I COHESION in TRANSLATION

Home , Genre studies

COHESION IN TRANSLATION: A CORPUS STUDY OF HUMAN- TRANSLATED, MACHINE-TRANSLATED, AND NON-TRANSLATED TEXTS (RUSSIAN INTO ENGLISH)

A dissertation submitted

to Kent State University in partial

fulfillment of the requirements for the

degree of Doctor of Philosophy

Tatyana Bystrova-McIntyre

December, 2012

Dissertation written by Tatyana Bystrova-McIntyre B.A., Tver State University, Russia, 1999 M.A., Kent State University, OH, 2004 M.A., Kent State University, OH, 2006 Ph.D., Kent State University, OH, 2012

APPROVED BY

______, Chair, Doctoral Dissertation Committee Brian J. Baer (advisor)

______, Members, Doctoral Dissertation Committee Gregory M. Shreve

______, Erik B. Angelone

______, Andrew S. Barnes

______, Mikhail Nesterenko

ACCEPTED BY

______, Interim Chair, Modern and Classical Language Studies Keiran J. Dunne

______, Dean, College of Arts and Sciences Timothy Chandler

iii

TABLE OF CONTENTS

LIST OF FIGURES ...... XIII

LIST OF TABLES ...... XX

DEDICATION ...... XXXVI

ACKNOWLEDGEMENTS ...... XXXVII

CHAPTER 1: INTRODUCTION ...... 1

1.1 Background ...... 2

1.1.1 Global textual features and translation: Text as a culture-

specific phenomenon ...... 2

1.1.2 New realities of the translation industry today ...... 5

1.1.2.1 Machine translation as a reality of modern

translation practice ...... 5

1.1.2.2 Editing and post-editing in the translation industry

today ...... 9

1.1.2.3 Translation memory tools and their effect on

translators' work ...... 10

1.2 Statement of the problem ...... 11

1.3 Goals and components of this study ...... 12

1.4 Research questions ...... 14

1.5 Organization of this study ...... 15

CHAPTER 2: LITERATURE REVIEW ...... 16

2.1 Non-translated vs. translated texts ...... 16

2.1.1 The notion of translationese ...... 16

2.1.2 General laws of translation ...... 18

2.1.3 Interference of the source language vs. translation universals

...... 19

2.1.4 Methodology of studies involving translated and non-

translated texts ...... 22

2.1.5 Applicability of studies involving translated and non-

translated texts ...... 22

2.2 Cohesion, global textual features, and related terms ...... 23

2.2.1 Text organization and discourse analysis ...... 23

2.2.2 Cohesion as a linguistic phenomenon ...... 26

2.2.3 Cohesion and texture...... 28

2.2.4 Cohesion vs. coherence ...... 28

2.2.5 Standards of textuality ...... 30

2.2.6 Definition of cohesion in the present study ...... 31

2.2.7 Categories of cohesion ...... 32

2.2.7.1 Grammatical cohesion ...... 33

2.2.7.2 Lexical cohesion ...... 46

2.2.8 Other global textual features ...... 50

2.2.8.1 Nominalization ...... 51

2.2.8.2 Lexical density ...... 52

2.2.8.3 Average word length ...... 52

2.2.8.4 Average sentence length ...... 53

2.2.8.5 Passives ...... 54

2.2.8.6 Prepositional phrases ...... 55

2.2.9 Cohesion and other global textual features as a language- and

culture-specific phenomenon ...... 55

2.3 Studies of cohesion and other global textual features ...... 59

2.3.1 Studies of cohesion in the context of teaching and evaluating

writing ...... 59

2.3.2 Studies of cohesion in the context of text comprehension ... 62

2.3.3 Studies of cohesion as a genre-specific and text-type specific

phenomenon ...... 63

2.3.4 Studies of cohesion in the context of translation studies ..... 65

2.3.5 Studies of other global textual features in the context of

translation studies...... 69

2.3.6 Cohesion and other global textual features and translation

expertise ...... 71

2.4 Conclusions ...... 73

CHAPTER 3: METHODS AND MATERIALS ...... 76

3.1 Corpus linguistics as a methodology to analyze cohesion and other

global textual features ...... 76

3.2 The framework for studying cohesion and other global textual features

...... 78

3.3 Materials ...... 83

3.3.1 Corpus compilation ...... 83

3.3.1.1 Literary corpus ...... 89

3.3.1.2 Newspaper corpus ...... 96

3.3.1.3 Scientific corpus...... 98

3.4 Methods and procedures ...... 101

3.4.1 Corpus database ...... 101

3.4.2 Corpus annotation ...... 102

3.4.3 Data collection ...... 103

3.4.4 Data analysis: General procedures for descriptive statistical

analysis and significance testing ...... 106

CHAPTER 4: TEXTUAL CHARACTERISTICS OF LITERARY, NEWSPAPER, AND

SCIENTIFIC TEXTS ACROSS NON-TRANSLATED, HUMAN-TRANSLATED,

AND MACHINE-TRANSLATED CORPORA ...... 111

4.1 Cohesive and other global textual characteristics selected for the genre

description of the corpora ...... 111

4.2 Cohesive characteristics of literary, newspaper, and scientific texts 113

vii

4.2.1 Third-person pronominal cohesive devices in literary, newspaper, and scientific texts ...... 113

4.2.2 Possessive pronouns in literary, newspaper, and scientific texts ...... 116

4.2.3 Demonstrative pronouns in literary, newspaper, and scientific texts ...... 120

4.2.4 Definite article in literary, newspaper, and scientific texts 124

4.2.5 Comparative cohesive devices in literary, newspaper, and scientific texts ...... 128

4.2.6 Reference cohesive devices in literary, newspaper, and scientific texts ...... 131

4.2.7 Conjunction cohesive devices in literary, newspaper, and scientific texts ...... 135

4.3 Textual characteristics of literary, newspaper, and scientific texts .. 140

4.3.1 Nominalization in literary, newspaper, and scientific texts 140

4.3.2 Lexical density in literary, newspaper, and scientific texts 144

4.3.3 Average word length in literary, newspaper, and scientific texts ...... 148

4.3.4 Average sentence length in literary, newspaper, and scientific texts ...... 152

4.3.5 Passives in literary, newspaper, and scientific texts ...... 156

viii

4.3.6 Prepositional phrases in literary, newspaper, and scientific

texts ...... 160

4.4 Conclusions ...... 164

CHAPTER 5: RESULTS AND ANALYSIS: ASSOCIATION OF COHESIVE

DEVICES AND OTHER GLOBAL TEXTUAL FEATURES WITH THE METHOD OF

TEXT PRODUCTION...... 171

5.1 Association of cohesive devices and other global textual features with

the method of text production: Overview of variables under study...... 171

5.1.1 Cohesive devices ...... 172

5.1.1.1 Reference cohesive devices ...... 172

5.1.1.2 Conjunction cohesive devices ...... 173

5.1.1.3 Reference and conjunction cohesive devices ...... 174

5.1.2 Global textual features ...... 174

5.2 Association of cohesive devices and other global textual features with

the method of text production: Literary corpus ...... 175

5.2.1 Cohesive devices in non-translated, human-translated, and

machine-translated texts in the literary corpus ...... 175

5.2.1.1 Reference cohesive devices in the literary corpus

...... 175

5.2.1.2 Conjunction cohesive devices in the literary corpus

...... 202

5.2.1.3 Total reference and conjunction cohesive devices in

the literary corpus ...... 208

5.2.2 Global textual features in non-translated, human-translated,

and machine-translated texts in the literary corpus ...... 211

5.2.2.1 Nominalization ...... 211

5.2.2.2 Lexical density ...... 212

5.2.2.3 Average word length ...... 216

5.2.2.4 Average sentence length ...... 217

5.2.2.5 Passives ...... 218

5.2.2.6 Prepositional phrases ...... 218

5.3 Association of cohesive devices and other global textual features with the method of text production: Newspaper corpus ...... 223

5.3.1 Cohesive devices in non-translated, human-translated, and

machine-translated texts in the newspaper corpus ...... 223

5.3.1.1 Reference cohesive devices in the newspaper corpus

...... 223

5.3.1.2 Conjunction cohesive devices ...... 250

5.4.1.3 Reference and conjunction cohesive devices in the

newspaper corpus ...... 259

5.3.2 Global textual features in non-translated, human-translated,

and machine-translated texts in the newspaper corpus ...... 263

5.3.2.1 Nominalization ...... 263

5.3.2.2 Lexical density ...... 266

5.3.2.3 Average word length ...... 269

5.3.2.4 Average sentence length ...... 269

5.3.2.5 Passives ...... 271

5.3.2.6 Prepositional phrases ...... 272

5.4 Association of cohesive devices and other global textual features with the method of text production: Scientific corpus ...... 279

5.4.1 Cohesive devices in non-translated, human-translated, and

machine-translated texts in the scientific corpus ...... 279

5.4.1.1 Reference cohesive devices in the scientific corpus

...... 279

5.4.1.2 Conjunction cohesive devices in the scientific

corpus ...... 304

5.4.1.3 Reference and conjunction cohesive devices in the

scientific corpus ...... 311

5.4.2 Global textual features in non-translated, human-translated,

and machine-translated texts in the scientific corpus ...... 313

5.4.2.1 Nominalization ...... 313

5.4.2.2 Lexical density ...... 315

5.4.2.3 Average word length ...... 317

5.4.2.4 Average sentence length ...... 320

5.4.2.5 Passives ...... 322

5.4.2.6 Prepositional phrases ...... 325

5.5 Conclusions ...... 336

CHAPTER 6: CONCLUSIONS, RECOMMENDATIONS, AND FUTURE

DIRECTIONS ...... 337

6.1 Summary of results ...... 338

6.1.1 Summary of results for the literary corpus ...... 338

6.1.2 Summary of results for the newspaper corpus ...... 349

6.1.3 Summary of results for the scientific corpus ...... 362

6.2 Practical recommendations for linguists working in the translation

industry and translator trainers...... 373

6.3 Limitations of this study ...... 376

6.4 Future research directions ...... 378

REFERENCES ...... 382

xii

LIST OF FIGURES

Fig. 3.1 Example of CLAWS POS-tagged text ...... 103

Fig. 4.1 Means and standard error for 3rd person pronominal cohesive devices for NT,

HT, and MT in the literary, newspaper, and scientific corpora ...... 116

Fig. 4.2 Means and standard error for possessive pronouns for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 120

Fig. 4.3 Means and standard error for demonstratives for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 123

Fig. 4.4 Means and standard error for the definite article for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 127

Fig. 4.5 Means and standard error for comparative devices for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 131

Fig. 4.6 Means and standard error for reference cohesive devices NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 134

Fig. 4.7 Means and standard error for conjunction cohesive devices for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 139

Fig. 4.8 Means and standard error for nominalization for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 143

xiii

Fig. 4.9 Means and standard error for STTR for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 147

Fig. 4.10 Means and standard error for average word length for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 151

Fig. 4.11 Means and standard error for average sentence length for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 155

Fig. 4.12 Means and standard error for passives for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 159

Fig. 4.13 Means and standard error for prepositions for NT, HT, and MT in the literary, newspaper, and scientific corpora ...... 163

Fig. 5.1 Means and standard error (± 1 SE) for 3rd person singular objective pronominal cohesive devices in literary texts ...... 179

Fig. 5.2 Means and standard error (± 1 SE) for 3rd person plural objective pronominal cohesive devices in literary texts ...... 181

Fig. 5.3 Means and standard error (± 1 SE) for 3rd person singular subjective pronominal cohesive devices in literary texts ...... 184

Fig. 5.4 Means and standard error (± 1 SE) for 3rd person plural subjective pronominal cohesive devices in literary texts ...... 186

Fig. 5.5 Means and standard error (± 1 SE) for the total of 3rd person pronominal cohesive devices in literary texts ...... 187

Fig. 5.6 Means and standard error (± 1 SE) for possessive pronouns in literary texts ... 191

xiv

Fig. 5.7 Means and standard error (± 1 SE) for singular demonstratives in literary texts

...... 194

Fig. 5.8 Means and standard error (± 1 SE) for definite article "the" in literary texts .... 197

Fig. 5.9 Means and standard error (± 1 SE) for the total number of reference cohesive devices in literary texts ...... 201

Fig. 5.10 Means and standard error (± 1 SE) for additive conjunction devices in literary texts ...... 205

Fig. 5.11 Means and standard error (± 1 SE) for adversative conjunction devices in literary texts ...... 206

Fig. 5.12 Means and standard error (± 1 SE) for the total number of conjunction devices in literary texts ...... 207

Fig. 5.13 Means and standard error (± 1 SE) for total reference and conjunction devices in literary texts ...... 210

Fig. 5.14 Means and standard error (± 1 SE) for STTR in literary texts ...... 214

Fig. 5.15 Means and standard error (± 1 SE) for prepositional phrases with "for" in literary texts ...... 221

Fig. 5.16 Means and standard error (± 1 SE) for 3rd person singular neuter pronominal cohesive device ("it") in newspaper texts ...... 227

Fig. 5.17 Means and standard error (± 1 SE) for 3rd person plural subjective pronominal cohesive device ("they") in newspaper texts...... 231

Fig. 5.18 Means and standard error (± 1 SE) for the total of 3rd person pronominal cohesive devices in newspaper texts ...... 233

Fig. 5.19 Means and standard error (± 1 SE) for possessive pronouns in newspaper texts

...... 236

Fig. 5.20 Means and standard error (± 1 SE) for definite article "the" in newspaper texts

...... 241

Fig. 5.21 Means and standard error (± 1 SE) for general comparative adjectives in newspaper texts ...... 245

Fig. 5.22 Means and standard error (± 1 SE) for superlative degree adverbs in newspaper texts ...... 246

Fig. 5.23 Means and standard error (± 1 SE) for the total number of reference cohesive devices in newspaper texts ...... 249

Fig. 5.24 Means and standard error (± 1 SE) for additive conjunction devices in newspaper texts ...... 253

Fig. 5.25 Means and standard error (± 1 SE) for adversative conjunction devices in newspaper texts ...... 255

Fig. 5.26 Means and standard error (± 1 SE) for causal and continuative conjunction devices in newspaper texts ...... 256

Fig. 5.27 Means and standard error (± 1 SE) for the total number of conjunction devices in newspaper texts ...... 257

xvi

Fig. 5.28 Means and standard error (± 1 SE) for total reference and conjunction cohesive devices in newspaper texts ...... 261

Fig. 5.29 Means and standard error (± 1 SE) for nominalization in newspaper texts .... 265

Fig. 5.30 Means and standard error (± 1 SE) for STTR in newspaper texts ...... 268

Fig. 5.31 Means and standard error (± 1 SE) for prepositional phrases with "of" in newspaper texts ...... 275

Fig. 5.32 Means and standard error (± 1 SE) for the total number of prepositional phrases in newspaper texts ...... 276

Fig. 5.33 Means and standard error (± 1 SE) for 3rd person singular neuter pronominal cohesive device ("it") in scientific texts ...... 283

Fig. 5.34 Means and standard error (± 1 SE) for the total of 3rd person pronominal cohesive devices in scientific texts ...... 286

Fig. 5.35 Means and standard error (± 1 SE) for possessive pronouns in scientific texts

...... 289

Fig. 5.36 Means and standard error (± 1 SE) for singular demonstratives in scientific texts

...... 292

Fig. 5.37 Means and standard error (± 1 SE) for definite article "the" in scientific texts296

Fig. 5.38 Means and standard error (± 1 SE) for comparative devices in scientific texts

...... 301

Fig. 5.39 Means and standard error (± 1 SE) for the total number of reference cohesive devices in scientific texts ...... 303

xvii

Fig. 5.40 Means and standard error (± 1 SE) for additive conjunction devices in scientific texts ...... 307

Fig. 5.41 Means and standard error (± 1 SE) for causal and continuative conjunction devices in scientific texts ...... 308

Fig. 5.42 Means and standard error (± 1 SE) for the sum of conjunction devices in scientific texts ...... 309

Fig. 5.43 Means and standard error (± 1 SE) for total reference and conjunction cohesive devices in scientific texts ...... 312

Fig. 5.44 Means and standard error (± 1 SE) for STTR in scientific texts ...... 316

Fig. 5.45 Means and standard error (± 1 SE) for average word length in scientific texts

...... 319

Fig. 5.46 Means and standard error (± 1 SE) for average sentence length in scientific texts

...... 321

Fig. 5.47 Means and standard error (± 1 SE) for passives in scientific texts ...... 323

Fig. 5.48 Means and standard error (± 1 SE) for prepositional phrases with "for" in scientific texts ...... 328

Fig. 5.49 Means and standard error (± 1 SE) for the use of general prepositions in scientific texts ...... 330

Fig. 5.50 Means and standard error (± 1 SE) for the use of the preposition "of" in scientific texts ...... 332

xviii

Fig. 5.51 Means and standard error (± 1 SE) for the total number of prepositional phrases in scientific texts ...... 335

xix

LIST OF TABLES

Table 2.1 Personal pronouns in English ...... 35

Table 2.2 Personal pronouns in Russian ...... 36

Table 2.3 Possessive pronouns in English ...... 36

Table 2.4 Possessive pronouns in Russian ...... 37

Table 2.5 Demonstrative cohesive devices in English ...... 38

Table 2.6 Demonstrative cohesive devices in Russian ...... 38

Table 2.7 Comparison in English...... 40

Table 2.8 Comparison in Russian ...... 41

Table 3.1 The combined framework for the study of the use of cohesive devices and other

global textual features (adapted from Dong and Lan 2010) ...... 82

Table 3.2 Descriptive statistics for the English-language corpus ...... 88

Table 3.3 Russian source texts and their corresponding human translations ...... 90

Table 3.4 English non-translated literary texts and their authors ...... 94

Table 3.5 Descriptive statistics for the LC...... 96

Table 3.6 Descriptive statistics for the NC ...... 98

Table 3.7 Descriptive statistics for the SC ...... 101

Table 3.8 Calculations used to track cohesive devices and textual features in the corpus

(adapted from Dong and Lan 2010) ...... 104

Table 3.9 CLAWS conventions for the tags used in the process of data collection ...... 105

Table 4.1 Association of 3rd person pronominal cohesive devices with the genre (Literary,

Newspaper, Scientific) for NT, HT, and MT by one-way ANOVA ...... 113

Table 4.2 Pairwise comparisons of 3rd person pronominal cohesive devices across genres

(Literary, Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing

...... 114

Table 4.3 Descriptive statistics for the total numbers of 3rd person pronominal cohesive devices across genres (N = 50) ...... 114

Table 4.4 Association of possessive pronouns with the genre (Literary, Newspaper,

Scientific) for NT, HT, and MT by one-way ANOVA ...... 117

Table 4.5 Pairwise comparisons of possessive pronouns across genres (Literary,

Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 117

Table 4.6 Descriptive statistics for the possessive pronouns across genres (N = 50) ..... 118

Table 4.7 Association of demonstratives with the genre (Literary, Newspaper, Scientific) for NT, HT, and MT by one-way ANOVA ...... 121

Table 4.8 Pairwise comparisons of demonstratives across genres (Literary, Newspaper,

Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 121

Table 4.9 Descriptive statistics for demonstratives across genres (N = 50) ...... 122

Table 4.10 Association of the definite article with the genre (Literary, Newspaper,

Scientific) for NT, HT, and MT by one-way ANOVA ...... 124

xxi

Table 4.11 Pairwise comparisons of the definite article across genres (Literary,

Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 125

Table 4.12 Descriptive statistics for the definite article across genres (N = 50) ...... 125

Table 4.13 Association of comparative devices with the genre (Literary, Newspaper,

Scientific) for NT, HT, and MT by one-way ANOVA ...... 129

Table 4.14 Pairwise comparisons of significant comparative devices across genres

(Literary, Newspaper, Scientific) for NT by Tukey HSD post hoc testing ...... 129

Table 4.15 Descriptive statistics for comparative devices across genres (N = 50)...... 129

Table 4.16 Association of reference cohesive devices with the genre (Literary,

Newspaper, Scientific) for NT, HT, and MT by one-way ANOVA ...... 132

Table 4.17 Pairwise comparisons of reference cohesive devices across genres (Literary,

