tests comparing the register-controlled corpora with the reference corpus retrieve a significant difference for TOU, making this the most probable representative of candidate for instruction with respect to this feature. FICTION contains significantly more personal pronouns than the reference texts. None of the other registers differs significantly from the reference corpus in their usage of personal pronouns.

132 

 English intralingual register variation

Among the relatively few personal pronouns, second person pronouns are expected to figure prominently in instructional texts. The t-tests identify significantly higher proportions of second person pronouns in relation to all pronouns in INSTR, TOU and WEB, making these the most likely representatives of instruction with respect to this indicator.

FICTION INSTR POPSCI SHARE SPEECH TOU WEB

×

× ×

×

× ×

×

× ×

×

× × ×

Second person pronouns

×

Absence of personal pronouns

×

Active voice

×

Imperative mood

Present tense

ESSAY

Material processes

Table 12. Significant results for instruction indicators in EO

×

×

× ×

The overview of registers potentially having an instructional goal according to the individual indicators as depicted in Table 12 suggests that the register INSTR (printed in bold in the table) best represents the goal type instruction. 1.2.4 Narration The last goal type to be discussed here is narration. The indicators identified for this goal type are above average frequency of mental and verbal processes and a tendency to use verbal/clausal structures; this is represented by the frequent use of finites at the expense of nominal structures, which, in turn, results in a low frequency of nouns. Moreover, attributive adjectives, nominalisations, subordination and a strong reliance on personal pronouns, past tense forms and interrogatives constitute further indicators. So far, we have discussed material and relational processes in connection with goal orientation. Now, we will focus on the frequency of mental and verbal processes. Mental processes are likely to be used in narrative texts to express the cognitive, perceptive and affective experiences of the characters featuring in the narrative. Verbal processes, in turn, should be frequently used to report on the characters’ utterances.

Field of discourse 

diff. to ER in percentage points

4.0

 133

Essay Fiction

2.0

Instr Popsci

0.0

Share Speech

–2.0

Tou Web

–4.0 –6.0 –8.0

mental

verbal

Figure 13. Mental and verbal processes in EO compared to ER

Figure 13 represents the deviation of the eight CroCo registers from the baseline in terms of mental and verbal processes. While we must bear in mind the exploratory character of this indicator, FICTION appears to differ distinctly from the other registers in that it contains more mental processes than the reference corpus. With respect to verbal processes, this register does not differ from the other registers as all the registers contain fewer verbal processes than the baseline. This means that it is not possible to identify a register serving as a candidate for narration with respect to the frequency of verbal processes. Nevertheless, FICTION can be seen as a candidate for the narrative goal type with respect to mental process frequency. Grammatical structures that spell out the relations between processes, participants and circumstances are expected to occur more frequently in narrative texts than structures compressing these relations into nominal groups. Without a comprehensive transitivity analysis, this characteristic should be retrievable from the proportion of finite verbs in relation to all tokens (since finites are required to create clauses) and the relative sparseness of nouns, attributive adjectives and nominalisations (since these aid the compression of clausal structures, see above). The t-tests only produce significant results for the proportion of finites per all tokens in those registers with clearly lower frequencies of finites, i.e. for ESSAY, SHARE, TOU and WEB. Consequently, no candidates for narration result from this indicator. This also applies to the relative absence of attributive adjectives. The t-tests do not produce any significant decrease for this indicator in comparison to the reference corpus.

134 

 English intralingual register variation

The only register in the study having a significantly lower mean value for nouns and nominalisations than the reference corpus is FICTION. The other registers either differ only slightly or have a significantly higher frequency of nominalisations (see exposition). FICTION is thus the only probable representative of the narrative goal type with respect to the relative absence of nouns and nominalisations. As discussed in section 1.2 of chapter 4, de Beaugrande and Dressler (1981) include subordination as an indicator for narrative texts. However, the results of the present study do not appear to substantiate this assumption. Only POPSCI contains significantly more subordinating conjunctions in relation to the number of all tokens, a finding which suggests the presence of more subordinated clauses in POPSCI than in the reference corpus. Personal pronouns should be frequently used in narrative texts to refer to the persons or things (in third person neuter) being described. In this sense, pronouns should be a more appropriate cohesive device in narrative texts than, for instance, lexical cohesion (see section 1.1 of this chapter). According to the t-tests, the only register containing a significantly higher proportion of personal pronouns per all tokens than the reference corpus is FICTION. TOU contains a significantly lower proportion, none of the other registers differs significantly from the reference corpus. Again, this means that FICTION is the sole register representing a narrative character with respect to this indicator. The frequency of present tense was discussed previously in connection with the other goal types. Now, past tense is taken into account. We would expect texts realising a narrative goal to show a significantly higher proportion of past tense forms per all finites than the reference corpus. Among the eight registers, FICTION is the only one which contains a significantly higher proportion. Five other registers have a significantly lower proportion, and two (SHARE and WEB) do not differ significantly from the reference corpus. FICTION is therefore again the only candidate for narration with regard to this indicator. Narrative texts are expected to contain many descriptions of dialogues between characters. These dialogues are expected to contain a higher frequency of interrogatives than the reference corpora. The t-tests retrieve a significantly higher number of interrogatives per all sentences in FICTION, once more making this register the only candidate for narration with respect to frequency of interrogatives. All the other registers display either insignificant or significantly lower mean values for this indicator. The picture of the narrative goal type emerging from Table 13 is obvious. FICTION is the most probable representative of narration with respect to six out of ten indicators. Only one other register, POPSCI, contains a significant finding for one of the narration indicators at all. FICTION can therefore be classified as clearly representing this goal type.