Newspaper, Scientific) for NT and HT by Tukey HSD post hoc testing ...... 133

Table 4.18 Descriptive statistics for reference cohesive devices across genres (N = 50)

...... 133

Table 4.19 Association of conjunction cohesive devices with the genre (Literary,

Newspaper, Scientific) for NT, HT, and MT by one-way ANOVA ...... 136

Table 4.20 Pairwise comparisons of conjunction cohesive devices across genres (Literary,

Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 137

Table 4.21 Descriptive statistics for conjunction cohesive devices across genres (N = 50)

...... 137

xxii

Table 4.22 Association of nominalization with the genre (Literary, Newspaper,

Scientific) for NT, HT, MT by one-way ANOVA ...... 140

Table 4.23 Pairwise comparisons of nominalization across genres (Literary, Newspaper,

Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 141

Table 4.24 Descriptive statistics for nominalization across genres (N = 50) ...... 142

Table 4.25 Association of STTR with the genre (Literary, Newspaper, Scientific) for NT,

HT, and MT by one-way ANOVA ...... 145

Table 4.26 Pairwise comparisons of STTR across genres (Literary, Newspaper,

Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 145

Table 4.27 Descriptive statistics for STTR across genres (N = 50)...... 146

Table 4.28 Association of average word length with the genre (Literary, Newspaper,

Scientific) for NT, HT, and MT by one-way ANOVA ...... 149

Table 4.29 Pairwise comparisons of average word length across genres (Literary,

Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 149

Table 4.30 Descriptive statistics for average word length across genres (N = 50) ...... 150

Table 4.31 Association of average sentence length with the genre (Literary, Newspaper,

Scientific) for NT, HT, and MT by one-way ANOVA ...... 153

Table 4.32 Pairwise comparisons of average sentence length across genres (Literary,

Newspaper, Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 153

Table 4.33 Descriptive statistics for average sentence length across genres (N = 50) ... 154

xxiii

Table 4.34 Association of passives with the genre (Literary, Newspaper, Scientific) for

NT, HT, and MT by one-way ANOVA ...... 156

Table 4.35 Pairwise comparisons of passives across genres (Literary, Newspaper,

Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 157

Table 4.36 Descriptive statistics for passives across genres (N = 50) ...... 157

Table 4.37 Association of prepositions with the genre (Literary, Newspaper, Scientific)

for NT, HT, and MT by one-way ANOVA ...... 160

Table 4.38 Pairwise comparisons of prepositions across genres (Literary, Newspaper,

Scientific) for NT, HT, and MT by Tukey HSD post hoc testing ...... 161

Table 4.40 Differences in mean scores for the cohesive characteristics and global textual

features found to be significantly different at the 0.05 level for NT, HT, and MT ...... 165

Table 5.1 Association of 3rd person pronominal cohesive devices with the method of text

production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 177

Table 5.2 Pairwise comparisons of significant 3rd person pronominal cohesive devices

(NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 177

Table 5.3 Descriptive statistics for 3rd person pronominal cohesive devices in literary

texts (N = 50) ...... 178

Table 5.4 Association of possessive pronouns with the method of text production (NT,

HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 190

Table 5.5 Pairwise comparisons of possessive pronouns (NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 190

xxiv

Table 5.6 Descriptive statistics for possessive pronouns in literary texts (N = 50) ...... 190

Demonstratives are represented by two groups: ...... 192

Table 5.7 Association of demonstrative cohesive devices with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 192

Table 5.8 Pairwise comparisons of singular demonstrative cohesive devices (NT, HT, and

MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 193

Table 5.9 Descriptive statistics for demonstrative cohesive devices in literary texts (N =

50) ...... 193

Table 5.10 Association of definite article "the" with the method of text production (NT,

HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 196

Table 5.11 Pairwise comparisons of definite article "the" (NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 196

Table 5.12 Descriptive statistics for the use of the definite article "the" in literary texts (N

= 50) ...... 196

Table 5.13 Association of comparative cohesive devices with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 198

Table 5.14 Descriptive statistics for comparative devices in literary texts (N = 50) ...... 198

Table 5.15 Association of the total number of reference cohesive devices with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA

...... 199

xxv

Table 5.16 Pairwise comparisons of the total number of reference cohesive devices (NT,

HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 200

Table 5.17 Descriptive statistics for the total number of reference cohesive devices in literary texts (N = 50) ...... 200

Table 5.18 Association of conjunction cohesive devices with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 203

Table 5.19 Pairwise comparisons of conjunction cohesive devices (NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 203

Table 5.20 Descriptive statistics for conjunction cohesive devices in literary texts (N =

50) ...... 204

Table 5.21 Association of total reference and conjunction cohesive devices with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way

ANOVA ...... 208

Table 5.22 Pairwise comparisons of total reference and conjunction cohesive devices

(NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 209

Table 5.23 Descriptive statistics for total reference and conjunction cohesive devices in literary texts (N = 50) ...... 209

Table 5.24 Association of nominalization with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 211

Table 5.25 Descriptive statistics for nominalization in literary texts (N = 50) ...... 212

xxvi

Table 5.26 Association of STTR with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 212

Table 5.27 Pairwise comparisons of STTR (NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 213

Table 5.28 Descriptive statistics for STTR in literary texts (N = 50) ...... 213

Table 5.29 Association of average word length with the method of text production (NT,

HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 216

Table 5.30 Descriptive statistics for average word length in literary texts (N = 50) ...... 216

Table 5.31 Association of average sentence length with the method of text production

(NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 217

Table 5.32 Descriptive statistics for average sentence length in literary texts (N = 50) 217

Table 5.33 Association of passives with the method of text production (NT, HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 218

Table 5.34 Descriptive statistics for passives in literary texts (N = 50) ...... 218

Table 5.35 Association of prepositional phrases with the method of text production (NT,

HT, and MT) in the corpus of literary texts by one-way ANOVA ...... 219

Table 5.36 Pairwise comparisons of prepositional phrases with "for" (NT, HT, and MT) for the corpus of literary texts by Tukey HSD post hoc testing ...... 220

Table 5.37 Descriptive statistics for prepositional phrases in literary texts (N = 50) ..... 220

Table 5.38 Association of 3rd person pronominal cohesive devices with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA 225

xxvii

Table 5.39 Pairwise comparisons of significant 3rd person pronominal cohesive devices

(NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing . 225

Table 5.40 Descriptive statistics for 3rd person pronominal cohesive devices in newspaper texts (N = 50) ...... 226

Table 5.41 Association of possessive pronouns with the method of text production (NT,

HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 234

Table 5.42 Pairwise comparisons of possessive pronouns (NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 235

Table 5.43 Descriptive statistics for possessive pronouns in newspaper texts (N = 50) 236

Table 5.44 Association of demonstrative cohesive devices with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA 238

Table 5.45 Descriptive statistics for demonstrative cohesive devices in newspaper texts

(N = 50) ...... 238

Table 5.46 Association of definite article "the" with the method of text production (NT,

HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 240

Table 5.47 Pairwise comparisons of definite article "the" (NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 240

Table 5.48 Descriptive statistics for the use of the definite article "the" in newspaper texts

(N = 50) ...... 240

Table 5.49 Association of comparative cohesive devices with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ...... 243

xxviii

Table 5.50 Pairwise comparisons of general comparative adjective cohesive devices (NT,

HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 244

Table 5.51 Descriptive statistics for comparative devices in newspaper texts (N = 50) 244

Table 5.52 Association of the total number of reference cohesive devices with the method

of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way

ANOVA ...... 248

Table 5.53 Pairwise comparisons of the total number of reference cohesive devices (NT,

HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 248

Table 5.54 Descriptive statistics for the total number of reference cohesive devices in

newspaper texts (N = 50) ...... 249

Table 5.55 Association of conjunction cohesive devices with the method of text

production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA 251

Table 5.56 Pairwise comparisons of conjunction cohesive devices (NT, HT, and MT) for

the corpus of newspaper texts by Tukey HSD post hoc testing ...... 251

Table 5.57 Descriptive statistics for conjunction cohesive devices in newspaper texts (N

= 50) ...... 252

Table 5.58 Association of the total reference and conjunction cohesive devices with the

method of text production (NT, HT, and MT) in the corpus of newspaper texts by one- way ANOVA ...... 259

Table 5.59 Pairwise comparisons of the total reference and conjunction cohesive devices

(NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing . 260

xxix

Table 5.60 Descriptive statistics for total reference and conjunction cohesive devices in newspaper texts (N = 50) ...... 260

Table 5.61 Association of nominalization with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA...... 263

Table 5.62 Pairwise comparisons of nominalization (NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 264

Table 5.63 Descriptive statistics for nominalization in newspaper texts (N = 50) ...... 264

Table 4.64 Association of STTR with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 267

Table 4.65 Pairwise comparisons of STTR (NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing...... 267

Table 4.66 Descriptive statistics for STTR in newspaper texts (N = 50) ...... 267

Table 5.67 Association of average word length with the method of text production (NT,

HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 269

Table 5.68 Descriptive statistics for average word length in newspaper texts (N = 50) 269

Table 5.69 Association of average sentence length with the method of text production

(NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA...... 270

Table 5.70 Descriptive statistics for average sentence length in newspaper texts (N = 50)

...... 270

Table 5.71 Association of passives with the method of text production (NT, HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 271

xxx

Table 5.72 Descriptive statistics for passives in newspaper texts (N = 50) ...... 271

Table 5.73 Association of prepositional phrases with the method of text production (NT,

HT, and MT) in the corpus of newspaper texts by one-way ANOVA ...... 272

Table 5.74 Pairwise comparisons of prepositional phrases with "of" (NT, HT, and MT) for the corpus of newspaper texts by Tukey HSD post hoc testing ...... 273

Table 5.75 Descriptive statistics for prepositions in newspaper texts (N = 50) ...... 274

Table 5.76 Association of 3rd person pronominal cohesive devices with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ... 280

Table 5.77 Pairwise comparisons of significant 3rd person pronominal cohesive devices

(NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ... 281

Table 5.78 Descriptive statistics for 3rd person pronominal cohesive devices in scientific texts (N = 50) ...... 282

Table 5.79 Association of possessive pronouns with the method of text production (NT,

HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 287

Table 5.80 Pairwise comparisons of possessive pronouns (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 288

Table 5.81 Descriptive statistics for possessive pronouns in scientific texts (N = 50) ... 288

Table 5.82 Association of demonstrative cohesive devices with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ... 290

Table 5.83 Pairwise comparisons of singular determiners (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 290

xxxi

Table 5.84 Descriptive statistics for demonstrative cohesive devices in scientific texts (N

= 50) ...... 291

Table 5.85 Association of definite article "the" with the method of text production (NT,

HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 295

Table 5.86 Pairwise comparisons of definite article "the" (NT, HT, and MT) for the

corpus of scientific texts by Tukey HSD post hoc testing ...... 295

Table 5.87 Descriptive statistics for the use of the definite article "the" in scientific texts

(N = 50) ...... 295

Table 5.88 Association of comparative cohesive devices with the method of text

production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ... 299

Table 5.89 Pairwise comparisons of comparative cohesive devices (NT, HT, and MT) for

the corpus of scientific texts by Tukey HSD post hoc testing ...... 299

Table 5.90 Descriptive statistics for comparative devices in scientific texts (N = 50) ... 300

Table 5.91 Association of the total number of reference cohesive devices with the method

of text production (NT, HT, and MT) in the corpus of scientific texts by one-way

ANOVA ...... 302

Table 5.92 Pairwise comparisons of the total number of reference cohesive devices (NT,

HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 302

Table 5.93 Descriptive statistics for the total number of reference cohesive devices in

scientific texts (N = 50) ...... 302

xxxii

Table 5.94 Association of conjunction cohesive devices with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ... 305

Table 5.95 Pairwise comparisons of conjunction cohesive devices (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 305

Table 5.96 Descriptive statistics for conjunction cohesive devices in scientific texts (N =

50) ...... 306

Table 5.97 Association of reference and conjunction cohesive devices with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA

...... 311

Table 5.98 Pairwise comparisons of reference and conjunction cohesive devices (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 312

Table 5.99 Descriptive statistics for total reference and conjunction cohesive devices in scientific texts (N = 50) ...... 312

Table 5.100 Association of nominalization with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 314

Table 5.102 Association of STTR with the method of text production (NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 315

Table 5.103 Pairwise comparisons of STTR (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 316

Table 5.104 Descriptive statistics for STTR in scientific texts (N = 50) ...... 316

xxxiii

Table 5.105 Association of average word length with the method of text production (NT,

HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 317

Table 5.106 Pairwise comparisons of average word length (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 318

Table 5.107 Descriptive statistics for average word length in scientific texts (N = 50) . 318

Table 5.108 Association of average sentence length with the method of text production

(NT, HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 320

Table 5.109 Pairwise comparisons of average sentence length (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 320

Table 5.110 Descriptive statistics for average sentence length in scientific texts (N = 50)

...... 321

Table 5.111 Association of passives with the method of text production (NT, HT, and

MT) in the corpus of scientific texts by one-way ANOVA ...... 322

Table 5.112 Pairwise comparisons of passives (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 322

Table 5.113 Descriptive statistics for passives in scientific texts (N = 50) ...... 323

Table 5.114 Association of prepositional phrases with the method of text production (NT,

HT, and MT) in the corpus of scientific texts by one-way ANOVA ...... 326

Table 5.115 Pairwise comparisons of significant prepositional phrases (NT, HT, and MT) for the corpus of scientific texts by Tukey HSD post hoc testing ...... 326

Table 5.116 Descriptive statistics for prepositional phrases in scientific texts (N = 50) 327

xxxiv

Table 6.1 Differences in mean scores for cohesive characteristics and global textual features in the literary corpus ...... 339

Table 6.2 Pairwise comparisons of significant cohesive and global textual characteristics

(NT, HT, and MT) by Tukey HSD post hoc testing in the literary corpus ...... 341

Table 6.3 Differences in mean scores for the cohesive characteristics and global textual features in the newspaper corpus ...... 349

Table 6.4 Pairwise comparisons of significant cohesive and global textual characteristics

(NT, HT, and MT) by Tukey HSD post hoc testing in the newspaper corpus ...... 352

Table 6.5 Differences in mean scores for the cohesive characteristics and global textual features in the scientific corpus...... 362

Table 6.6 Pairwise comparisons of significant cohesive and global textual characteristics

(NT, HT, and MT) by Tukey HSD post hoc testing in the scientific corpus ...... 364

xxxv

DEDICATION

To my son Matthew William McIntyre,

Who was born when this dissertation was being completed.

xxxvi

ACKNOWLEDGEMENTS

I would like to express my deepest gratitude to my long-time advisor, mentor, and colleague Dr. Brian James Baer, whose guidance and encouragement have helped me through the years of the journey called "graduate school"—from the beginning of my

Master's program in Translation to the end of my doctoral studies. His wise teachings and

continuous support, combined with the diverse and stimulating environment of the

Institute of Applied Linguistics at Kent State University, have helped me grow as a

scholar, translator, teacher, and person.

I would like to thank the members of my dissertation committee Dr. Gregory

Shreve, Dr. Erik Angelone, Dr. Andrew Barnes, and Dr. Mikhail Nesterenko for their

thoughtful review of my dissertation and insightful comments during the defense. Their

suggestions helped me to fine-tune this dissertation.

My sincere gratitude goes to the entire faculty and staff of the Department of

Modern and Classical Language Studies for their dedication to the success of their students in our relatively young but bold doctoral program.

I would also like to thank Dr. Alexander Alekseenko for providing invaluable advice on statistics, and my supportive friend Nataliya Alekseenko for steering me towards her, back then, fiancé Alexander when I was in need of such advice.

xxxvii

I would like to thank my parents Yury and Maria Pastushenkov, as well as my

grandparents on both sides of my family, for instilling values of life-long learning in me.

Their great scholarly and pedagogical achievements have inspired my research and teaching career.

A special thank-you goes to my wonderful husband Bill McIntyre, whose love, encouragement, and wisdom have made the successful completion of this dissertation possible.

Finally, I would like to thank everyone else who cared about my research, teaching, and life—my friends all around the world, my academic and translation colleagues, my supportive students, and my up-and-coming linguist brother Dmitry.

Tatyana Bystrova-McIntyre

November 15, 2012

Kent, OH

xxxviii

CHAPTER 1: INTRODUCTION

This dissertation compares the use of cohesive devices and other global textual features

across corpora of three groups of texts produced by different methods: texts written in

English (non-translated texts), texts translated into English from Russian by human translators (human-translated texts), and texts translated into English from Russian by a machine translation tool (machine-translated texts). This study looks at global features of these texts in corpora of three genres: literary, newspaper, and scientific.

I use the term global to refer to textual features, that is, those features that appear throughout an entire text, as opposed to features that appear only at a sub-sentential level.

Cohesive devices are an important subset of such features; they tie textual segments together into a whole by creating "connections among the elements within the discourse"

(Campbell 1995: 5-6). The present study looks at reference and conjunction cohesive devices as identified by the pioneers of cohesion studies Michael Alexander Kirkwood

Halliday and Ruqaiya Hasan (1976). Detailed accounts of these devices and of the notion of cohesion are provided in Chapters 2 and 3. Other global textual features included in this study are nominalization, lexical density, average word length, average sentence length, passives, and prepositional phrases. These features are defined and discussed in

Chapter 2.

This empirical study applies a more comprehensive research approach than has been used earlier for analyzing global features of texts. It sheds light on global characteristics of translated texts—whether and how human and machine translations differ from non-translated texts—contributing to existing studies of translation universals and laws of translation. Furthermore, by including machine-translated texts as a third dimension of research, in addition to typically studied non-translated and human- translated texts, this study addresses questions of whether translation universals and laws of interference are at work in machine-translated texts to the same extent as in human- translated texts. The results of this study are then used to design practical recommendations for translators, editors of human and machine translations, translation evaluators, and translator trainers.

This chapter presents background for this research, discusses its significance, goals, and research questions, and outlines its structure.

1.1 Background

1.1.1 Global textual features and translation: Text as a culture-specific phenomenon

Text organization and global textual features as its core are important components of successful writing, and have been of interest to linguists for a long time. How do writers or speakers represent their envisioned scenes and stories in a linear mode of text? How do readers or listeners decode such representations in the process of reading? Additionally—

and more importantly for the purposes of this study—how is text organization different

across cultures?

In 1960s, Robert Kaplan introduced the term "contrastive rhetoric." While working in an English as a Second Language program, he observed that the writing of

learners of English as a second language often displayed features of global text

organization that seemed flawed to native speakers of English. Related to this

observation, Kaplan also noticed similarities in global textual features in groups of

international students whose native languages were the same or similar (Mohamed-

Sayidina 1993: 7).

These observations became the basis of Kaplan's argument that discourse

organization is a phenomenon that differs from culture to culture (Mohamed-Sayidina

1993: 7). Kaplan suggested that international students writing in English transfer culture-

specific patterns of textual organization from their native languages into their English

writing, echoing the Sapir-Whorf hypothesis of linguistic relativity:

[T]he organization of a paragraph, written in any language by any individual who

is not a native speaker of that language, will carry the dominant imprint of that

individual's culturally-coded orientation to the phenomenological world in which he lives

and which he is bound to interpret largely through the avenues available to him in his

native language (Kaplan 1972: 1).

Consonant with Kaplan's contrastive rhetoric and other related theories, a number

of translation scholars have pointed out that cohesive devices and other global features of text organization are language and text-type specific (Hatim and Mason 1990, Baker

1992, Blum-Kulka 2000, Teich and Fankhauser 2004, among others). In their work, translators have to deal with these different specificities, adjusting cohesive and other

global textual features in the target text so as to re-situate the source text in a different

culture.

The language- and culture-specificity of cohesive and other global textual features raises a number of questions related to translation. To what extent do translators adjust global textual features in their work? To what extent do translations reflect the unique textual features of their source texts? To what extent does any of this influence readers' comprehension or interfere with how translations are received in the target culture? How does machine translation cope with culture-specific global features of texts? Do beginning translators differ from expert translators in their approaches to translating global features of text organization?

The present study aims to answer some of these questions by illuminating similarities and differences in the use of cohesive and other global features across non- translated, human-translated, and machine-translated texts. It looks at thirty-two cohesive and other textual parameters across these three methods of text production and in corpora of three genres (literary, newspaper, and scientific). The premise of this study is that although cohesive devices and other global textual features have been understudied in

empirical translation research, they are vitally important in the practice and training of translators and editors of human-translated and machine-translated output.

1.1.2 New realities of the translation industry today

The translation industry today is very different from that of the recent past. It is rapidly changing with the increased demands of globalization and swift growth of Internet and information technologies. Research in translation studies must keep up with these changes, focusing on new realities of the translation industry, such as machine translation, computer-assisted translation tools, and more. For this reason, the present study includes machine-translated output as a part of its design and looks at similarities and differences in the use of cohesive devices and other global features across three groups of texts—non-translated, human-translated, and machine-translated. In addition, the study includes three genres, including two non-literary ones (newspaper and scientific).

The sub-sections below deal with the issues of machine translation, (post-)editing, and translation memory tools as realities of the translation industry that contribute to the need to study cohesive and other global textual features of non-translated, human- translated, and machine-translated texts.

1.1.2.1 Machine translation as a reality of modern translation practice

Machine translation (MT) has been around since the 1960s, when it was expected to soon replace human translation. This proved an overly optimistic hope, and research in MT

slowed down. Since the late 1980s, with the increased power and decreased price of

computers, MT development is once again on the rise (Hutchins 1995: 433-437).

The growth of online communication, as well as the demands of businesses targeting customers abroad, have contributed to the need for more and faster translation in general, giving a new impetus to MT in particular. For example, Internet users often want new electronic content to be translated instantaneously and at no cost—exactly what such MT tools as GoogleTranslate and Yahoo! BabelFish provide (Gaspari and Hatchins

2007: 200-201). With developments in statistical machine translation and an increased ability to build very large electronic bilingual corpora needed for it, such entities as

Google are constantly improving their MT tools.

Translators should not necessarily fear for their fate. In fact, they might celebrate—according to the U.S. Bureau of Labor Statistics, "employment of interpreters and translators is projected to increase 22 percent over the 2008–18 decade, which is much faster than the average for all occupations." However, ignoring MT is no longer an option. MT has become an integral part of many translators' work. It is now included as a part of major translation tools, such as SDL Trados Studio. Many Language Service

Providers (LSPs) and government organizations hire translators to post-edit MT output, which is often considered to be a faster and cheaper way to provide translations. For instance, SDL, a major LSP, markets its "intelligent machine translation," which integrates its customized MT technology, SDL LanguageWeaver, with human post-

editing skills and is claimed to significantly reduce companies' translation budgets and

production times.

Given the explosive growth of MT in recent years, it would seem relevant to

supplement the studies of global features in human-translated texts by similar studies of

MT output. The present study does this. It supplements more traditional studies of human-translated and non-translated texts, including both machine- and human-translated outputs for the Russian source texts studied.

In fact, the inclusion of machine-translated texts into the study constitutes a relative novelty. While a number of studies have shown that translated texts exhibit certain peculiar characteristics when compared with non-translated texts (Blum-Kulka

1986, Eskola 2004, Mauranen 2000, Olohan and Baker 2000, Puurtinen 1995, Schaeffner

& Adab 2001, Tirkkonen-Condit 2004, among others), translations performed by machines have not been included in the majority of these studies. Only recently have researchers started including machine-translated texts into their studies (Khanna et al.

2011, Volansky et al. 2011). With MT on the rise, analyzing machine-translated texts may prove beneficial for training future translators to work with MT and for assessing the quality of MT output. It might also help identify MT output in the first place.

Not surprisingly, recent research does find differences between machine- translated and human-translated texts. In their study of English non-translated texts and

Hebrew into English human-translated and machine-translated texts, Volansky et al.

(2011) find that these three groups of texts display differences in such features as average

sentence and word length, passive-verb ratio, type-token ratio, punctuation, etc. They use

this information to design a system that would automatically differentiate between

English non-translated texts, texts translated into English by human translators, and texts

translated into English by MT tools.

Volansky et al. (2011) point out that "better awareness of the translation process

can improve the quality of statistical machine translation (STM) systems," citing works

by Gamon et al. (2005) and Kurokawa et al. (2009) (2). While the present study does not

aim at improving STM directly, its methodology and results might prove useful for this

purpose as well.

Some studies that look at machine-translated texts find that while they differ from human-translated texts, MT may be a reasonable (and free of charge) alternative to human translation, at least for some text-types. For instance, Khanna et al. (2011), who look at medical translations produced by professional translators vs. machines, find that while machine-translated sentences had significantly lower fluency (grammatical

correctness) scores compared with professional translation, there are similarities in terms

of adequacy (information preservation) and meaning (connotation maintenance) scores.

The present study might help illuminate more similarities and differences between non-

translated, human-translated, and machine-translated texts in view of cohesion and other

textual features, and provide insight into the work of translators and editors.

1.1.2.2 Editing and post-editing in the translation industry today

Editing is another translation field that may benefit from the results of this study insofar

as cohesion and text organization are vital to editing texts, especially at the macro-textual level. Brian Mossop, in his book Revising and Editing for Translators, talks about

smoothness (his own term for cohesion) as one of the twelve main revision parameters to

be considered by editors, noting that a lack of smoothness may occur due to "careless

imitation of the word order or the connector words … of the source text" (2001: 106).

With the use of MT skyrocketing, the demand for professional translators who are

experienced in post-editing MT output has been growing. Some studies show that, at least

for some text-types, post-editing MT output may be faster than translating from scratch.

For example, in the study by Carl et al. (2011), translation speeds for MT plus post-

editing are found to be, on average, faster than for human translation. The researchers

also report a modest increase in translation quality for post-edited output.

Demand for editors and post-editors clearly exists in today's translation market, as has been indicated by the results of my study of two hundred ProZ.com job postings, collected on ten consecutive days in September-October, 2010. According to the analysis of the services required, while translations were called for in the majority of the postings

(61.5%), checking of translations was the second service in this list (with 22.5% of all postings), and editing (including post-editing and copy-editing) came in third (with 19%)

(Bystrova-McIntyre 2010).

The results of the ProZ.com survey support Mossop's claim that editing skills may

help translation graduates find translation-related work in technical writing or editing,

edit their own translations, and deal with poorly written source texts (2001: iv), thus

making translation students more marketable and versatile in the job market. To the

extent that cohesion is among the major revision parameters for editors of translations

(Mossop 2001: 107), translation trainees should be familiar with cohesion as it relates to

translation and the editing of one’s own or other translators’ work. Nonetheless, the role

that cohesion and other textual characteristics play in the construction of well-written texts is often underestimated. For instance, out of five leading textbooks on monolingual editing (Baskette, Sissors, and Brooks 1992, Berner 1991, Dragga and Gong 1989,

Harrigan 1993, Samson 1993), only one—Dragga and Gong's Editing: The Design of

Rhetoric (1989)—listed cohesion in its index.

1.1.2.3 Translation memory tools and their effect on translators' work

The ability to work with translation memories (TMs)—databases that store segments of aligned source and target content—is a must in the translation industry today. Among main benefits of TMs are improved consistency of translated content, reduced cost

(especially for large-scale projects), and increased speed.

One can conjecture that the use of TMs, which forces translators to function more at the level of shorter textual segments, rather than at the level of the text as a whole, has an effect on the global textual features of translations. While there appear to be no studies

that directly measure differences in cohesion and other textual features in translations involving the use of translation tools vs. those that do not, multiple studies indicate that translation tools influence the process and product of translation (Dragsted 2005, O'Brien

2006a, 2006b, Guerberof 2008, Bowker 2005, Ribaz López 2007, Wallis 2006, among others). For instance, Dragsted (2004, 2006) finds that the segmentation rules used by

TM software "do not correlate well with the way expert human translators 'chunk' a source text" (Garcia 2009: 27)—a finding that might influence global-level features of translations. Pym notes, referring to Dragsted's research of 2004, that "the segmentation imposed by the software makes rather more impositions than authorial segmentation ever did" (2007: 7).

The proposed research adds to the array of studies in this area, but emphasizes differences in cohesive and other global textual features across human-translated, machine-translated, and non-translated texts. Its results may help design better policies and training for today's translation industry, where global teams of linguists extensively work with TM and MT tools.

1.2 Statement of the problem

Gideon Toury (1995) argues that the translation process is influenced by interference on the part of the source text. Additionally, scholars claim that translations tend to exhibit unique characteristics regardless of the source language, a phenomenon known as

"translation universals" (Baker 1996). Following this, it may be hypothesized that the

method by which a text is produced (non-translated, human-translated, or machine- translated) will be reflected in the use of cohesive devices and other global textual features. It may also be hypothesized that the text-type will influence the use of such devices.

These influences, however, have not been widely studied across large collections of texts. Moreover, few comprehensive methodologies have been applied to the empirical

study of such influences. In view of recent developments in the language industry

outlined above, more research and improved methodologies are needed to inform the

work of modern translators, editors of human and machine translations, translation

evaluators, and translator trainers. The present study suggests and tests a more comprehensive methodology for studying cohesive and other global features of texts, aiming to illuminate such features across genres in English non-translated, human- translated, and machine-translated texts.

1.3 Goals and components of this study

This study aims to illuminate the use of cohesive devices and other global features in texts produced by three different methods:

1. Texts translated by human translators from Russian into English (human-

translated texts)

2. Texts translated by MT tools from Russian into English (machine-translated texts)

3. Texts written in English (non-translated texts)

It looks at cohesive and other global textual features across three text-types:

1. Literary texts

2. Newspaper texts

3. Scientific texts

The goals of this study are as follows: (1) to distinguish global features of non- translated, human-translated, and machine-translated texts; (2) to contribute to the arsenal of research methodologies available to translation scholars, and (3) to design recommendations for translators, editors of machine-translated output, and trainers.

Consistent with its goals, this study has both a discovery component and a hypothesis-testing component. In addition, its methodology is designed to be more comprehensive than others that have come before.

1. Discovery component. The study produces new data on the use of cohesive devices and other textual features in human-translated, machine-translated, and non- translated texts, as well as across text-types. It uses computer tools to study thirty-two variables related to cohesion and global textual features (outlined in Chapter 3) in a relatively large corpus of texts, revealing phenomena not visible to the naked eye. This new knowledge may serve as a more solid basis for formulating future hypotheses in this area of translation research. It will inform the work translators, editors of human and machine translations, translation evaluators, and trainers. In addition, these data may also be useful for researchers looking to programmatically determine if a text is a translation.

2. Hypothesis-testing component. The study aims to test one main hypothesis concerning the groups and types of texts involved:

The method of text production (non-translated, human-translated, or machine-

translated) influences the use of cohesive devices and other global textual features

in a specific text-type.

3. Improved methodology. The study tests a more comprehensive methodology than has been used in earlier studies for analyzing cohesion and global textual features in translated and non-translated texts. Empirical translation studies need solid methodologies designed and/or tuned specifically to answer their unique research questions, and this study offers one such methodology for future use by other translation scholars.

1.4 Research questions

This study addresses three major research questions:

1. Does the way texts are produced (non-translated, human-translated, or machine-

translated) influence the use of cohesive devices and other global textual features

in literary texts for the Russian-into-English language pair, and if so, how?

2. Does the way texts are produced (non-translated, human-translated, or machine-

translated) influence the use of cohesive devices and other global textual features

in newspaper texts for the Russian-into-English language pair, and if so, how?

3. Does the way texts are produced (non-translated, human-translated, or machine-

translated) influence the use of cohesive devices and other global textual features

in scientific texts for the Russian-into-English language pair, and if so, how?

1.5 Organization of this study

This dissertation consists of six chapters. Chapter One provides a background and introduction to the study, states its research problem, and outlines the purposes and components of this study. The major research questions are also formulated in this chapter. Chapter Two offers a literature review, discussing and defining the key concepts employed in the study and outlining existing research in global textual features in both non-translation and translation disciplines. Chapter Three describes the methods and materials used in this research. Chapter Four provides an overview of text-type specificities for literary, newspaper, and scientific texts based on the data analyzed in this study. Chapter Five reports on the analysis of the cohesive and other textual features included in the study. The data and analysis for each corpus are reported in individual sections of the chapter. Chapter Six provides a summary of the results, discusses implications of the results for those working in the language industry and training future linguists, provides practical recommendations for translators and (post-)editors, and lists limitations of the study. It also outlines future research directions suggested by this study.

CHAPTER 2: LITERATURE REVIEW

This chapter provides an overview of the literature pertaining to studies of cohesion and

other global textual features in both non-translation and translation contexts. It outlines

the main concepts related to differences across translated and non-translated texts, and

discusses definitions of cohesion and other global textual features under study. Since

translation studies draws extensively on research in other fields, such as general

linguistics, psycholinguistics, and anthropology, to name but a few, the chapter reviews

existing research on cohesion and other global features of texts in both monolingual and multilingual settings.

2.1 Non-translated vs. translated texts

This section provides an overview of the research on differences between translated and non-translated texts. It discusses the concept of translation laws and gives an overview of the research into translation universals. Additionally, the section touches upon the common methodology of research into characteristics of translated and non-translated

texts. In conclusion, it discusses some potential applications of this research.

2.1.1 The notion of translationese

As noted above, research in translation studies indicates that there are differences

between translated and non-translated texts (Gellerstan 1986, Blum-Kulka 1986, Eskola 16

2004, Mauranen 2000, Olohan and Baker 2000, Puurtinen 1995, Schaeffner & Adab

2001, Hansen 2003, Tirkkonen-Condit 2004, among others). Researchers find differences

in vocabulary use, grammatical features, and other textual characteristics, as discussed in more detail below. Gellerstam (1986), whose study finds differences between non- translated Swedish texts and English-into-Swedish translations, uses the term

"translationese" to refer to these peculiar characteristics of translated texts. Olohan (2004) defines translationese as "a common description for translated language that appears to be influenced by the source language, usually in an inappropriate way or to an undue extent"

(90).

While the term is used by and large pejoratively (Olohan 2004: 90), such influences from the source text do not necessarily have to be viewed as negative. Toury

(1995), whose general laws of translation are discussed below, mentions that such discourse transfer can be both negative and positive (Routledge Encyclopedia of

Translation Studies 1998: 290). Indeed, as the following overview of pertinent research

shows, the discussion of such peculiarities of translated texts typically does not involve their condemnation.

This study looks at differences in the use of cohesive and other global textual features between non-translated, human-translated, and machine-translated texts, and

does not employ the term "translationese" to refer to such differences due to its overall

pejorative connotation. Rather than refer to a set of criteria to determine quality, the

assumption I make in this study is that the closer a translated text approaches the style of

a non-translated text, the more it conforms with the norms of the target culture. This

proximity to target culture norms is how I understand quality in the context of this study.

2.1.2 General laws of translation

Many translation researchers view differences between translated and non-translated texts as proof of the existence of general laws of translation. In fact, Gideon Toury (1995) maintains that one of the main goals of descriptive translation studies is to discover such general laws of translation and understand the norms involved in the translation process.

As mentioned above, Toury sets forth two general laws of translation: the aforementioned law of interference and the law of growing standardization. The law of interference leads to translationese, or "fingerprints" of the source language in the target text: "in

translation, phenomena pertaining to the make-up of the source text tend to be transferred

to the target text" (1995: 275). According to the law of growing standardization, "the

special textual relations created in the source text are often replaced by conventional

relations in the target text, and sometimes they are ignored altogether" (Routledge

Encyclopedia of Translation Studies 1998: 290).

According to Volanski et al. (2011), the "combined effect of these laws creates a

… text that partly corresponds to the source text and partly to texts written originally in

the target language but in fact is neither of them" (2). Frawley (1984) refers to such texts

as "hybrids," a term that is later revisited by Adab and Schaeffner (2001) and other

scholars. The hybrid nature of translated texts may affect cohesive and other global textual features.

2.1.3 Interference of the source language vs. translation universals

Some researchers trace unique features of target texts (TT) to their source texts' (ST) language (cf. Toury's law of interference). For instance, Johansson and Hofland (2000), in their study of English and Norwegian modal particles, find that particles used in translation are influenced by the source text. Gellerstam (1996) explains the overuse of the verb tillbringa in texts translated into Swedish by the fact that translators have to render the English verb "spend."

Other researchers claim that translations exhibit certain characteristics regardless of the source language (Baker 1996). These universal characteristics are known as translation universals, "one of the most intriguing and controversial topics in recent translation studies," as it was put in John Benjamin's marketing summary of the book

Translation Universals: Do They Exist? Mona Baker (1996) lists four categories of such features of translated texts: explicitation ("the tendency to spell things out in translation"), simplification ("the idea that translators subconsciously simplify the language or message"), normalization ("the tendency to conform to patterns and practices that are typical of the target language, even to the point of exaggerating them"), and leveling out ("the tendency of translated text to gravitate around the centre of any continuum rather than move towards the fringes") (Baker 1996: 176-177).

Explicitation, in particular, is prominent in research on translation universals. For instance, Vanderauwera (1985), in a study of Dutch-into-English translations of novels, finds that translators add cultural information to aid TT reader comprehension, repeat lexical items to avoid ambiguity, and fill in elliptical structures to make translation clearer. Blum-Kulka's explicitation hypothesis (1986) holds that translations may be more redundant than their source texts. Øverås (1998) finds explicitation shifts in Norwegian-

English and English-Norwegian translations, with lexical explicitation occurring more frequently than grammatical explicitation. Olohan and Baker (2000) analyze the use of the optional 'that' in translated and non-translated texts, reporting that it is more frequent in translated texts. They use this finding to support the notion of explicitation in translation.

Simplification is a focus of another group of studies in translation universals.

Laviosa, based on her extensive research into differences between translated and non- translated corpora (1998a, 1998b, 2002), finds that translated texts are characterized by a narrower range of vocabulary, a lower ratio of lexical to running words, and a lower average sentence length. In addition, translated texts tend to under-represent linguistic items of the target language that do not have direct equivalents in the source language

(Tirkkonen-Condit 2004).

Normalization is, in a way, related to simplification, and its existence is supported by several studies. Vanderauwera (1985) suggests that translators may break up sentences or omit certain information in an attempt to meet target language standards. May (1997),

in her study of French and Russian translations of Virginia Wolf and William Faulkner, argues that translators normalize punctuation, making it more simple. Øverås (1998) discovers shifts in collocation and metaphorical patterns from unusual to more usual in the target language. Kenny's study of hapax legomena (words that occur only once in a text) finds that translators tend to normalize them (2002).

Finally, leveling out has not been given much attention (Baker 1996: 184). As

Olohan points out, it also "appears to pose [the] most difficulty for measurement" (100).

Olohan suggests looking for reduced variance in a corpus of translated and non-translated texts in order to identify the presence of leveling out. Some studies, however, have not found this to be the case. For instance, Laviosa (1998b) reports low variance for both translated and non-translated fictional texts in terms of lexical density and proportion of high-frequency words, and higher variance in translated texts in terms of sentence length.

The present study hypothesizes that translation universals and Toury's laws of interference may account for differences in the use of cohesive and other global textual features across non-translated, human-translated, and machine-translated texts. The study also raises a question of whether the notion of translation universals is pertinent to machine-translated output, hypothesizing that such universals may be a product of the human brain.

2.1.4 Methodology of studies involving translated and non-translated texts

Most studies dealing with characteristic features of translation employ methods from corpus linguistics and use monolingual translational corpora, compiled from texts translated into a certain language (typically English), and comparable (non-translated) corpora of texts originally written in that language (e.g., Laviosa 1998a and 1998b,

Olohan and Baker 2000, among others). With such corpora, researchers investigate the use of certain lexical or grammatical items (e.g., particular words, type-token ratio, nominalization, passive forms, prepositional phrases) or other features of texts (e.g., length of words, sentences, or paragraphs) in translated and non-translated texts. Some researchers (e.g., Tirkkonen-Condit 2002) use such corpora with human participants, with the goal to determine whether readers can differentiate between translated and non- translated texts.