Field of discourse 

 135

POPSCI SHARE

×

Subordination

Absence of nominalisations

Absence of attrib. adjectives

Finites

Absence of nouns

×

×

Interrogatives

INSTR

Past tense

FICTION

Personal pronouns

ESSAY

Verbal processes

Mental processes

Table 13. Significant results for narration indicators in EO

×

×

×

×

SPEECH TOU WEB

In a final step of interpretation, we now have to identify the goal types assigned to all of the registers under investigation in this study. We have seen that ESSAY, POPSCI and SPEECH represent the argumentative goal type. ESSAY and POPSCI also scored the highest number of indicators for exposition. Here, the first interpretation in terms of argumentation was overruled because both registers carry more indicators suggestive of exposition. Nevertheless, this ambiguity underlines the similarities existing between these two goal types. The instructional goal type is best realised by our INSTR register and narration is realised by FICTION. The remaining three registers SHARE, TOU and WEB were not identified as best representatives of a given goal type, but may still be characterised by one of the four goal types discussed here. SHARE and TOU can be assigned to exposition as they score the highest number of indicators (four for SHARE and five for TOU) for this goal type. TOU also appears to show some traits of the instructional goal type since it still contains four significant results for this type. It gives information about places of interest and instructs the addressee on what to do there. WEB, finally, was a candidate for two indicators in exposition and instruction respectively. This suggests that this register either does not represent a single goal type or consists of highly heterogeneous texts. The findings discussed in view of goal orientation exhibit some vagueness with respect to certain goal types, in particular to the distinction between exposition and argumentation. This may be attributable to a number of reasons:

136 

 English intralingual register variation

–– The global application of goal types not only to whole texts but even to entire registers may not work for all the registers discussed here. This applies particularly to WEB, a register consisting of a rather heterogeneous mix of texts. –– The registers may not be the best representatives of the goal types discussed here. This holds for argumentation in particular. –– The features analysed here may not be the most appropriate indicators for all goal types. As discussed in section 3 of chapter 3, the validity of the research design is a guiding principle, the present analysis shows that it is a constant problem that is not easy to solve. The analysis should therefore be seen as a first attempt since, to our knowledge, no one so far has attempted to analyse goal types in this detailed way which involves concrete linguistic evidence rather than the discussion of individual examples. In this regard, the present approach resembles Biber’s (1988, 1989, 1995) methodology. –– Some of the features that can be expected to represent good indicators for a given goal type were not available for all registers. Future work will have to prove whether the inclusion of these indicators will improve the quality of the analysis of goal orientation. Despite the limitations of the analysis, this type of interpretation appears to represent a step towards a more systematic identification of goals pursued in various registers.

2 Tenor of discourse The second register variable, tenor of discourse, addresses those aspects of a context of situation which refer to the relationship between sender and addressee. It is investigated in terms of the two subdimensions social role relationship covering the linguistic effects of the (different) statuses of the interactants in society and social distance, the subdimension representing the (communicative) background shared by the interactants. 2.1 Social role relationship The subdimension social role relationship looks into those features of a register resulting from the social status held by the interactants. If the status is not equal but hierarchical, this should have an impact on how the interactants express themselves. As described in section 2.2 of chapter 4, this subdimension is further broken down into the two intermediate categories level of authority and level of expertise, which are then operationalised so as to obtain observable

Tenor of discourse 

 137

indicators. We expect to find only limited variation between the registers in the corpus. None of the registers is expected to display a clear asymmetry in the level of authority between the interactants expressed in the texts. Differences should merely arise in terms of expertise. Some of the texts can be assumed to require a high level of expertise on the part of the addressee in order to maintain an equal relationship between the interactants.

2.1.1 Level of authority The intermediate category level of authority is intended to shed light on those features of a register resulting from the degree of authority assumed by one of the interactants. As derived in section 2.2 of chapter 4, the first indicator to be analysed for level of authority is mood. A sender with a high level of authority should be in a position to make demands at his/her discretion. This may be reflected in an above average frequency of imperatives, used to express demands for goods and services. The opposite case, where the sender has a lower level of authority than the addressee, may also be found in the corpus. In this case, the sender may simply provide information and use the declarative mood for this purpose; or s/he may demand information from the addressee and use the interrogative mood. The interrogative mood may also open up the interaction, inviting a contribution from the addressee, thereby reducing the asymmetry of the exchange. The social role relationship may, of course, also be equal. In this case, we expect the declarative mood to be the neutral option, particularly in monologic communication. In dialogic interaction among peers, other, more varied distributions of the mood options are conceivable as well. These are likely to be distributed similarly among the interactants. The fact that mood options, e.g. declarative mood, may be interpreted in different directions underlines the importance of taking into consideration several indicators in the interpretation process. Declaratives are assumed to be neutral with respect to level of authority. They may also be used by a sender possessing less authority than the addressee. The ANOVA for declaratives per all sentences in the English originals produces a significant difference between the between-groups and the within-groups indices (F(8,118)=9.751; p