2.1.5 Applicability of studies involving translated and non-translated texts

In addition to shedding light on the nature of translated and non-translated texts, studies

such as this one might inform the work of translators, trainers and trainees, and

translation evaluators. It may also be useful for researchers seeking to differentiate

between translated and non-translated texts (Baroni et al. 2006). Automated methods of

indentifying translated texts could be a useful (self-) assessment tool for those involved in

translation (cf. Baroni et al. 2006), as well as for researchers who extract parallel corpora

from the web, when potential parallel texts for inclusion need to be identified and

assessed (Resnik et al. 2003). Baroni et al. also note that automated techniques of

distinguishing between translated and non-translated texts might be useful in identifying

multicultural plagiarism.

2.2 Cohesion, global textual features, and related terms

This section discusses and defines concepts and terms related to cohesion and other global textual features. It starts off with an overview of text and discourse organization, as well as such related concepts as "genre" and "text-type." Further, it provides a thorough discussion of different definitions of cohesion and its relation to other linguistic terms, such as texture, coherence, and standards of textuality. Also included is a detailed outline of various categories of cohesion as set forth by Halliday and Hasan, as well as other researchers' viewpoints on the issue. In addition to cohesion, other textual features incorporated within the framework of the present study are discussed and defined. These include nominalization, lexical density, average word length, average sentence length, passives, and prepositional phrases.

2.2.1 Text organization and discourse analysis

Cohesion and other global features that occur at the level of text rather than isolated words and phrases at the subsentential level are key attributes of text organization. Text organization refers to ways that producers of written and oral communication structure information within a construct that "form[s] a totality … [with] its own characteristics"

(Caron 1992: 153). Such a "totality" is often referred to as "discourse"—the term that

covers communication in its situational and social contexts. The term "organization" can be defined as "the sum of relations which hold between the units of text… and between each unit and the whole" (Goutsos 1997: 138). Studies of text organization fall within the scope of discourse analysis.

Modern discourse analysis includes studies of various levels and dimensions of discourse, as well as of cognitive processes and memory representations related to discourse, and is thus "not a simple enterprise" (van Dijk 1985: 5-10). In its nature, discourse analysis is interdisciplinary, encompassing methods and levels of "analysis of language, cognition, interaction, society, and culture" (van Dijk 1985: 10), which gives it much in common with translation studies, also interdisciplinary in its nature. Van Dijk states that discourse is "a manifestation of all these dimensions of society" (1985: 10)—a statement that might also be made of translation.

Specific definitions of discourse and discourse analysis vary—for instance,

Jaworski and Coupland (1999: 1-3) list ten definitions of discourse drawn from various works. The current study does not aim at providing an overview of such definitions, and understands discourse in a broad sense of communication in its situational and social contexts. In this sense, translation studies may appropriately be added to the list of the disciplines that, according to Tannen (1989: 7f), discourse analysis embraces. (In 1989, her list included linguistics, anthropology, sociology, psychology, literature, rhetoric, philology, speech communication, and philosophy.)

As Hoey (1991a) points out, discourse organization is often approached from the

viewpoint of genre or text-type 1 (Halliday and Hasan 1985, Martin 1985, Ventola 1987)

(13). In fact, as Dooley and Levinsohn emphasize, most analysis that is done on discourse

"can only be genre-specific," because "many linguistic observations about a given text

cannot be generalized"—they only hold for a specific text-type (2001: 7). As Longacre

(1996) puts it,"[t]he linguist who ignores discourse typology can only come to grief" (7).

The concept of genre or text-type is central to translation studies—translators train and

work within text-types, as do writers, typically specializing in only a few. The text-types

selected for the present study include modern prose, newspaper commentary devoted to

international news, and scientific writing.

The present study focuses on cohesive and other global textual features and

analyzes them in the context of translation, looking at non-translated, human-translated,

and machine-translated texts of different text-types. The framework for the present study

includes cohesive and other textual features relevant to translator's competence

(Campbell 1998), such as nominalization, average word length, lexical diversity (e.g.,

1 The two terms are often used interchangeably. In the context of translation studies in the U.S., the term "text-type" seems to be preferred.

type/token ratio), passives, and prepositional phrases. The study limits its scope to written

discourse.

2.2.2 Cohesion as a linguistic phenomenon

Cohesion is a linguistic category that has interested linguists, especially those working

with the English language, for decades—breaking ground with pioneering works of

Gleason (1968), Hasan (1968), Quirk et al. (1972), Enkvist (1973), Halliday and Hasan

(1976), Gutwinski (1976), de Beaugrande (1980), Hoey (1983), and other, more recent,

scholars. The notion of cohesion "is easily perceived but not easy to define" (Stoddard

1991: 13). This study adapts a multidimensional model of cohesion based on the definitions discussed below.

Cohesion is a unifying mechanism that ties textual segments, such as phrases, sentences, or paragraphs, into a whole by "connections among the elements within the discourse" (Campbell 1995: 5-6). According to Gutwinski, who was among the first scholars to devote considerable attention to cohesion, these connections, or relations, occur on the grammatical stratum, while being "signaled by certain grammatical and lexical features reflecting discourse structure on a higher … stratum" (1976: 28). These features represent "textual connectivity of sentences and clauses" (Gutwinski 1976: 28).

Another pioneer of cohesion and textuality studies, Robert de Beaugrande, argues that cohesion includes "the procedures whereby surface elements appear as progressive occurrences such that their sequential connectivity is maintained and made recoverable"

(1980: 19). In LeTourneau's words, "cohesion … denotes the 'syntax of texts' with respect to the conformity of syntactic structure to the demands of information structure" (2007:

145).

The monumental work on cohesion in English is that of Halliday and Hasan

(1976), and many works draw on their extensive description. Halliday and Hasan stress that the concept of cohesion is a semantic one—"it refers to relations of meaning that exist within the text, and that define it as a text" (1976: 4). Cohesion occurs “where the interpretation of some element in the text is dependent on that of another" (Halliday and

Hasan 1976: 4). Thus, we cannot successfully decode a cohesive element in a text without awareness of another element that may occur earlier or later.

Following the pioneers in the study of cohesion, The Handbook of Discourse

Analysis combines their views and defines cohesion as "the set of resources for constructing relations in discourse which transcend grammatical structure" (2001: 35). It can be seen as "residing in the semantic and grammatical properties of the language"

(Bex 1996: 91). Therefore, the term "cohesion" is two-fold: it includes both the observable surface structure markers and their cohesive functions that contribute to creating what is referred to in linguistics as "texture."

Halliday and Hasan are sometimes criticized for restricting the use of the term

"cohesion" only to relations that occur across sentence boundaries (Herbst 2010: 284), thus neglecting intra-sentential cohesive ties. However, as Herbst (2010) clarifies, "it is important to realize that they only do this in order to focus on the textual aspect of

cohesion" (284). In fact, as Herbst further notes, Halliday and Hasan themselves

emphasize that the "parts of a sentence or a clause 'cohere' with each other" and so

display texture (1976: 6). Later, Halliday (1985) himself suggests the 'clause complex' as the basic unit for studying cohesion. Following this viewpoint, which is shared by many researchers (notably, de Beaugrande and Dressler 1981: 50), I will use the term

"cohesion" to apply to both inter- and intra-sentential relations in a text.

2.2.3 Cohesion and texture

A text is "any passage, spoken or written, of whatever length, that forms a unified whole," and any such passage is characterized by texture (Halliday and Hasan 1976: 1).

Texture, achieved through cohesion, among other things, thus may be defined as "the property of a text of being an interpretable whole (rather than unconnected sentences)"

(Teich and Fankhauser 2004: 131). Texture is the process during which the flow of meaning is channeled into a digestible current of discourse "instead of spilling out formlessly in every possible direction" (Halliday 1994: 311). Cohesion is an important element of texture since it helps the reader to relate phrases, sentences, and paragraphs within a meaningful whole of a text. As Koch puts it, "cohesion is the text-forming component of the linguistic system" (2001: 2).

2.2.4 Cohesion vs. coherence

Cohesion uses the linguistic level of texts to relate "successive elements which constitute discourse" (Caron 1992: 161). This differentiates cohesion from coherence: the latter uses

the conceptual level to relate textual elements and "involves connections between the

discourse and the context in which it occurs" (Campbell 1995: 5). Koch's definition is

more succinct: "cohesion is described as a textual phenomenon whereas coherence is a mental one" (2001: 2). In Blum-Kulka's words, coherence is "a covert potential meaning relationship among parts of a text, made overt by the reader or listener through processes of interpretation" (1986: 298-299). Neubert & Shreve (1992: 94) describe a coherent text as having "an underlying logical structure that acts to guide the reader through the text."

The two terms are closely related. They are cognates, going back to the Latin cohaerēre, "to cling" or "to stick." As we have already seen, and as Dooley and

Levinsohn point out, "cohesion … can be defined briefly as the use of linguistic means to signal coherence," and these linguistic means serve as "clues to assist the hearers [or readers, in our case] in coming up with an adequate mental representation [or coherence]"

(2001: 27). These terms have been used in this way by many scholars—Grimes 1975,

Halliday and Hasan 1976, de Beaugrande and Dressler 1981, Brown and Yule 1983, among others.

One important difference between these two phenomena is that "coherence also works without cohesion but not the other way around" (Koch 2001: 2). Koch's example

of a non-coherent text that has several cohesive ties illustrates this difference very well:

"Father was home. Home is here. Here is there. There was mother" (2).

Since the two terms are interrelated, some scholars consider them confusing—for instance, Mossop (2001) suggests replacing them with more self-explanatory terms,

"smoothness" (to refer to cohesion) and "logic" (to refer to coherence). The present

research uses the traditional terms, originally borrowed by translation studies from

linguistics, to be consistent with the majority of literature on the subject.

2.2.5 Standards of textuality

Discussions of cohesion, coherence, and texture frequently involve the seven standards of

textuality set forth by de Beaugrande and Dressler in their seminal work Introduction to

Text Linguistics (1981). The seven standards of textuality include cohesion (a network of

surface relations), coherence (a network of conceptual relations), intentionality

(concerned with text producers and their intentions), acceptability (concerned with the reader's ability to interpret the text), informativity (based on the balance of new and given information), situationality (concerned with the relevance of a text to its context), and intertextuality ("relationship between a given text and other relevant texts encountered in

prior experience" (Neubert and Shreve 1992: 117), which, for instance, allows readers to

recognize a poem as a poem, etc.).

The standards of textuality are of direct relevance to translation studies, especially

when these studies focus on texts as units of translation (rather than smaller units, such as

sentences or paragraphs). The present research closely examines one of these textual

standards—cohesion.

2.2.6 Definition of cohesion in the present study

When cohesion is discussed in the framework of translation studies, its language- and culture- specificity is of vital importance. This fact is reflected in some definitions of cohesion in translation studies. For instance, Blum-Kulka defines cohesion as "an overt relationship holding between parts of the text, expressed by language specific markers"

(299).

For the purposes of this study, cohesion will be defined as a set of overt, language-specific resources that tie a text together at a global level. As per Halliday and

Hasan (1976), such linguistic resources in English consist of five distinct categories— reference, substitution, ellipsis, conjunction, and lexical cohesion (13). These categories are described in more detail in 2.2.7. The present study looks at reference and conjunction cohesive devices only, since they are most suitable for an automated corpus analysis. Other global textual features included in this study are nominalization, lexical density, average word length, average sentence length, passives, and prepositional phrases. These features are defined and discussed in Chapter 2.

Following the path set forth by Kaplan's contrastive rhetoric, this dissertation assumes that cohesion differs across languages and cultures. As Dooley and Levinsohn

point out, "[e]ach language will, of course, have its own range of devices which can be

used for cohesion" (2001: 27). Halliday and Hasan based their description on English-

language writing, and their framework for the study of cohesion may not be as well-

suited for other languages. However, since the present study deals only with the texts

translated into English or originated in English, the use of Halliday and Hasan's

categorization is appropriate.

It may be important to note that the present study does not deal with possible

expansion of the definition of cohesion—e.g., the inclusion of such features as syntactic

parallelism, functional sentence perspective (which deals with information arrangement),

or graphic devices (such as typography, enumeration, or chart types), as suggested by

Campbell (1995: 7) and other scholars. Such an expansion may ultimately be useful, since Halliday and Hasan's arguments were based mostly on their analysis of written literary works; however, it is outside the realm of this research.

2.2.7 Categories of cohesion

Halliday and Hasan's categorization has been used and expanded by other scholars. For example, Dooley and Levinsohn suggest six types of cohesion categories that they consider to be cross-linguistic (2001: 27). They include identity, lexical relations, morphosyntactic patterns, signals of relations between propositions (conjunctions), and intonation patterns. The categories of identity, lexical relations, and conjunctions overlap with Halliday and Hasan's categories of lexical repetition and replacement, reference

(pronouns), substitution, ellipsis, and conjunction. Intonation is a category that is difficult to study in written texts. For these reasons, Halliday and Hasan's categorization, tested by time and practice, was chosen for the current research.

Halliday and Hasan distinguish between two main types of cohesion—

grammatical cohesion and lexical cohesion. These two main types are further broken

down into concrete groups of cohesive resources.

2.2.7.1 Grammatical cohesion

Grammatical cohesion is realized through overt grammatical means of a language.

Halliday and Hasan name three of their five categories as being purely grammatical

cohesive ties: reference, ellipsis, and substitution. Conjunction combines both

grammatical and lexical features.

2.2.7.1.1 Reference

Reference is a property in which an item in a text (a presupposing element), "instead of

being interpreted semantically in [its] own right," requires recourse to another item (a

presupposed element) (1976: 31). It is "the most applied cohesive device in texts" (Koch

2001: 6). In English, reference is achieved through pronouns, demonstratives, the definite

article, comparatives, and such adverbs as "here," "there," "now," and "then."

Depending on whether the presupposed element occurs within the text or outside

of it, reference is endophoric (textual) or exophoric (situational). As Thompson puts it,

exophoric references point "outwards to the world," while endophoric references point

"inwards to the text" (1996: 149). Exophoric references can refer to the context of the

situation or discourse (e.g., the use of the first and second person pronouns), to assumed

cultural knowledge (e.g., "the current president of the United States"), or another artistic

work. Exophoric (situational) references are excluded from the scope of the present work since this study uses automated corpus analysis tools and thus can only focus on overt grammatical relationships within a given text that can be detected by a computer program.

Endophoric, or textual, references are among the main cohesive phenomena included in the present research. They may be anaphoric or cataphoric (Halliday and

Hasan 1976: 33); in the case of an anaphoric reference, the presupposed item precedes the presupposing item, while in the case of a cataphoric reference, it follows the presupposing item. Koch observes that in written discourse, "anaphoric reference is more often used than cataphoric reference" (2001: 4).

Halliday and Hasan further distinguish reference as being personal, demonstrative, or comparative (1976: 37). Personal reference is "reference by means of function in the speech situation" (37) and is created with the help of personal pronouns, possessive determiners, and possessive pronouns. As mentioned above, the first and the second person personal pronouns are often exophoric (situational) and refer to discourse participants outside of the text (speaker, addressee(s), writer, or reader(s)) (51). Third person pronouns tend to be endophoric and refer to something immediately within the text, and are thus cohesive (51). Thus, only third-person pronouns are included in the scope of this research.

Personal pronouns seem to be common cohesive devices in both English and

Russian. They help writers avoid unnecessary repetitions. However, Russian is

characterized by grammatical gender (i.e., inanimate singular nouns carry gender

distinction (masculine, feminine, or neuter), so the noun цветок ('flower') is a masculine

noun and thus requires a masculine personal pronoun (он) when the writer resorts to reference. English inanimate objects do not carry gender distinction (with very few exceptions, such as "ship," which is often referred to as "she"), and thus can be referenced by only one pronominal form when singular—"it."

Tables 2.1 and 2.2 illustrate nominative forms of personal pronouns for English and Russian. As mentioned above, only third person personal pronouns are included in this study.

The use of possessive pronouns in notably different in English and Russian.

Possessive pronouns or other modifiers are often obligatory in English (e.g., "He took his hat off" – the sentence without "his" would be grammatically incorrect in English). In

Russian, however, it is not only permissible but also more typical to have a noun without any explicit modifier (e.g., compare a translation of this same sentence in Russian "Он

снял шляпу"—with no modifier before the noun шляпа ('hat').

Table 2.1 Personal pronouns in English Singular Plural First person I we Second person you you Third person he, she, it they

Table 2.2 Personal pronouns in Russian Singular Plural First person я мы Second person ты вы Third person он, она, оно они

Moreover, Russian has a particular possessive pronoun свой, which has no direct equivalent in English, and can approximately be rendered with the English as "(one's) own." It may be used with all persons but is especially common (and often obligatory) with the third person. The use of other possessive pronouns in place of свой might cause differences in meaning. For instance, "Он взял со стола свою шляпу" may be rendered in English as "He took his (his own) hat from the table" (note the use of свой in Russian in place of the English "his"). If we replace свой with the Russian possessive pronoun

его, the meaning will change, and the Russian sentence will indicate that "he took his

(someone else's) hat from the table."

Tables 2.3 and 2.4 illustrate nominative forms of English and Russian possessive pronouns. For Russian, only masculine forms are provided as examples.

Table 2.3 Possessive pronouns in English Singular Plural First person my our Second person your your Third person his / her / its their

Table 2.4 Possessive pronouns in Russian Singular Plural First person мой наш Second person твой ваш Third person его / её / его их свой свои

Demonstrative reference is realized "by means of location, on a scale of

proximity" in relation to the speaker (space) or the moment of speech (time) (Halliday

and Hasan 1976: 37-38) and is "a form of verbal pointing" (57). In most cases, according

to Koch, "demonstratives signal that something was just mentioned" (2001: 5). In terms of semantics of demonstrative reference, Halliday and Hasan distinguish between the dimension of "near" and the dimension of "not near" (1976: 38). The adverbial demonstratives of place and time ("here," "there," now, and "then") typically function as

adjuncts in a clause and occur outside of a noun phrase (or a "nominal group," per

Halliday and Hasan), while nominal demonstratives ("this," "these," "that," "those," and

"the") tend to occur as elements of a nominal group (57-58). The English definite article

may be described as "the most neutral item amongst the demonstratives" (Thompson

1996: 150); it "merely indicates that the item in question is specific and identifiable; that

somewhere the information necessary for identifying it is recoverable" (Halliday and

Hasan 1976: 71).

English demonstratives (excluding the definite article "the," which does not exist

in Russian) are summarized in Table 2.5, and Russian demonstratives are summarized in

Table 2.6.

Table 2.5 Demonstrative cohesive devices in English Demonstrative pronouns Adverbs Category Singular Plural Place Time Near this these here now Not-near that those there then

Table 2.6 Demonstrative cohesive devices in Russian Demonstrative pronouns Adverbs Category Singular Plural Place Time Masculine Feminine Neuter Near этот эта это эти тут сейчас Not-near тот та то те там потом

The major differences between the English and Russian demonstratives include:

1) Russian demonstratives can be masculine, feminine, or neuter (этот

мужчина ('this man'), эта женщина ('this woman'), этот цветок ('this flower').

2) Russian demonstrative pronouns are declined, and have different forms in different grammatical cases.

3) Russian adverbs of place have a different form for location and direction (тут

('here' as location)—сюда ('here' as direction); там ('there' as location)—туда ('there' as direction).

4) Russian has no definite or indefinite articles; thus, the concept of definiteness has to be realized by other means, such as word order or the explicit demonstrative pronouns listed above. This difference may influence translators working from Russian into English, since they have to introduce articles or replace Russian demonstratives with

English definite articles.

Comparative reference is realized with the help of identity, similarity, or difference (in the case of general comparison), or numerative and epithetic expressions

(in the case of particular comparison) (Halliday and Hasan 1976: 76-77). Any comparison "includes two things that are being compared; and any comparative attached to one entity or concept thus implies the existence of the other entity or concept"

(Thompson 1996: 149). Some examples of general comparison include the use of "same,"

"equal"," such," "similar," "other," "different," etc. Particular comparison is realized with the help of numerative comparative devices (e.g., "more," "fewer," "less") or epithetical comparatives (e.g., comparative adjectives, such as "better," "worse," "larger," "smaller,"

"more difficult," "less difficult," etc.).

Both Russian and English have morphological and analytical comparatives. The possibility of a morphological comparative in English often depends on prosody (the number of syllables—e.g., the comparative adjective "nicer" is possible, while

"beautifuller" is not), while in Russian, this is not the case. English morphological comparatives can be used as attributes ("a nicer dress"), while Russian morphological

comparatives cannot be used as attributes. (Russian does not allow for such expressions

as "Это красивее платье.")

Although involved analysis of differences and similarities between English and

Russian is outside the scope of the present research, such examples illustrate that there

are differences substantial enough to pose a potential challenge.

General and particular comparison in English and Russian are illustrated in Tables

2.7 and 2.8. These tables represent re-arranged versions of the corresponding tables in

Halliday and Hasan (1976: 76) and Simmons (1981: 66), respectively. For Russian, only the singular masculine form is provided for adjectives and determiners that change their form according to the number and gender of the corresponding nouns. Both adjectival and adverbial forms are provided in the same cell, in keeping with Halliday and Hasan's approach.

Table 2.7 Comparison in English General comparison Particular comparison Numerative Identity Similarity Difference Epithet (quality) (quantity) comparative more, fewer, adjectives & less, further, adverbs (e.g., other, same, equal, such, similar; additional; better); different, else; identical; so, similarly, so-, as-, so-, as-, more-, differently, identically likewise equally- + less-, equally- + otherwise quantifier (e.g., comparative adj. so many) & adverbs (e.g., equally good)

Table 2.8 Comparison in Russian General comparison Particular comparison Numerative Epithet Identity Similarity Difference (quantity) (quality) лучший, худший, разный, compound похожий, различный, больше, comparative тот же самый, подобный, другой, иной; меньше, еще; adjectives, одинаковый; сходный; разнообразно, такой + равно, так; так + quantifier подобно, различно, по- adjective; похоже, (e.g., так то же самое другому, иначе так + adverb сходно много) лучше, хуже, заранее, позже, etc.

2.2.7.1.2 Substitution

Substitution is the second source of grammatical cohesion distinguished by Halliday and

Hasan. It is defined as "the replacement of one item by another" (1976: 88). Halliday and

Hasan devote a considerable amount of attention to various instances of substitution in

English (88-141). They also point out differences between substitution and other sources

of cohesion, such as reference and ellipsis (see next section). They emphasize that

substitution is "a relation between linguistic items, such as words and phrases; whereas

reference is a relation between meanings" (89).

Substitution in English can be nominal (achieved by the use of "one/ones"

or "the same" in place of a noun phrase, as in "We have no coal fires; only wood ones"),

verbal (realized with the help of "do"/"did" in place of a verb, as in "Never a woman in

Windsor knows more of Anne's mind than I do2"), and clausal (realized through the use of "so" and "not," when they replace an entire clause, as in "'Is there going to be an earthquake?' ' It says so'") (Halliday and Hasan 1976: 91-130).

According to Simmons, substitution is "not a viable category of cohesion in

Russian" (1981: 64). Since this study centers on the usefulness of automated tools, and substitution is little tractable to those tools, it is outside the scope of this study.

2.2.7.1.3 Ellipsis

Ellipsis, the third source of grammatical cohesion in English per Halliday and Hasan, is similar to substitution – "it can be defined as substitution by zero" (1976: 89). It occurs when an item is omitted and no tangible substitution happens. In the case of ellipsis,

"there is a presupposition, in the structure, that something is to be supplied, or

'understood,'" and a sense of incompleteness is present (144). In most cases, according to

Halliday and Hasan, "the presupposed item is present in the preceding text" (144).

Halliday and Hasan distinguish between three types of ellipsis: nominal, verbal, and clausal (cf. types of substitution). Nominal ellipsis occurs within nominal groups, when a modifier in such a group replaces the noun and starts functioning as a head noun

2 Halliday and Hasan quote Shakespeare's The Merry Wives of Windsor.

(Halliday and Hasan's example from Lewis Carroll is "Four other Oysters followed them,

and yet another four" (148)). Verbal ellipses occur within a verbal group (e.g., "'Have you

been swimming?' 'Yes, I have,'" in which case, the verbal form "swimming," from the

present perfect continuous tense "have been swimming," is elided (167)). Finally, clausal

ellipses are observed when a modal or propositional element of a clause can be elided.

According to Halliday and Hasan, a modal element of a clause includes the subject and

the finite element of the verb phrase, while a propositional element includes the

remaining parts of the verb phrase, as well as objects, complements, or adjuncts (197).

For instance, in "'What was the Duke going to do?' 'Plant a row of poplars in the park,'"

the modal element is omitted; while in "'Who was going to plant a row of poplars in the

park?' 'The Duke was,'" the propositional element is omitted (197-198).

While Halliday and Hasan emphasize the difference between ellipsis and substitution, some researchers account for them as one category. For instance, Thompson describes two types of ellipsis, ellipsis proper ("a gap") and substitution (where a gap is filled with "a substitute form") (1996: 153). This difference in categorization little impacts the present research. It illustrates an eternal tendency of scientists to either

"lump" or "split" categories and sub-categories.

Russian is rich in ellipsis. Comparing the use of ellipses in Russian into English translations (or the reverse), however, lies outside the goals of this study. While this study focuses on the usefulness of the applicability of a computer-based approach to the

study of many cohesive devices, the study of ellipses would likely require considerable

qualitative analysis.

2.2.7.1.4 Conjunction

Conjunction is yet another cohesive device discussed by Halliday and Hasan, and is quite different in nature from reference, substitution, or ellipsis. According to Halliday and

Hasan, the cohesive function of conjunctions is indirect and is realized "by virtue of their

specific meanings," which "presupposes the presence of other components in the

discourse" (226). Conjunctions relate "linguistic elements that occur together in

succession," thus creating ties between segments of text (Halliday and Hasan 1976: 227).

They combine "any two textual elements into a potentially coherent complex semantic unit" (Thompson 1996: 156).

While reference, substitution, and ellipsis are "clearly grammatical" because they involve closed systems (e.g., such systems as those of person, number, proximity, degree of comparison, or presence/absence), conjunction is "on the border-line of the grammatical and lexical" (303). As Halliday and Hasan point out, "the set of conjunctive elements can probably be interpreted grammatically in terms of systems, but … some conjunctive expressions involve lexical selection as well, e.g., "moment" in "from that

moment on" (303-304).

Hoey (1991a), among others, suggests discounting conjunction as a cohesive tie

on the grounds of "its quite different function in text formation" (9), pointing out that it is

"better treated as part of a larger system of semantic relations between clauses" (5).

Regardless, in the current research, conjunction will be included, since conjunction is still

part of textual organization, and may yield interesting results.

In Halliday and Hasan's Cohesion in English, four types of conjunction are

identified: additive, adversative, causal, and temporal (238). Additive conjunction is "a

generalized semantic relationship in the text-forming component of the semantic system"

(234), more commonly known as coordinating conjunction, such as that realized with

"and" or "besides." Adversative conjunction is a relation used as "contrary to expectation" (250). Some examples of adversative conjunctions include "but," "yet,"

"however," "(al)though," "nevertheless," etc. Causal conjunction is a cause-effect relation

(of reason, result, or purpose) and is expressed by such examples as "so," "thus," "hence,"

"therefore," etc. (256). Lastly, temporal conjunction is a relation of sequence in time, when one element in a text "is subsequent to the other" (261). Examples of temporal conjunctions include "then," "thereupon," "later," etc.

Conjunctions are abundant in both English and Russian. Simmons suggests accepting Halliday and Hasan's categorization of conjunction for the Russian language

(1981: 69). As examples, Simmons lists the Russian и ('and') as additive conjunction,

однако ('however') as adversative conjunction, так ('so') as causal conjunction, and

потом ('then') as temporal conjunction. Since the present study concentrates on English

texts only, further analysis of conjunction in Russian is not relevant.

2.2.7.2 Lexical cohesion

Unlike grammatical cohesion, lexical cohesion is realized through lexis, or vocabulary

(Halliday and Hasan 1976: 318). Since the present study focuses on the analysis of grammatical cohesion (including conjunction), the following overview of lexical cohesion is tangential, and therefore will be kept brief. That said, lexical cohesion is a rich topic worthy of its own study, and may be tractable with corpus tools. It is a tempting target for future studies.

Halliday and Hasan differentiate between two aspects of lexical cohesion— reiteration ("the repetition of a lexical item, or the occurrence of a synonym of some kind, in the context of reference; that is, where the two occurrences have the same referent") and collocation (the use of "a word that is in some way associated with another word in the preceding text, because it is a direct repetition of it, or is in some sense synonymous with it, or tends to occur in the same lexical environment") (1976: 318-319). In Halliday and Hasan's framework, collocations point to some semantic relationships and include, for example, superordinates, hyponyms, and antonyms.

Cohesive repetitions do not necessarily have to be lexical. Repetitions occur at the clause/sentence level as well. Some scholars (e.g., Gutwinski 1976 and Gleason 1965) use the term "enation" for the repetitions of syntactically similar sentences. Gleason states that two sentences are "enate" if "they have identical structures, that is, if the elements

(say, words) at equivalent places in the sentences are of the same classes, and if constructions in which they occur are the same" (1965: 199). Conversely, Enkvist terms

the repetition of syntactically and phonologically similar clauses and sentences "iconic

linkage" (1973). Other scholars (e.g., Quirk et al. 1972 and James 1983) use such terms

as "formal parallelism" or "structural parallelism."

Gutwinski points out that the cohesive function of enation (or iconic linkage, or

structural parallelism) may be reinforced by lexical cohesion, as well as other features of

grammatical cohesion (1976). He also points out that syntactic similarity, or enation, may

be complete or partial.

It should be noted that the concept of collocation defined by Halliday and Hasan

in their 1976 work differs from the one widely used in corpus linguistics, which dates

back to the works of Firth (1957) and Sinclair (1966), with Firth's famous quote "You

shall know a word by the company it keeps" (11). In corpus linguistics, collocations are

defined as "characteristic co-occurrence patterns of words," or, simply put, "words that

'go together' or words that are often 'found in each other's company'" (Bowker and

Pearson 2002: 32). In this interpretation, they are of major interest to corpus linguistics, since they represent "one type of word behavior that can be identified with the help of a corpus" (32). As Hoey (1999) puts it, collocation "as the relationship a lexical item has with items that appear with greater than random probability in its (textual) context … is in principle statistically demonstrable (as long as one processes enough text)" (7-8).

In her chapter in Understanding Reading Comprehension (1984), Hasan re-works the earlier model of lexical cohesion from 1976, developing such sources as repetition, synonymy, antonymy, hyponymy, and meronymy. Among other researchers who devoted

attention to lexical cohesion were Winter (1977), Francis (1985), Hoey (1991a),

McCarthy (1991), Martin (1992, 2001), and Matthiessen (together with Halliday, 1999).

Francis (1985) focused her attention on anaphoric nouns; Hoey (1991a) related "lexical

patterning to how lexical cohesion operates over larger stretches of text"; McCarthy

(1991) discussed lexical cohesion and discourse-organizing words; Winter (1977) focused on the anaphoric function of lexis, while Halliday and Matthiessen (1999) continued developing an ideational semantics (Lexical Cohesion and Corpus Linguistics

2006: 2). Martin combined Hasan's categories with Halliday and Hasan's earlier model and integrated the ideas of some of the other scholars mentioned above, proposing a

modular perspective for analyzing cohesion within a broader framework for analyzing discourse (Martin's contribution to The Handbook of Discourse Analysis, 2001). In addition, collocation was "factored out into various kinds of 'nuclear' relation, involving elaboration, extension, and enhancement (as developed by Halliday 1994 for the clause complex)" (2001: 38). Martin's term for lexical relations deployed "to construe institutional activity" was "ideation" (38).

Halliday and Hasan's treatment of lexical cohesion is considered insufficient by some linguists. For instance, in their introduction to Lexical Cohesion and Corpus

Linguistics, Flowerdew and Mahlberg (2009) point out that in Halliday and Hasan's

detailed description of different sources of cohesion in English, they give the shortest

treatment to lexical cohesion ("less than twenty pages," as Hoey (1991a) notes). In her

chapter in the said volume, Mahlberg states that lexical cohesion should be assigned a

more central role in linguistic research (2009: 105). She is not the first to take this stance.

For instance, Stotsky (1983) asserts that Halliday and Hasan's analysis does not

adequately describe the types of lexical ties found in written texts and proposes dividing

lexical cohesion into semantically-related words (e.g., words related by repetition,

synonymy, inclusion into an enumeration or a set) and collocationally-related words

(related through co-occurrence in contexts).

Hoey, who devoted a considerable amount of attention to lexical cohesion, observes that lexical cohesion "is the single most important form of cohesive tie" (1991a:

9). He illustrates it with Halliday and Hasan's own sample analysis, where lexical cohesion accounts for over forty percent of ties they identify (1991a: 9). According to

Hoey, "[l]exical cohesion is the only type of cohesion that regularly forms multiple relationships" between textual elements (1991a: 10). Hoey argues that the study of cohesion in texts is much more about lexis than previously believed: "the study of cohesion in text is to a considerable degree the study of patterns of lexis in text" (1991a:

10). In his Patterns of Lexis in Texts (1991a), Hoey provides a more detailed treatment of lexical repetition than do Halliday and Hasan.

Some scholars, such as Mahlberg (2009), argue for a different approach to studying cohesion categories, due to the fact that we should not assume that "lexical and grammatical phenomena can be clearly distinguished" (103). Mahlberg calls for "a corpus theoretical approach" to the description of cohesion, where cohesion is seen in a new light: "cohesion is created by interlocking lexico-grammatical patterns and overlapping

lexical items" (103). However, as the classification established in 1976 by Halliday and

Hasan has been far more widely applied, the present study uses it for its research

framework.

2.2.8 Other global textual features

In addition to studying cohesive devices in translated and non-translated texts, the present

study looks at other features that characterize texts globally. The additional features are

included following Dong and Lan's model for translation evaluation (2010). In their

model, Dong and Lan combine the categories of grammatical cohesion set forth by

Halliday and Hasan (1976) with Campbell's (1998) categories for evaluating textual

competence of translators. In his research on translation into a second language,

Campbell suggests that textual competence of translators may be empirically studied by

measuring such textual features of target texts as nominalization, average word length,

lexical diversity, passives, and prepositional phrases. The additional parameter of average

sentence length is also included into the framework, since some scholars find that

translated and non-translated texts exhibit significant differences in average sentence

length (Laviosa 1998a, 1998b).

The rationale for including other textual features into the present analysis, combining them with Halliday and Hasan's grammatical cohesion categories, is two-fold.

First, it makes the research framework more comprehensive, eliciting additional empirical data from the corpus. Second, this inclusion may allow the findings to be more

closely related to studies in translation competence and translation pedagogy. The results of this study may benefit researchers in these fields by providing benchmark data for translated and non-translated texts, especially for the Russian into English language pair, and by informing design of translation pedagogy and evaluation tools. Last but not least,

Campbell himself termed his exclusion of cohesion from his translator competence model

"a notable omission" (1998: 159).

2.2.8.1 Nominalization

Halliday (1994) broadly defines nominalization as a process "whereby any element or group of elements is made to function as a nominal group in the clause" (41). Following

Dong and Lan's research, the present study looks at a narrower part of nominalization tractable with Writer's Workbench. It measures only the number of noun forms of verbs as its nominalization count (Writer's Workbench Manual).

When translating between languages, especially when they are as structurally different as English and Russian (with English being an analytical language, and Russian, a synthetic one), translators might use nominalization or de-nominalization as a strategy.

Differences in the use of nominalization across languages might influence the process and product of translation. In addition, the use of nominalization is predicted to vary across genres.

Some linguists study nominalization as a feature that can be exploited ideologically by writers and editors. For instance, Fowler (1991) suggests that

nominalization "offers substantial ideological opportunities" (80), since it allows for

more information to go unexpressed and thus permits the "habits of concealment,

particularly in the areas of power-relations and writers' attitudes."

2.2.8.2 Lexical density

Texts can be described in terms of their lexical density. Type-token ratio is a measure of lexical density that is based on mathematical division of the number of unique words and

their forms (types) by the total number of words in a text (tokens). A high type-token

ratio is an indication that a text contains many different lexical items, so a higher

proportion of the words in that text have a specific meaning (Westin 2002: 77). A low

ratio "shows that few specific words are used while the more general ones are frequent"

(ibid.).

Westin notes that the category of type-token ratio was frequently used by

psycholinguists starting in 1960-1970s (e.g., Biber 1988) in their studies of differences

between oral and written genres (77). Written genres tend to have a higher type-token

ratio than spoken genres, possibly due to differences in production circumstances (Chafe

and Danielewicz 1987).

2.2.8.3 Average word length

Average word length is another textual feature that has been used as a measure of lexical

diversity, and "is included into several norm-based writing tests and in readability

formulas" (Troia 2009: 365). According to Troia, "[t]he measure of word length,

operationalized as the number of letters in a word, capitalizes on the fact that word length

and frequency are inversely related; as length increases, frequency decreases" (365). For

this reason, he notes the use of "lower frequency words as a developmental measure of

more mature writing" (365).

Average word length is a component of common readability formulas, including

the Flesch Readability Test, the Bormuth Mean Cloze formula adopted by the College

Entrance Examination Board and others. A comprehensive overview of readability

formulas can be found in DuBay's 2004 white paper.

2.2.8.4 Average sentence length

Sentence length is a measure that is often included into style guides and readability

formulas (Flesch Readability Test, Gog Index, the Clear River Test, etc.). Flesch

incorporates sentence length into his original "Reading Ease" readability index as early as

the late 1940s. The present study employs average sentence length as an additional

textual parameter since it has been shown that this may exhibit significant differences in

translated and non-translated texts (Laviosa 1998a and 1998b).

Studies of sentence length date back to the early 20th century. For instance, Kitson

(1921) finds that the average sentence length is different for different English-language newspapers (e.g., it was shorter in the Chicago American than in the Post of his times).

Vogel and Washburne (1928) include sentence length and prepositional phrases into the ten structural characteristics they used for their empirical study of readability. In Gary

and Leary's comprehensive study of readability (1935), longer sentences were found to be

more difficult for readers. In fact, the study found that average sentence length is among

the best predictors of textual difficulty. The study also showed negative correlation

between the number of prepositional phrases and readability.

2.2.8.5 Passives

Passives are an important characteristic of genres. It has received considerable attention

in linguistics, and translation studies might benefit from looking at this category as well.

The basic discourse function of passives is "to switch the focus of action in the active

clause from subject/actor to object/patient" (Westin 2002: 118). Research suggests that

the frequency of passives varies between text-types. For instance, Svartvik (1966: 152-

154) finds that texts pertaining to science have the highest number of passives, followed by texts pertaining to news and arts, while advertising has the fewest. Westin finds that the use of agentless passives in English newspaper editorials has decreased over time

(2002: 120), which might indicate "a drift away from a language that Biber characterizes

as 'abstract, technical, and formal' (Biber 1988: 112-113)."

The use of passives has also been linked to ideological purposes. For example,

Kress (1983: 127-128) emphasizes that passives allow for focusing on the affected entity and representing the action as an attribute of the affected entity while omitting the agent

of the sentence.

2.2.8.6 Prepositional phrases

Westin (2002) suggests that the use of prepositional phrases is "an effective way of

packing high amounts of information into idea units and of expanding the size of them,"

and so this feature is often found in informational discourse (71). In Russian, many

relations that would require a preposition in English may be expressed synthetically (e.g.,

"United States of America" is rendered in Russian as "Соединенные Штаты Америки,"

with the of-relation expressed synthetically with the help of the genitive case). For this

reason, translators working from Russian into English often have to introduce

prepositional phrases where there were none in the source text, which might be reflected

in the number of prepositional phrases in translated texts.

The frequency of prepositional phrases varies across text-types. In their study of

conversations, lectures, personal letters, and academic writing, Chafe and Danielewicz

(1987) find that prepositional phrases are frequent in all of these genres, while being

unusually frequent in academic writing. Biber's corpus study (1988) supports these

results.

2.2.9 Cohesion and other global textual features as a language- and culture-specific phenomenon

One of the main reasons why cohesion and other textual features pose problems for translators is their language- and culture-specificity (e.g., Hatim and Mason 1990, Baker

1992, Blum-Kulka 2000, Teich and Fankhauser 2004). As Blum-Kulka points out,

cohesive relationships in texts are linked to a language's grammatical system, as well as

to stylistic preferences for types of cohesive markers in each language involved in

translation (299-300).

In different cultures and languages, readers may have different background

knowledge of the world and particular subjects, and "discourses … force us to draw upon

all we know about our culture, language, and world" (Everett 1992: 19). For this reason,

as Mona Baker points out, even a "simple cohesive relation of co-reference cannot be

recognized … if it does not fit in with a reader's prior knowledge of the world" (1992:

220). Baker analyzes an extract from A Hero from Zero, pointing out that in it, there are no explicit cohesive ties between "Harrods" and "the splendid Knightsbridge store."

While a British reader, in this case, would not need an explicit cohesive tie between the two phrases due to his/her background knowledge (most British readers are familiar with the famous store and know that it is in Knightsbridge), a reader in a different culture might not be able to make such a connection without the translator explicitating cohesive ties between "Harrods" and "the splendid Knightsbridge store" (Baker 1992: 220-221).

Cross-cultural and cross-linguistic differences also lie in the structures of the languages involved in translation. In a small study of cohesion and coherence in one

Chinese text and its English translation, Yeh (2004) finds that "cohesion of a text in

English is constituted by reference items, such as 'he' or 'they,' while … in Chinese, [it] might be realized by the existence of a topic chain" (249). Lonsdale observes that due to the fact that English makes very few overt distinctions in terms of gender, number, or

subject-verb agreement, translators should pay special attention to making references

clear in an English-language target text (Lonsdale 1996: 219). In her study of Bulgarian

and English legal texts, Yankova (2006) finds a higher incidence of lexical cohesion in

English texts than Bulgarian texts, which might be predicted to play a role in translators'

decision-making processes. She also describes a higher number of synonyms and

collocations in Bulgarian texts (although this finding was not statistically significant). In

her study of English and Norwegian originals and translations, Hasselgård finds evidence

suggesting that English needs explicit markers of cohesion more often than Norwegian,

which may be explained by the fact that "the word order in Norwegian is more flexible

than that of English, particularly in sentence-initial position" (1997: 16).

The language and culture specificities of cohesion and other features of text

organization inevitably influence the process and product of translation (Johansson and

Hofland 2000). Toury, when discussing his translation law of interference, mentions that

“phenomena pertaining to the make-up of the source text tend to be transferred to the target text” (1995: 275). Pym, in his article on Toury's translation laws, explains this

"make-up" as "a set of segmentational and macrostructural features" (2007: 7), which would include cohesive devices and other features of text organization that are often transferred from the source to the target text. In European languages, Pym notes, there is

"a default norm by which one translates sentence for sentence, paragraph for paragraph, text section for text section." (Pym 2007: 7). This default behavior, Pym posits, may explain what Toury describes as "interference."

Non-interference, in its turn, "necessitates special conditions and/or special efforts

on the translator's part" (Toury 1995: 275). These special efforts would be targeted at

changing global textual features, such as cohesive devices, sentencing, or paragraphing.

And according to Pym, translators usually "need a very good reason to change a

paragraph break" (2007: 7). Elisabeth Le notes that "[t]ranslators are torn between the

apparent need to respect sentence and paragraph boundaries and the risk of sounding

unnatural in the target language" (2004: 267). Some empirical studies, for example, Baer

and Bystrova-McIntyre (2009), suggest that such global textual features as sentencing

and paragraphing are different across languages, as well as across the text-types included in their studies.

All this, again, stresses that translation inevitably involves changes, or shifts (to use Blum-Kulka's term) at a global level that should take into account the cultural and linguistic specificity of cohesion and other features of text organization. Lonsdale, in her textbook on teaching translation from Spanish to English, stresses that "different languages use cohesive devices (reference, substitution, conjunction, lexical and syntactic cohesion, chunking of information in sentences and paragraphs) differently," and translators must consider adjusting cohesive devices from the source language to the target language (215). Without such adjustments, translations may sound unnatural to the target reader.

The present study seeks to illuminate textual shifts that English translators consider when working with Russian texts first-hand, editing translations done by other

humans, or post-editing MT output. It establishes target language benchmarks by analyzing English non-translated texts, and then compares them to the results of analyses of human-translated and machine-translated texts from Russian into English. It also addresses the text-type specificity of cohesion and other features of text organization.

2.3 Studies of cohesion and other global textual features

This section starts with an overview of linguistic studies of cohesion in monolingual settings, including the contexts of writing evaluation and pedagogy, text comprehension, and genre and text-type specificity of cohesion categories. The reason for including an overview of linguistic studies that are not directly relevant to translation is that translation studies can benefit greatly from employing the findings and methodologies of linguistics, sociolinguistics, discourse analysis, and other disciplines. The section then proceeds to cover studies of cohesion in the context of translation, discussing its difficulties and reporting on achievements.

2.3.1 Studies of cohesion in the context of teaching and evaluating writing

Cohesion is an important and fascinating research topic in the context of teaching and evaluating writing. As discussed above, cohesion is an integral part of creating texts.

Still, some scholars point out the need to incorporate cohesion and coherence into writing textbooks at a more detailed and organized level. For instance, Kolln informally examined fifteen random books on writing (five for high school juniors and seniors, ten for college freshmen) and found that three (all college level) had no index entry for

"coherence," and only one (also college) included the word "cohesion" as an entry (1999:

94). Kolln also points out that textbooks on writing often include only parts of the concept of cohesion—e.g., while transition devices are mentioned in all fifteen books she looked at, only a few bring up parallelism and repetition in a positive light, and only one included a detailed framework of cohesion as outlined by Halliday and Hasan (95). While

Kolln's analysis may not be representative in a scholarly sense, it suggests that more attention should be given to cohesion and coherence in writing (and, consequently, in translation, since the latter involves writing as its major component).

Meanwhile, Chiang (2003) finds that the features of cohesion are considered an important predictor of writing quality by both native and non-native writers of English who rated foreign language writing samples using 20 discourse and grammatical features.

In fact, "all except three of the 30 raters based their perceptions of overall quality [of writing samples] primarily on either of the two discourse features: coherence and cohesion" (471). Of the four areas of evaluation used in this study (coherence, cohesion, syntax, and morphology), "regression analysis showed that … cohesion was the best predictor of writing quality of all the four areas" (471).

In the context of writing, cohesion is often examined in relation to the perceived quality of writing. For instance, Witte and Faigley (1981) examine cohesion in low- and high-rated essays and report a significant difference in the frequency of cohesive ties, finding that in low-graded essays, 20.4 percent of all the words contributed to cohesion, while in high-graded essays, the number was significantly higher—31 percent.

The number of ties, however, may not be the most representative feature of successful cohesion—"Quantity is not necessarily quality" (a common wisdom sited by

Kolln 1999: 96). Some researchers find no significant differences in the number of cohesive ties across differently rated writing samples. For instance, Tierney and

Mosenthal (1983) report that there is no causal relationship between propositional measures of cohesive ties within topic and coherence rankings within topic (215-229).

Hasan (1984), when counting the number of cohesive ties in children's writing and measuring coherence using readers' responses, also finds no easy correlation between the two. Irwin (1986), in her study of cohesion factors in children's textbooks, finds no significant differences in the total numbers of cohesive ties and connective concepts across the selected grade levels (64).

The number of cohesive ties proves more useful when examined across multiple dimensions. Many researchers report significant differences in the number of ties across different text modes, as well as when more granular types of cohesive ties are differentiated. Crowhurst (1987), in her study of cohesive ties across grade levels and modes (narration and argument), finds that the mode makes more difference than the grade, with 25.5 percent as the highest number of ties for narrations produced by sixth graders. In addition, Crowhurst's results indicate significant differences across the use of specific cohesive devices (e.g., the use of synonymy and collocation increased with age, which may be related to children's vocabularies expanding as they grow up) (1976: 192).

Irwin (1986), whose study was already mentioned above, finds significant differences in

terms of the frequency of shared arguments across main clauses, with lower-level textbooks displaying more cohesion than upper-level textbooks (1986: 64). This points to a usefulness of a more granular study of cohesive ties. For this reason, the present study looks at the use of cohesive ties in specific, granular, categories, in texts of different production modes (translations vs. non-translated texts), and across several text-types.

Last but not least, a significant body of research on cohesion exists for experiments involving impaired participants (Cherney et al. 1998, Fine 1994, etc.).

2.3.2 Studies of cohesion in the context of text comprehension

Cohesion is closely linked to readability and comprehension. Horning (1991) emphasizes that "[s]tudies of cohesion in reading show that cohesion makes a substantial contribution to readability" (136). In one such study, Irwin (1980) measures reading time and recall for two versions of a passage that differ in the number of cohesive ties, with one version containing about twice as many ties as the other, and finds that an increased number of cohesive ties improves readers' comprehension.

Of course, comprehension depends not only on the properties of the text itself.

McCutchen and Perfetti (1982) point out that the ability of readers to create the sense of coherence in a text depends on their knowledge of the topic and textual conventions.

Charolles goes further and states that "[i]n the end, it all depends on the receiver, and his

[sic] ability to interpret indications present in the discourse" (1983: 95). Comprehension

also develops with practice and experience on the part of the reader. Chapman (1987), for

instance, finds that readers' awareness of cohesive ties improves over time.

The translation scholar Mona Baker brings all this together, stating that textual

coherence is "a result of the interaction between knowledge presented in the text and the

reader's own knowledge and experience of the world" (1992: 219). Many linguists, such

as Barthes (1975), Eco (1979), Iser (1974), or Steiner (1975), have argued that, in the process of reading, the reader of a text becomes its producer rather than its receiver

(Bassnett-McGuire 1980: 79). The reader's experience and knowledge may be influenced by a number of factors, including age, sex, race, nationality, education, occupation, and political or religious affiliation (Baker 1992: 219).

Translators have to deal with text comprehension both as readers of the source text and as producers of the target text for the reader with a potentially different background knowledge and experience. Cohesion is an important factor in text comprehension, and is thus of strong relevance to translation studies.

2.3.3 Studies of cohesion as a genre-specific and text-type specific phenomenon

As suggested earlier, research on cohesive ties suggests that their usage differs across genres and text-types. As Mahlberg emphasizes, "[c]ohesive links are genre-specific"

(2006: 107). Mahlberg goes on to explain that, for instance, narrative texts that "deal with a central character … can provide many examples of reference and chains of reference items," while newspaper articles, in contrast, may be "more likely candidates to illustrate

lexical relationships where sentences share three or more lexical links" of the type

discussed in Hoey (1991a), who describes different categories of lexical repetition (e.g.,

complex repetition involving words that share a lexical morpheme but are not identical,

such as "argue" and "argument") (2006: 107).

The genre-specificity of cohesive ties is supported by research. For example,

Smith and Frawley (1983) study the use of conjunctions in fiction, religious texts, journalism, and science. Their findings suggest that conjunctions used in fiction are more similar to the ones used in religious texts than to those used in journalism and science.

The conjunctions in science tend to be additives and hypotheticals that indicate causes

and logical succession. Verikaitė (2005) analyzes conjunctive discourse markers

(additive, adversative, causal, and temporal) in the genres of textbooks and scientific

research articles. According to her results, "the frequency of occurrence of a particular conjunction varied across the genres," for example, the causal and temporal conjunction were used more frequently in the research articles—suggesting that the variance may be due to genre constraints (68).

It should be noted that cohesion and other textual features also differ across

media, or modes, of production—oral vs. written. For example, in oral discourse, we find

more repetition than in written discourse. Aaron (1998) explains this as follows: "in

written language there is a limit to how much repetition can be tolerated by readers" (3).

Dooley and Levinsohn (2001) point out differences in organization between written and oral discourse, for example, written texts tend to be more tightly organized; they tend to

introduce information at a faster pace and have longer groupings of sentences (16-17).

Tanskanen (2006) finds differences in lexical cohesion (reiteration and collocation) across different genres on a spoken-written continuum (face-to-face conversations— prepared speeches—mailing list language—academic writing). Since the present study involves written texts only, additional discussion of difference across oral and written media is outside its scope.

The present study does involve methods of text production in the context of translation studies—it looks at English non-translated texts, texts translated from Russian into English by human translators (human-translated texts), and texts translated from

Russian into English by MT tools (machine-translated texts).

2.3.4 Studies of cohesion in the context of translation studies

In translation studies, cohesion has gained increased attention relatively recently, with its rise marked by the "textual turn" in translation studies, when global features, such as cohesion, were recognized to be of central importance (Neubert and Shreve 1992). Still, few studies have yet been done to isolate cohesive features of translated texts; as Chau

Hu points out, the issue of cohesion "in the theory and pedagogy of translation … has not been adequately studied thus far" (1999: 33). Not surprisingly then, until only recently, the American Translators Association (ATA) error marking framework had no explicit category of cohesion and few other categories that recognized phenomena above the level of the sentence (American Translators Association website). So while we know that

expert translator behavior is marked by a global, top-down approach to text creation

(Jääskeläinen 1990; Tirkkonen-Condit and Jääskeläinen 1991; Kussmaul 1995), the

world of professional translation has not fully benefited from this knowledge. A part of

the translation profession continues to concentrate, like novice translators themselves, on

the sub-sentential level, overlooking cohesion as a vital feature of translation.

This situation can "in part be explained by the fact that the qualities that constitute

cohesion are generally difficult to pinpoint and isolate" (Baer and Bystrova 2009: 163).

As Wilson (quoted in Callow 1974: 10-11) noted regarding the first translation of the

bible into Dagbani:

For a native speaker it was difficult to express what was wrong with the earlier

version, except that it was ‘foreign.’ Now, however, a comparison … has made clear that

what the older version mainly suffers from are considerable deficiencies in ‘discourse

structure,’ i.e., in the way the sentences are combined into well-integrated paragraphs, and these in turn into a well-constructed whole.

In addition, the construction of cohesion in translated texts may be complicated, as Baker (1992: 125) points out, by a tension between syntax and thematic patterning, requiring recasting not for the sake of semantics, understood in a limited sense, but for the sake of cohesion. Blum-Kulka refers to such language-specific recastings as "shifts"

(299).

Another possible reason that cohesion has been understudied in translation may be that it, unlike "unique items" (Tirkkonen-Condit 2004) or hapax legomena (Kenny

2001), is rather difficult to investigate with automated tools. The number of features studied with automated tools is typically limited. For instance, while Baker (1992) discusses automated studies of lexical density and type-token ratio as indicators of cohesion in texts, this omits other cohesive devices.

When studying cohesive devices in translation, many researchers perform their text analyses and annotation manually, which is labor-intensive (e.g., studies by Hoey

1991a , Hu 2004, Yeh 2004, and Zhao et al. 2009) and limits the number of texts included in such studies. For instance, the study by Zhao et al. (2009) includes only 15

English originals and their Chinese translations, manually annotated using Halliday and

Hasan's categories of reference, substitution, ellipsis, conjunction, and lexical cohesion.

Manual annotation has its advantages and disadvantages. As Teich and

Fankhauser point out, the inter-rater reliability in manual annotation is often less than satisfactory (132). In many studies, inter-rater reliability is not even checked. Further, smaller-sized corpora may be less representative of a broader population. However, manual annotation may allow for a more detailed analysis of the few selected texts, as shown in the studies by Øverås (1998) and Zhao et al. (2009).

Corpus studies, the advantages of which are discussed in more detail in Chapter 2, provide opportunities for automated studies that do not involve human annotation. A famous example of a corpus-based study of the optional "that" in translated English texts

was done by Olohan and Baker (2000). The authors analyze the inclusion or omission of

the optional 'that' with the reporting verbs 'say' and 'tell' using well-known corpora—the

Translational English Corpus (TEC) and the British National Corpus (BNC). Olohan and

Baker's data show that the 'that'-connective "is far more frequent in TEC than in BNC,

and conversely that the 'zero'-connective is more frequent for all forms of both verbs in

the BNC corpus than in TEC" (141). These results are interpreted as providing evidence

for explicitation as a feature of translated texts. The present study aims at expanding the

number of cohesion features elucidated by a corpus study.

Some researchers study cohesion in translation in the realm of translation

universals, searching for evidence supporting their existence (see Section 1.2.3). For

instance, Øverås (1998), who studies cohesion in an English-Norwegian parallel corpus,

finds that translated texts exhibit an overall rise in the level of cohesion, which suggests explicitation in translation. Her findings point to "shifts in conjunctive and reference cohesion through addition and expansion in the specification of nouns" by means of determiners, substitution, and shifts in lexical cohesion (Kruger 2002: 90). Blum-Kulka's explicitation hypothesis is also based on the use of cohesive markers in translation

(1986). She observes that a target text may be more redundant than a source text, and this

"redundancy can be expressed by a rise in the level of cohesive explicitness" in the target text (2004: 300).

Studies of cohesion in translation often include a limited number of text-types.

For instance, Zhao et al. (2009), mentioned above, look at cohesive devices in medical

texts only. The researchers find more similarities than differences in the use of cohesive

devices in medical originals and their translations, with the only difference found "in the

employment of reference in terms of occurrence frequencies" (2009: 313). The authors

conclude that this is due to the need to maintain "precision, clarity, and logicality" in

translated medical texts. To develop a broader model of cohesion in translation, more

text-types need to be included.

It seems useful to compare the use of cohesive devices in translated texts to their

use in non-translated texts of the same text-type, as was done in Olohan and Baker's analysis (2000). Such an empirical comparison might help illuminate differences between translated and non-translated texts. As Kruger points out, "[c]orpus-based research in translation is concerned with revealing both the universal and the specific features of translation" (2004: 70).

2.3.5 Studies of other global textual features in the context of translation studies

Studies of translation product and process have also employed other features of text organization to characterize texts and aid translators' work and training. For instance,

Byrne (2006) devotes his entire book to improving the quality of technical translations with the help of usability testing, which includes readability tests and other measures of text characteristics. He emphasizes how important it is for technical translators to be familiar with best practices of document design and testing procedures for the variety of technical test-types—the knowledge that today's technical writers spend years acquiring

but translators often lack (2006: ix). Among other things, Byrne mentions such categories

as average sentence length, average word length, nominalization, and passive

constructions. Related to studies of cohesion in translation, Byrne, based on the results of

his empirical study, suggests that iconic linkage has a positive effect on the usability of

software user guides. 'Iconic linkage' is Juliane House's term for "the repetition or reuse of target language translations for source language sentences which have the same meaning but different surface properties "(164).

Translation researchers have used the study of textual features to illuminate differences between translated and non-translated texts. For example, studies done by

Laviosa (1998a, 1998b) look at lexical density and sentence length in translational and non-translational corpora in the genres of newspaper articles and narrative prose. For newspaper articles, Laviosa's findings suggest that translated texts have a lower lexical density (a lower percentage of content words versus grammatical words, in Laviosa's terms), and display a lower average sentence length (1998a). For narrative prose, Laviosa

finds that the lexical density is significantly lower in translated texts, while the mean

sentence length, contrary to the findings for the genre of newspaper articles, is

significantly higher (1998b: 561).

Laviosa's translational corpora include translated texts from a variety of source

languages, since her research aims at pinpointing features that translated texts exhibit

regardless of their source language. However, it is of interest to look at particular

language pairs; as Laviosa herself cautiously hypothesizes, "the average sentence length

may be particularly sensitive, in the narrative subject domain, to the influence of different

source languages, as well as the author's particular style" (1998b: 565).

To give another example of how textual features are used in translation research,

Khanna et al. (2011) consider sentence length and textual complexity when studying translations of sentences from a medical pamphlet performed by professional (human) translators and a machine translation tool. They find that while sentence length is not associated with scores given by graders to human and machine translations, complexity is associated with the graders' preference scores. In their study, the graders preferred more complex English sentences produced by professional translators.

Interestingly, some textual features have been studied in the framework of

ideology. For instance, some scholars view nominalization as a feature that can be

exploited ideologically by translators and editors. Kuo and Nakamura (2005), who study

ideological changes in English-Vietnamese translations, find that nominalization is one of the devices that may be used "to mask the 'reality' presented in the original news text"

(411). On the other hand, attempts to "de-vilify" nominalizations have also been made

(Martin 2008).

2.3.6 Cohesion and other global textual features and translation expertise

The notions of cohesion and other textual features that characterize texts globally are relevant to studies of expertise in translation. The ability to function at a macro-textual level, which involves dealing with the question of cohesion and other global features of

texts, is often cited as a trait of translation experts. Being able to function at a macro-level

and take global aspects of a task into account are characteristic of experts in other fields

as well. In their overview to The Nature of Expertise, Glaser and Chi (1988: xviii) report

that "experts see and represent a problem in their domain at a deeper (more principled)

level than novices; novices tend to represent a problem at a superficial level." The famous

1981 study on the categorization of physics problems among experts and novices by Chi,

Feltovich, and Glaser demonstrated that experts used underlying physics principles to

categorize problems, while novices used surface features as the basis for problem

categorization.

Translation researchers have indicated that novices in translation tend to

concentrate on the most straightforward and superficial level of words and phrases. For

instance, Shreve (2002) mentions the tendency of novice translators to “view the

translation as a sequence of exclusively lexical problems” (164). Shreve notes that

novices often ignore more complex structures because they do not recognize them or do not think of them as part of the scope of the translation activity (165). Expert translators, on the other hand, have an ability to “recognize that the sentence might not be the appropriate level to work with as a unit” (Séguinot 1999: 92). Differences in dictionary use and overall approach to translation (microcontext vs. macrocontext), reported in

Krings (1986), support this claim as well. The results reported in Pouget (1998),

Tirkkonen-Condit & Jääskeläinen (1991), Kussmaul (1995), and Angelone (2010) also suggest a more global approach of professional translators to their task.

For this reason, the present study of cohesion and other global textual features

might facilitate the development of curricula aimed at helping translation students shorten

their track to becoming experts in their profession by addressing the global feature of

textual cohesion. Calls for introducing cohesion and coherence into translation instruction

have already been made (e.g., Chau Hu 1999). Based on the results of this study, data-

informed pedagogical interventions can be developed to explicitly teach issues of

cohesion and textual organization in translation. Such interventions would encourage

student translators "to consider the target text globally, as a product involving a variety of

features above and beyond lexis, for which they are professionally responsible" (Baer and

Bystrova 2009: 163).

2.4 Conclusions

This chapter provided an overview of literature pertaining to cohesion and other global

features of texts in both monolingual and multi-lingual settings. It also outlined the main concepts and theories related to differences across translated and non-translated texts, as well as across different genres of texts. In particular, it discussed Toury's law of interference and the concept of translation universals, which maintain that translated texts possess characteristics different from non-translated texts in the same language. Toury's law of interference relates this to influences of source texts, while the concept of translation universals presupposes that unique characteristics of translations exist regardless of the source language.

This study focuses on cohesion as one of the central global features of texts.

Consequently, this chapter extensively described developments in the concept of cohesion in monolingual and multilingual studies, from the pioneering works by Halliday and Hassan (1976), Gutwinsky (1976) and others to more modern views of such scholars as Hoey (1983, 1991a, 1991b), Campbell (1995), Dooley and Levinsohn (2001), and

Mahlberg (2009), to name but a few. The chapter included definitions and descriptions of grammatical and lexical categories of cohesion in both English and Russian. Further, it looked at the concept of cohesion in translation (Blum-Kulka 2004, Baker 1996, etc.).

The chapter also gave considerable attention to other global features of texts and their importance in monolingual and translation research. Such features included nominalization, passives, prepositional phrases, average sentence and word length, and lexical density.

The chapter also provided an overview of studies of other global textual features in monolingual settings, including the contexts of writing evaluation and pedagogy, text comprehension, and genre and text-type specificity of global textual categories. Then, it covered studies of cohesion and other global features in the context of translation, discussing their difficulties and reporting on their achievements.

The chapter then introduced methodologies used to study global features of texts, which often involve methods of corpus linguistics. Studies of electronic corpora permit analyses of global text features in a more objective way and at a larger scope than manual analyses, helping illuminate linguistic features not visible to the naked eye. The chapter

also outlined practical applicability of studies such as this one in designing ways to

distinguish between human-translated, machine-translated, and non-translated texts and

for developing pedagogical interventions in training translators and (post)-editors.

In translation studies, global features of texts have gained increased attention relatively recently—with a "textual turn" in translation studies that recognized such global features to be of vital importance (Neubert and Shreve 1992). Still, few studies have yet been done to isolate global features of translation; as Chau Hu points out, the issue of cohesion "in the theory and pedagogy of translation … has not been adequately studied thus far" (1999: 33). In the current practice of translation, where content is translated by global translation teams using a plethora of translation technologies, translators' attention is often focused on segments below the textual level, threatening the maintenance of a global textual orientation. These realities of the industry make it vitally important to conduct empirical studies of such global features. As discussed in this chapter, studies of global features should include larger corpora, more text-types, and more comprehensive methodologies—all of which this dissertation seeks to address.

CHAPTER 3: METHODS AND MATERIALS

This chapter begins with a general overview of corpus linguistics methods as tools to analyze texts. It then outlines the framework used in this research to study cohesion and other features of texts, and presents the methods, tools, procedures, and materials employed in this dissertation. In addition, it describes the process of compiling the three-

dimensional, multi-genre corpus used for this study.

3.1 Corpus linguistics as a methodology to analyze cohesion and other global textual

features

The present study employs automated methods of corpus linguistics to investigate the use

of cohesive devices and other global textual features in a number of different text-types and across translated and non-translated texts. The corpus-based automated approach allows for comparison among substantial numbers of texts, including non-translated texts and texts translated into English by humans and machines.

Defined as a "collection of electronic texts assembled according to explicit design criteria," a corpus is usually compiled with the goal of "representing a larger textual population" (Zanettin 2002: 11), providing linguists with a "much more solid empirical basis than had ever been previously available" (Granger 2003: 18) (Baer, Bystrova-

McIntyre 2009: 161). Prior to corpus studies, linguists were limited to "what a single

individual could experience and remember" (Sinclair 1991: 1). As Sinclair (1991) notes,

due to enormous developments in computer technology over the last several decades,

corpus linguistics became possible and allowed for "the emergence of a new view of

language," when much more data and much more powerful tools to analyze it became

available to linguists (1).

Translation studies adopted methods and tools of corpus linguistics somewhat

later than linguistics itself did. In her preface to the book Introducing Corpora in

Translation Studies, Maeve Olohan (2004: 1) remarks, referencing Graeme Kennedy

(1998: 1), that, while linguistics has employed electronic corpora for more than three decades, "the use of corpora […] in translation studies has a short history, spanning no more than ten years." This makes corpus-based translation studies a fruitful area of development (Baer, Bystrova-McIntyre 2009: 161). The use of corpora allows researchers to compile databases of unprecedented size and to analyze data from them within seconds (Sinclair 1991: 1).

Corpus linguistics as a methodology involves "large text corpora … [and] derives its results from the analysis of such corpora" (Herbst 2010: 34). According to Bowker

(2001), the value of using a corpus-based approach in translation studies lies "in its broad scope, the authenticity of texts, and in the fact that data are in machine-readable form"

(Baer, Bystrova-McIntyre 2009: 161). Moreover, modern computational tools and methods for quantitative and qualitative analyses (word frequencies, concordances, etc.) allow for more empirical and objective research (Bowker 2001: 346).

Applying the techniques of corpus analysis to the study of cohesion may address

certain criticisms of Halliday and Hasan's cohesion theory and the "perceived lack of

utility of quantitative analysis" of cohesive devices (Campbell 1995: 8). A presumed lack

of utility, for instance, is noted by Hendricks (1998: 104), who deems the mass of data

that arises from applying Halliday and Hasan's framework manually to longer texts

"practically useless." As noted earlier, corpus analysis techniques allow for easier handling of masses of data, helping to detect patterns and assess the statistical significance of analyses.

The corpus-based approach also allows researchers to study a more comprehensive list of cohesive devices and other textual features, especially with the relatively new framework that I employ in this study. This framework, developed by

Dong and Lan (2010), includes a variety of cohesive markers identified by Halliday and

Hasan (1976), as well as several features of text organization suggested by Campbell

(1998), thus increasing the scope of coverage for cohesive and textual features that were

previously lacking in most automated analyses, which focused on such features as type-

token ratio, paragraph length, lexical density, readability indices, or individual words.

3.2 The framework for studying cohesion and other global textual features

The present study employs the framework for studying cohesive markers and other global textual features designed by Da-Hui Dong & Yu-Su Lan (2010). Dong and Lan applied their framework to translations produced by different groups of Chinese-into-English

translators (novices, native experts, and non-native experts). The results of their study were published by a major translation publisher, St. Jerome Publishing, in 2010. The present study uses a different application of Dong and Lan's framework, focusing on differences and similarities that might exist between human-translated, machine-

translated, and non-translated texts, with English as the target language, and Russian as

the source language.

Dong & Lan combine parts of Halliday & Hasan's (1976) framework with one

part of Stuart Campbell's model of translation competence—textual competence (1998).

Below, I outline the parts of both models that were used by Dong and Lan.

As described in Chapter 2, Halliday and Hasan (1976) differentiate two main

components of cohesion: grammatical and lexical. The cohesive devices that constitute

Halliday and Hasan's framework include: reference, substitution, ellipsis, conjunction,

and lexical cohesion. For their framework, Dong & Lan choose two of these five

categories: reference and conjunction (50). For the term "reference," I use the definition

offered by Dong & Lan: "[r]eference is a relation between an element of the text and

another element that it is linked to which gives it meaning" (50). Several pronoun

referents and demonstratives are covered by the term (50):

− Pronominal

− Demonstrative

− Definite article

− Comparative

Conjunction is defined as "the semantic linkages between related sentences,"

which includes coordinating as well as subordinating conjunctions "used to express assistive, adversative, causal, and temporal relationships between and within sentences"

(Dong & Lan 2010: 50-51):

− Additive ("and," "also," etc.)

− Adversative ("however," "yet," "but," etc.)

− Causative ("so," "therefore," "consequently," etc.)

− Temporal ("then," "next", "finally," etc.)

− Continuative ("anyway," "nonetheless," etc.)

Halliday & Hassan's model is supplemented with a component of Campbell's

model of translation competence (1998). The idea of combining parts of the two models

belongs to Dong & Lan, and, in my opinion, makes the resulting model more

comprehensive.

Campbell's model was originally designed for assessing translation into a second

language and is now well accepted in studies of translation expertise. It includes three

types of competence—textual competence, translators' disposition, and monitoring

competence. Campbell's textual competence is based on Bachman's definition of textual

competence as "the knowledge of the conventions for joining utterances together to form

a text" (1990: 88). In Campbell's terms, it is "the capacity to deploy grammar and lexis

above the level of the sentence" (1998: 60). It also involves mastery of genre conventions

and "sensitivity to naturalness" (61). Disposition is a personal competence of a translator.

Campbell claims that translators' vocabulary choices "reveal tendencies of risk-taking versus prudence and persistence versus capitulation" (3). Finally, the term "monitoring competence" describes translators' ability to self-edit and to assess their own work (126-

127).

Of these three competences, textual competence yields most easily to corpus studies. In fact, to assess textual competence, Campbell uses a model designed by Biber to describe structural genre norms in corpora. Biber's model is quite extensive (1988:

223-245), and, among a wide variety of textual features, includes nominalization, average word length, lexical diversity (e.g., type/token ratio), passives, and prepositional phrases.

These features, as well as sentence length, are included in this study.

As a result of combining Halliday & Hasan's cohesion model with Campbell's components of textual competence, Dong & Lan suggested the framework outlined in

Table 3.1. Their framework was supplemented by an additional textual feature—sentence length. It was also slightly modified to address possible oversights of the researchers.

While Dong & Lan used their framework to study translation competence, I use this framework to study characteristics of texts translated from Russian into English by humans and machines when compared with non-translated texts in an attempt to pinpoint aspects of cohesion in translation in the Russian-English language pair. Table 9 provides an overview of the modified framework adapted from Dong and Lan (2010: 59).

Table 3.1 The combined framework for the study of the use of cohesive devices and other global textual features (adapted from Dong and Lan 2010) Components of previous models Dong and Lan's model (2010) Cohesive devices (Halliday and Hasan 1976) Reference devices Pronominal devices—personal Included (3rd person pronouns only, since other pronouns are less likely to be cohesive devices) Pronominal devices—possessive Added by the researcher Demonstratives Included Definite article Included Comparative devices Included Conjunction devices Additive devices Included Adversative devices Included Causal devices Replaced by subordinate conjunction Temporal devices Included Continuative devices Replaced by subordinate conjunction Global textual features (Campbell 1998) Nominalization Included Lexical density (type-token ratio) Included Word length (average) Included Passives Included Prepositional phrases Included Sentence length (average) Added by the researcher

3.3 Materials

This section outlines criteria and procedures used in compiling the corpus, and describes in detail the sub-corpora comprising it.

3.3.1 Corpus compilation

At the broadest level, monolingual corpora can be divided into two types (Zanettin 2011:

15). A corpus can be general—attempting to represent a "national variety" of a language.

Or it can be specialized—focusing on a particular variety of language: text type/genre

(e.g., news, scientific articles), domain/topic (e.g., international events, physics),

production method (e.g., oral language, written language), or a combination of these.

On the other hand, corpora used in translation studies are almost always

specialized, and the opportunities for specialization are numerous. They may be parallel

(consisting of source texts and their translations) or comparable (consisting of translated

and non-translated texts of similar criteria in a single language). Comparable corpora

should "cover a similar domain, variety of language and time span, and be of comparable

length" (Baker 1995: 233). Translational corpora may be unidirectional (consisting of

translations into one language only) or multidirectional (consisting of translations to and

from a certain language) (Zanettin 2008). In addition, translational corpora might include

texts produced by different groups (such as professionals or learners) or methods (such as

human-translated or machine-translated). For this study, I built a set of specialized

corpora based on the combined criteria of text-type/genre and production method.

For the purposes of this study and future use, a monolingual comparable corpus of

English texts and a unidirectional parallel corpus of Russian source texts and their

English translations were compiled. The monolingual comparable corpus includes three

major categories of texts—texts translated from Russian into English by human

translators, texts translated from Russian into English by the Google machine translation

tool, and English non-translated texts. In the tables, figures, and graphs, these three

methods of text production will be referred to as follows:

Human-translated texts—HT;

Machine-translated texts—MT;

Non-translated texts—NT.

Quantitative analysis was performed on the monolingual English corpus only. The

corresponding parallel corpus of Russian source texts and their English translations was

compiled to produce machine translated texts for the monolingual English corpus and for

further qualitative analysis of specific textual features identified by the quantitative

analysis.

The corpus consists of three text-types—literary texts (modern prose), newspaper articles (international news/commentary), and scientific articles (physics). These sub- corpora will be referred to as follows:

Literary corpus—LC;

Newspaper corpus—NC;

Scientific corpus—SC.

All texts in the corpus were produced by professionals, and were published in books, literary magazines, scientific journals, and online resources of major Russian and

English newspapers that report professional editing of their articles. The corpus is synchronous: as of 2012, all the texts are considered to be modern (published in the past

10 years for the newspaper and scientific corpora, and the past 20 years for the literary

prose corpora).

Any corpus design entails considerations of representativeness, and involves

important decisions regarding the corpus size, text selection criteria, a balance of texts,

length of individual samples, mark-up, etc. (Baker 1995). These decisions depend on the

researcher's goals, his/her approach, and practical considerations (available time, funding,

etc.).

Typically, it is assumed that "the linguistic patterns which are observed can be

generalized beyond that corpus to the textual universe that corpus stands for" (Zanettin

2011: 15). However, as Leech points out, "difficulties of determining the size of the

textual universe and its sub-universes from which a corpus is to be sampled are

formidable" (2007: 139), and "representativeness remains the 'holy grail' of corpus

linguistics, something to strive for rather than something that can reasonably be attained"

(Zanettin 2011: 15). In the end, it is the researcher's call to make compilation decisions

based on best practices and practical considerations. As Ahmad (2008) notes, corpus

design "will always carry the unintended influence of the designer(s)" (61).

Even the best corpora compiled by large institutions and groups of researchers can

be criticized. Leech (2007) points out some flaws in the world-famous 100-million-word

British National Corpus (BNC). For instance, Leech notes that spoken language in it is

underrepresented, and that the genre proportionality may be flawed (136). This fact does

not undermine the benefits of corpus linguistics outlined in 2.1. As with many

methodologies, corpus studies work best when viewed as part of a larger effort to

describe linguistic phenomena, with both quantitative and qualitative methods used

together.

For specialized corpora, representativeness is improved by the fact that such

corpora cover much smaller text categories than general corpora. For this study, the sub-

corpora were compiled using strict criteria, such as text-types, publication types, publication years, authors, translators, lengths of samples, etc. (outlined in more detail in the following sections). The compilation criteria and practices used in this study were piloted and improved over several years of corpus compilation and research undertaken by this researcher and her advisor, documented in several articles and professional conference presentations (Bystrova-McIntyre 2006, 2007, Baer, Bystrova-McIntyre

2009). The present corpus represents substantial improvements over these prior efforts. In my opinion, these three sub-corpora are sufficiently representative of their specific language varieties for the goals of this study.

Still, the inclusion of a larger number of texts is an ever-present goal for corpus

researchers ("the bigger, the better" concept). However, the goal of largeness is more

realistically attainable by funded research groups rather than individuals. The compilation

process for the present corpus took several months of full-time work for one researcher, with gratis assistance from a database designer. Immediate tasks included locating source texts, their translations, and comparable non-translated texts (constrained by availability within the budget of the researcher, i.e., free or at a minimal cost), scanning and running character recognition for sources not available electronically, designing and maintaining a database to store and analyze the texts and metadata, running texts through machine translation and part-of-speech tagging software, collecting the data from the texts, and solving many unexpected problems arising in the process.

The comparable and parallel corpora compiled for the purposes of this study include 600 texts. Each of the three text-type sub-corpora includes 200 texts—50 Russian source texts, their corresponding English human and machine translations, and 50

English non-translated texts of comparable nature. From these, 450 English texts were used for the quantitative analysis.

For literary and scientific corpora, texts of at least 1,000 words were included.

When the texts ran longer than 2,000 words, the sample for inclusion was cut off at the end of the paragraph following the 2,000-word mark. This was done to make the corpus more manageable and to have texts of comparable length. For the newspaper corpus, the minimum 1,000-word requirement was not applied, since the genre of a newspaper article presupposes shorter texts.

The Russian segments in the parallel corpus were trimmed to match the English

samples obtained as described above. For the machine-translated part of the corpus, these

Russian samples were run through GoogleTranslate. The MT was performed on March

20, 2012 (a potentially important point, since GoogleTranslate is known to be under rapid development). As is well known, GoogleTranslate is a free translation tool, currently working for 58 different languages. GoogleTranslate is a statistical MT system, which analyzes truly monumental amounts of linguistic data—"it looks for patterns in hundreds of millions of documents" (GoogleTranslate's webpage). It supplements this process with the use of translation memories (TMs), which help fine-tune some translation solutions.

In fact, users can upload their own TMs into the Google Translator Toolkit, to improve quality of the output and tweak it for the domain of interest. For the purposes of the current project, no additional TMs were used.

The entire English-language corpus is comprised of 691,414 words. The descriptive statistics for the corpus are shown in Table 3.2.

Table 3.2 Descriptive statistics for the English-language corpus Descriptive parameters # of texts # words # sentences Human-translated corpora (English)* 150 229,100 12,177 Machine-translated corpora (English)* 150 219,429 11,577 Non-translated comparable corpora 150 242,885 13,174 (English) Totals for the English corpus 450 691,414 36,928 *These texts are translations of the texts in the Russian source-text corpora

Further compilation criteria were fine-tuned for each of the three corpora

individually, and are described in more detail in the following subsections.

3.3.1.1 Literary corpus

The literary corpus focuses on contemporary prose represented by short story writing. For

the LC, translations of Russian contemporary prose were scanned from individual

publications (e.g., Ludmila Ulitskaya's Sonechka), The New Yorker magazine (e.g.,

Ludmila Petrushevskaya's The Fountain House), and especially from collections of translated Russian stories, all published in English during the last 20 years. After scanning, optical character recognition was performed. Below is the list of the collections used for the literary corpus:

− The Penguin Book of New Russian Writing. 1996. Eds. Victor Erofeyev &

Andrew Reynolds)

− Rasskazy. New Fiction from a New Russia. 2009. Eds. Mikhail Iossel & Jeff

Parker

− Life Stories: Original Works by Russian Writers. 2009. Ed. Paul E. Richardson

− Russian Love Stories. An Anthology of Contemporary Prose. 2009. Ed. Nadya L.

Peterson

Since the corpus "should be representative in terms of the range of original

authors and of translators" (Baker 1995: 233), each author in the LC occurs only once,

which minimizes influences related to authors' individual styles. However, by necessity,

individual translators do occur multiple times in the corpus; there exist fewer Russian into English translators than authors, and a single translator often works with a number of

Russian authors. For this corpus, the occurrence of any one translator is limited to three times, to avoid major influences of individual translators' styles.

The source texts for the corpus of Russian contemporary prose were mainly obtained from the Internet. The Russian texts included in the corpus were available on websites of Russian online libraries or of individual authors. The process of corpus compilation was not easy, due to the limited availability of source texts and translations, as well as to the constraints of corpus compilation parameters (such as the inclusion of each author only once, the inclusion of each translator only three times, and the minimum length of the pieces).

The list of the Russian source texts with their corresponding human translations is presented in Table 3.3.

Table 3.3 Russian source texts and their corresponding human translations Author Translator(s) Title (Russian) 1 Alyokhin, Evgeni Mesopir, Victoria Ядерная весна 2 Astafyev, Victor Reynolds, Andrew Людочка 3 Babchenko, Arkady Allen, Nick Дизелятник 4 Boteva, Maria Mesopir, Victoria Дело в том кому веришь (sic.) 5 Danilov, Dmitry Robinson, Douglas Более пожилой человек 6 Dovlatov, Sergey Frydman, Anne Компромисс пятый 7 Epikhin, Nikolai Gusev, Mariya Рычева

8 Erofeev, Venedikt Mulrine, Stephen Василий Розанов глазами эксцентрика 9 Gallego, Ruben Schwartz, Marian Белое на черном 10 Geide, Marianna Gunin, Anna Чудовище Плещеева озера 11 Gelasimov, Andrei Bayer, Alexei Жанна 12 Goralik, Linor Iossel, Mikhail Говорит: (sic.) 13 Gorenshtein, Bromfield, Andrew С кошелочкой Friedrikh 14 Grishkovets, Yevgeny Richardson, Paul E. Спокойствие 15 Kabakov, Alexander Seluyanova, Anna Убежище 16 Kalinin, Vadim Gusev, Mariya Невероятная и печальная история Миши Штрыкова и его жестокосердной жены 17 Kamenetskaya, Maria Mikhailova, Julia Между летом и осенью 18 Kharitonov, Evgeny Tait, Arch Духовка 19 Khazanov, Boris Maizell, Sylvia Праматерь 20 Khurgin, Alexander Fisher, Anne O. Беруши 21 Kochergin, Ilya Gunin, Anna Потенциальный покупатель 22 Kozlov, Vladimir Gregovich, Andrea Праздник строя и песни & Iossel, Mikhail 23 Lipskerov, Dmitry Bayer, Alexei Эдипов комплекс 24 Lukyanenko, Sergei Bliss, Liv Сердце Снарка 25 Makanin, Vladimir Shayevich, Bela Однодневная война 26 Mamleev, Yury Mulrine, Stephen Тетрадь индивидуалиста 27 Moskvina, Marina Fisher, Anne O. Мусорная корзина для алмазной сутры 28 Muliarova, Elena Peterson, Nadya L. Хроника двух дней жизни Жени Д. 29 Pelevin, Viktor Favorov, Nora Онтология детства

Seligman 30 Petrushevskaya, Gessen, Keith & Дом с фонтаном Ludmilla Summers, Anna 31 Popov, Evgeny Porter, Robert Как съели петуха 32 Popov, Valery Doyle, James Любовь тигра 33 Prigov, Dmitry Shuttleworth, Mark Описание предметов 34 Prilepin, Zahar Ilinskaya, Svetlana & Убийца и его маленький друг Robinson, Douglas 35 Rubina, Dina Katz, Michael R. & Туман Komarov, Denis 36 Sadulaev, German Gunin, German Почему не падает небо 37 Senchin, Roman Mesopir, Victoria Тоже история 38 Shcherbakova, Galina Osorio, Rachel Три любви Маши Передреевой 39 Shklovsky, Evgeny Lindsey, Byron Бабель в Париже 40 Sokolov, Sasha Pobedinskaya, Olga Тревожная куколка & Reynolds, Andrew 41 Sorokin, Vladimir Hoffman, Deborah Черная лошадь с белым глазом 42 Starobinets, Anna Litman, Ellen Правила 43 Terts, Abram Tait, Arch Золотой шнурок (Andrew Sinyavsky) 44 Tolstaya, Tatyana Gambrell, Jamey Кысь 45 Ulitskaya, Ludmila Tait, Arch Сонечка 46 Voinovich, Vladimir Morley, Peter Роман (Трагедия) 47 Yerofeyev, Viktor Berdy,Michele A. Реабилитация Дантеса 48 Yuzefovich, Leonid Schwartz, Marian Гроза 49 Zobern, Oleg Gessen, Keith Шестая дорожка Бреговича 50 Zonberg, Olga Ilinskaya, Svetlana & Смилуйся, государыня рыбка Robinson, Douglas

The comparable non-translated literary sub-corpus consists of modern English- language prose harvested from Kindle and paper editions of collections of contemporary

American prose, as well as from the online archives of The New Yorker magazine's

Fiction section for the years of 2010-2012. Each author occurs only once. The list of the contemporary American prose collections is below:

− Henry Prize Stories 2011. Random House, Inc. Kindle Edition

− The Best American Short Stories 2011: The Best American Series. Eds. Geraldine

Brooks & Heidi Pitlor. Houghton Mifflin Company. Kindle Edition.

− The Best American Sampler: The Best American Series 2011. Houghton Mifflin

Company. Kindle Edition.

− The Best American Short Stories 2010: The Best American Series. Eds. Richard

Russo & Heidi Pitlor. Houghton Mifflin Company.

− The Best American Short Stories 2009: The Best American Series. Eds. Alice

Sebold & Heidi Pitlor. Houghton Mifflin Company.

− The Best American Short Stories 2008: The Best American Series. Eds. Salman

Rushdie & Heidi Pitlor. Houghton Mifflin Company.

− The Best American Short Stories 2005: The Best American Series. Eds. Michael

Chabon & Heidi Pitlor. Houghton Mifflin Company.

The list of the English non-translated texts and their authors is presented in Table

3.4.

Table 3.4 English non-translated literary texts and their authors 1 McGuane, Thomas A Prairie Girl 2 McDermott, Alice Someone 3 Chabon, Michael Citizen Conn 4 Sayrafiezadeh, Said A Brief Encounter with the Enemy 5 Lanchester, John Expectations 6 Bergman, Megan Mayhew Housewifely Arts 7 Millhauser, Steven Phantoms 8 Bissle, Tim A Bridge Under Water 9 Chimamanda Ngozi Adichie Ceiling 10 Egan, Jennifer Out of Body 11 Englander, Nathan Free Fruit for Young Widows 12 Keegan, Claire Foster 13 Goodman, Allegra La Vita Nuova 14 Havazelet, Ehud Gurov in Manhattan 15 Hurrocks, Caitlin The Sleep 16 Johnston, Bret Anthony Soldier of Fortune 17 Lipsyte, Sam The Dungeon Master 18 Makkai, Rebecca Peter Torrelli Falling Apart 19 McCracken, Elizabeth Property 20 Nuila, Ricardo Dog Bites 21 Oates, Joyce Carol ID 22 Powers, Richard To the Measures Fall 23 Row, Jess The Call of Blood 24 Saunders, George Escape from Spiderhead 25 Slouka, Mark The Hare's Mask 26 Calhoun, Kenneth Night Blooming

27 Doenges, Judy Melinda 28 Means, David The Junction 29 Shepard, Jim Your Face Hurtles Down at You 30 Tuck, Lily Ice 31 Watson, Brad Alamo Plaza 32 Barnes, Julian The Limner 33 Lethem, Jonathan Procedure in the Air 34 Simpson, Helen Diary of an Interesting Year 35 Akpan, Uwem Baptizing the Gun 36 Barry, Kevin Fjord of Killary 37 Bynum, Sarah The Erlking 38 Frame, Janet Gavin Highly 39 Lennon, Robert Eight Pieces for the Left Hand 40 Link, Kelly Stone Animals 41 Munro, Alice Silence 42 Chase, Katie Man and Wife 43 Krauss, Nicole From the Desk of Daniel Varsky 44 Tice, Bradford Missionaries 45 Mccorkle, Jill Magic Words 46 Moffett, Kevin One Dog Year 47 Rash, Ron Into the Gorge 48 Groff, Lauren Delicate Edible Birds 49 Shipstead, Maggie The Cowboy Tango 50 Tower, Wells Raw Water

The entire English literary corpus includes 297,866 words. Descriptive statistics for the LC are presented in Table 3.5.

Table 3.5 Descriptive statistics for the LC Descriptive parameters Non-translated Human-translated Machine-translated # of texts 50 50 50 # of words 102,983 102,395 92,488 # of sentences 6,972 7,017 6,456 Average sentence length 17.2 16.2 16.3

3.3.1.2 Newspaper corpus

The newspaper corpus is drawn from a genre of commentary on topics of international news. Russian source texts, their translations, and non-translated English articles were harvested from online sources. The articles were published between 2008 and 2012. The original Russian publications include five major news sources, each represented by 10 articles in the 50-text sub-corpus in an effort to make the corpus as balanced and representative as possible. The list of the publications is provided below:

− Izvestia

− Nezavisimaia gazeta

− Pravda

− Rossiiskaia gazeta

− Vzgliad

The corresponding translations were obtained from the publications' own English-

language versions, where available, and, in other cases, from the website

watchingamerica.com. All translations on watchingamerica.com undergo professional editing. Notably, each translation lists not only its translator, but also its editor—a rare phenomenon that may be viewed as a sign of respect and understanding for the profession of translation.

The non-translated English comparable texts were taken from five major English- language news sources. Each source is represented by 10 articles in an effort to make the corpus as balanced and representative as possible. The list of the publications is provided below:

− Chicago Tribune

− Los Angeles Times

− The New York Times

− Washington Post

− The Wall Street Journal

In all parts of the NC, the occurrence of individual authors is limited to two times.

The occurrence of individual translators was harder to control, since in some cases, their names were not listed (for instance, in the English version of Pravda).

In the process of compiling the NC, headlines, by-lines, captions, and acknowledgements were removed from the corpus samples, since they are likely to display text characteristics of their own (Perfetti 1987).

The entire English newspaper corpus includes 121,112 words. Descriptive statistics for the NC are presented in Table 3.6.

Table 3.6 Descriptive statistics for the NC Descriptive parameters Non-translated Human-translated Machine-translated # of texts 50 50 50 # of words 39,893 40,440 40,779 # of sentences 1,985 2,037 2,079 Average sentence length 20.7 20.2 20.1

3.3.1.3 Scientific corpus

The scientific corpus is comprised of texts on the subject of physics. This subject was selected because my father is a Russian physicist who has, over the years, resorted to my help as a translator for his publications. My father's familiarity, and my own, with the field made it easier to select and access publications to include in the SC.

The original Russian publications include five major physics journals that are routinely translated into English; each journal is represented by 10 articles published between 2008 and 2012 within the 50-text sub-corpus. To minimize the influence of individual writing styles, main authors were included only once. For the purposes of this study, a main author is defined as the author of a single-author work, two authors of a

two-author work, and the first author of a work with more than two authors. The list of

the journals, with their transliterated Russian and translated English titles, is provided

below:

− Fizika metallov i metallovedenie (The Physics of Metals and Metallography)

− Fizika tverdogo tela (Physics of the Solid State)

− Izvestiia akademii nauk (seria fizicheskaia) (Bulletin of the Russian Academy of

Sciences (Physics))

− Zhurnal tekhnicheskoi fiziki (Technical Physics)

− Pis'ma v zhurnal tekhnicheskoi fiziki (Technical Physics Letters)

The comparable English-language sub-corpus includes five prominent physics journals published in English, each represented by 10 articles in this 50-text sub-corpus.

Below is the list of these journals:

− Journal of Magnetism and Magnetic Materials

− Journal of Nanomaterials

− Journal of Physics (Condensed Matter)

− Physical Review B

− Physical Review Letters

For this sub-corpus, only the articles written by authors affiliated with U.S. universities and colleges were included. This was done to ensure that the English writing was as close to that of a native speaker of English as possible. As for the Russian sub-

100

corpus, to minimize the influence of individual writing styles, main authors were

included only once.

It should be noted that scientific international English is in itself a linguistic

phenomenon where the conventions of writing differ significantly from those in literary

prose and journalism, even when written by a native speaker of a language. What reads

well to a physicist in a scientific journal may not be intelligible to a reader of The New

Yorker magazine. As David Bellos points out in his non-scholarly but entertaining book

Is That a Fish in Your Ear?, "much of the English now written by natural and social scientists whose native language is other is almost impenetrable to non-specialist readers

who believe that because they are native English speakers they should be able to

understand whatever is written in English" (2011: 17), which might influence how native

speakers write in this domain.

In the process of compiling the SC, titles, abstracts, captions, and

acknowledgements were excluded from the corpus samples, since they are likely to

display peculiar text characteristics of their own, being specific genres by themselves

(Moon 1998). The lists of works cited were also excluded from the texts. When complex

formulas and signs caused trouble for the OCR or part-of-speech tagging software, the

formulas and signs were removed from the texts or replaced with a placeholder.

Moreover, formulas and signs are not part of this study, so their exclusion does not

influence it.

101

It is interesting to note that some of the Russian journals translated into English

do mention the names of the translators for some of their articles—among them are

Fizika tverdogo tela (Physics of Solid State), Pis'ma v zhurnal tekhnicheskoi fiziki

(Technical Physics Letters), and Zhurnal tekhnicheskoi fiziki (Technical Physics).

However, the information on translators was not included regularly, and was insufficient to control for the same translators' work in the corpus.

The entire English scientific corpus includes 272,436 words. The descriptive statistics for the SC are presented in Table 3.7.

Table 3.7 Descriptive statistics for the SC Descriptive parameters Non-translated Human-translated Machine-translated # of texts 50 50 50 # of words 100,009 86,265 86,162 # of sentences 4,217 3,123 3,042 Average sentence length 24.3 28.2 28.8

3.4 Methods and procedures

3.4.1 Corpus database

The compiled corpus is stored in a Microsoft Access database (named Disserata),

modified from its previous version (CorpusCopia) based on the results of the pilot study

conducted in January–April of 2011. In addition to the text samples, Disserata stores the

following metadata for each text: author(s), translator(s) (when applicable), work title,

102

year of publication, publication venue, method of text production (non-translated, human-

translated, machine-translated), relevant links, and researcher's notes.

3.4.2 Corpus annotation

Most corpora are annotated based on the goals of specific research projects. Annotation is

"the practice of adding interpretative, linguistic information to an electronic corpus of

spoken and/or written language data" (Garside, Leech, & McEnery 1997: 2). In the

process of annotation, this interpretive, linguistic information is tagged using existing or

newly designed conventions. For the purposes of this study, part-of-speech tagging was performed for all English-language samples. The tagged samples are stored in Disserata as well.

To tag the corpus with part-of-speech information, the latest version of the

CLAWS part-of-speech (POS) tagger designed by the University of Lancaster

(http://ucrel.lancs.ac.uk/claws/) was purchased under the academic license. In addition to

parts of speech, CLAWS tags punctuation marks and sentence boundaries. The credibility

of the tool is established by the fact that it was used to annotate one of the largest and best known corpora of English—the British National Corpus. According to the official

CLAWS website, the tagger has been significantly improved over the years, and now consistently achieves 96-97% accuracy. The example of the POS-tagged text is provided in Fig. 3.1.

103

Fig. 3.1 Example of CLAWS POS-tagged text

3.4.3 Data collection

This POS information was used to collect data on the use of cohesive devices and

prepositional phrases. The data were extracted using Access database queries. To analyze

other textual features such as nominalization, lexical density (type-token ratio), passives, and word/sentence length, two more tools were used—WordSmith Tools (Scott 2009) and Writer's Workbench. Academic licenses for both tools were purchased by the researcher.

Table 3.8 presents further details on data collection, explaining the calculations and tools used to track cohesive devices and other textual features in this study. The system of calculations has been adapted from Dong and Lan (2010). The explanations for

104

the tags referenced in Table 3.8 are provided in Table 3.9. It should be noted that

CLAW's "AT" tag includes both "the" and "no" as articles; for this reason, Disserata also counted the word "the" alone.

Table 3.8 Calculations used to track cohesive devices and textual features in the corpus (adapted from Dong and Lan 2010) Dong and Lan's model (2010) Calculations Cohesive devices Reference devices Pronominal devices (3rd person CLAWS pronouns only) (PPH1+PPHO1+PPHO2+PPHS1+PPHS2) Pronominal devices (possessive CLAWS (APPGE) pronouns) Demonstratives CLAWS (DD1+DD2) Definite article CLAWS (AT) or Disserata Comparative devices CLAWS (JJR+JJT*+RGR+RGT) Conjunction devices Additive devices CLAWS (СС) Adversative devices CLAWS (CCB) Causal devices (subordinate CLAWS (CS) conjunction) Temporal devices CLAWS (RT) Continuative devices (subordinate CLAWS (CS) conjunction) Global textual features Nominalization Writer's Workbench program Lexical density (standardized type/token Wordsmith Tools ratio)

105

Word length (average) Writer's Workbench program Passives Writer's Workbench program Prepositional phrases CLAWS (IF+II+IO+IW) Sentence length (average) Disserata * Note: In Dong and Lan's model, this tag is different (JK). However, in their explanation, they talk about superlative comparative devices (the CLAWS tag "JJT"). The "JK" tag refers to catenative adjectives ("able" in "be able to", "willing" in "be willing to"), which are not relevant for this study. For this reason, the appropriate substitution was made.

Table 3.9 CLAWS conventions for the tags used in the process of data collection CLAWS abbreviation Explanation Cohesive devices Reference devices PPH1 3rd person singular neuter personal pronoun ("it") PPHO1 3rd person sing. objective personal pronoun ("him," "her") PPHO2 3rd person plural objective personal pronoun ("them") PPHS1 3rd person sing. subjective personal pronoun ("he," "she") PPHS2 3rd person plural subjective personal pronoun ("they") APPGE possessive pronoun, pre-nominal (e.g., "my," "your," "our") DD1 singular determiner (e.g., "this," "that," "another") DD2 plural determiner (e.g., "these," "those") AT definite article ("the") JJR general comparative adjective (e.g., "older," "better," "stronger") JJT general superlative adjective (e.g., "oldest," "best," "strongest")

106

RGR comparative degree adverb ("more," "less") RGT superlative degree adverb ("most," "least") Conjunction devices CC coordinating conjunction (e.g., "and," "or") CCB adversative coordinating conjunction (e.g., "but") CS subordinating conjunction (e.g., "if," "because," "unless," "so," "for") RT quasi-nominal adverb of time (e.g., "now," "tomorrow") Textual features (prepositional phrases) IF "for" (as preposition) II general preposition IO "of" (as preposition) IW "with," "without" (as prepositions)

3.4.4 Data analysis: General procedures for descriptive statistical analysis and

significance testing

Upon extraction, the data was analyzed with statistical software SPSS. Each variable

included in this study has been post-processed in order to carry out a comprehensive data analysis. Since the texts were of variable length, the variables for each text were normed to represent the value per 1,000 words. To do so, each finding was multiplied by 1,000 and then divided by the actual number of running words in the text. Descriptive statistics were then calculated for each variable, and significance testing was performed to determine whether significant differences exist between the variables of the three populations for the main research questions, as well as for the genre comparison provided

107

in Chapter 4. The major research questions of this study are addressed in Chapter 5. This

study addresses three major research questions: