Texts for Translation into Chinese

RIKKYO UNIVERSITY

Graduate School of Intercultural Communication

Identification and Rewriting of Negative Translatability Indicators in Human Translation: Optimizing Japanese Hybrid

Texts for Translation into Chinese

by Beibei HE

Thesis for the degree of Doctor of Philosophy

2020

i

Abstract

Japanese (JA) to Simplified Chinese (ZH-CN) translation is in high demand in Japan's translation industry. In particular, hybrid texts that feature technical content with a marketing function have become a trending genre for which clients have high quality expectations. However, these expectations tend to be disappointed, not least when it comes to corporate social responsibility (CSR) reports, the hybrid texts which are taken as the basic data for this study. A major reason is the neglect of the importance of the textual component of internationalization, i.e., source text optimization for better translation quality.

Moreover, because of the social, industrial, educational, and institutional background in China, Chinese translators may not be competent enough to produce quality translations for non-optimized JA source content. This is backed by the fact that a high incidence of translation errors has been identified across the ZH-CN CSR reports published by large-scale Japanese companies.

From the viewpoint of audience design (Bell, 1984), the case for text adaptation for an audience of Chinese translators with the aim of improving translation quality becomes compelling: it further enables buyers to launch their localized products or release their localized linguistic content on time and with a quality level acceptable to their end-users or target readers.

Despite the growing conviction that optimized source content will help both translators and translation buyers, precisely how to optimize remains an area under- researched in the academic domain and a topic with which both language service providers and buyers are struggling, not least when it comes to the JA to ZH-CN language pair.

In response to the lack of a systematic method of JA source optimization for

quality improvement in the human translation (HT) setting, the first objective of this

ii

study is to establish a set of rewrite rules to remedy flaws in texts authored in JA. This entails developing a methodology for identifying problematic ST features that inhibit translation quality through analyzing the cause of translation errors. For cases in which translation errors are found to be induced by problematic source features, this study aims to establish a framework for eliminating such source features and thereby reducing translation errors, i.e., source optimization for improved translatability.

However, the first challenge is the absence of any error annotation scheme that is tailored for the JA to ZH-CN language pair, is customized for hybrid texts, and accounts for issues in both source texts (STs) and target texts (TTs). In other words, there is no applicable meta-language that can be used to identify the issues in JA to ZH- CN translations of hybrid texts.

An applicable error annotation scheme is necessary for assessing and defining quality in various scenarios in the translation industry. It is also essential for developing effective rewrite/pre-editing rules to improve, respectively, HT and machine translation (MT) quality. The reason is that the effectiveness of a rewrite rule (in both MT and HT settings) is often validated via comparative quality assessment of the translation of the original ST sentences and that of the rewritten/pre-edited ST sentences. Without an applicable error annotation scheme, it is impossible to conduct a quality assessment that will enable a useful summative or formative evaluation.

Thus, in response to the lack of an appropriate error annotation scheme, the second objective of this study is to develop a revision scheme—revision in this study covers both corrections and improvements as defined by Mossop (2014)—that accounts for both problematic features in STs and errors in TTs, tailored for the specific language pair and genre.

Chapter 1 explains the background to this study, its objectives, research questions,

and research methods. What CSR reports are and why they are chosen as the basic data

for this study are also described. In order to justify the approach taken to improve JA

iii

to ZH-CN HT translation of high-profile hybrid texts, the main applications of quality assessment in the translation industry (to the present author’s knowledge) are summarized; source text optimization as an established industry approach to achieving better translation quality is introduced; the recent integration of MT and HT into translation workflows is described; and translation competences, with special reference to the circumstances of translators in China working into ZH-CN, are outlined.

Chapter 2 first reviews the studies into JA to ZH-CN translation in relation to localization and translation quality assessment (TQA) and reveals the absence of prior research on which the current study could build. Secondly, prior studies of both academic-based and industry-based TQA approaches are reviewed, starting from relating translation theories on quality assessment to the two major criteria—accuracy and fluency—in many classic and contemporary TQA schemes. A gap is identified in that there exist no error schemes that account for problematic ST features, and most error schemes are English (EN) oriented. Therefore, this study is considered novel in undertaking systematic TQA research by developing a revision scheme that accounts for not only errors in TTs but also problematic features in STs and is tailored for ZH- CN translations of JA hybrid texts. Thirdly, there is a review of prior research into the key concepts of: internationalization, source (text) optimization, audience design, text adaptation, pre-editing, controlled languages (CLs) with a focus on JA-based CLs.

Special emphasis is given to negative translatability indicators (NTIs)—a notion coined by O’Brien (2005), defined as “a linguistic feature, either stylistic or grammatical, that is known to be problematic for MT” and borrowed for use in the HT setting in this study.

It emerges that there exist limited JA-based CLs compared to EN-based CLs, and very few precedents reporting the results of experiments on human translatability. No studies on JA-based CLs with a focus on HT into ZH-CN were found.

Chapter 3 describes the experiment designed for developing a revision scheme

that accounts for both TTs and STs, focusing on ZH-CN translation of CSR reports

iv

authored in JA. Firstly, two widely used error typologies—the ATA and DQF-MQM schemes—are mapped into a novel scheme while introducing new categories that exist in neither ATA nor DQF-MQM, taking into account the hybridity of CSR reports and the particularity of the language combination. A set of categories is established to label source issues and added as a component of the novel scheme, reflecting the fact that problematic source features may inhibit translation quality. Secondly, an experiment is conducted to validate the novel revision scheme, using a parallel corpus created from the CSR reports published by 13 Japanese companies. The validation process is a loop of iterative annotation, discussion, resolution, and refinement by three assessors. The iteration is terminated once a satisfactory agreement, i.e., moderate agreement (Cohen’s kappa 0.41 to 0.60), is achieved. Through this iterative error annotation process, the study develops a revision scheme that accounts for not only TT but also ST issues, customized for JA to ZH-CN translation, and optimized for hybrid texts. Identifying JA to ZH-CN translation errors in publicly available CSR reports using the customized scheme is a precursor to identifying ST features that correlate with errors in the translations.

Chapter 4 presents an analysis of the errors (1,127 in total) identified in the experiment conducted in Chapter 3. Through analysis of the identified errors, this study also proposes a set of rewrite/pre-editing rules corresponding to each NTI (i.e., a problematic source feature). As a result, 281 are found to be associated with NTIs. The NTIs considered to have triggered the most translation errors number approximately 142. The CL typology proposed by O’Brien (2013) is adopted to categorize these NTIs.

Of these 142 NTIs, 113 are found to instantiate 16 of O’Brien’s 61 sub-categories, 29 not corresponding to O’Brien’s subcategories because they include syntactic patterns and stylistic issue specific to JA texts.

In Chapter 5, a corpus-based experiment is designed to test the impact of each

NTI on example sentences extracted from the corpus. Based on a 500,000-character

v

collection of CSR reports, this study establishes the frequency of each NTI in the corpus.

Taking 38 NTIs as a test NTI set, sample sentences containing NTIs are then rewritten according to the rewrite/pre-editing rules. The impact of eliminating each NTI is evaluated by comparing human translations of the original (with NTIs) and rewritten (NTIs removed) sentences. The results show the relative effectiveness of the rewrite/pre-editing rules in terms of accuracy, i.e., the test NTI set improves accuracy in 39.64% of cases while degrading it in only 9.87% of cases. It also shows that the test NTI set improves and degrades fluency to a very similar extent, i.e., it has no clear positive or negative impact.

Since the corpus-based experiment tests the impact of the sample NTIs in isolation, Chapter 6 conducts a follow-up text-based pre-editing experiment to gauge the impact of the NTI set on running texts in order to better approximate the impact of rewriting the NTIs in a production scenario. As expected, the NTIs are not evenly represented within the randomly selected texts. Mirroring the results of the corpus- based experiment, rewriting of the running texts shows improvement in accuracy in 48.89% of cases on average, which is encouraging. However, no significant improvement on fluency is observed. It is noteworthy that the results of the corpus- based experiment and those of the text-based experiment are consistent.

Chapter 7 concludes the thesis by discussing the results of the previous chapters,

points out the major contributions and limitations of this study, and outlines future plans

towards HT-oriented JA-based CLs for improving ZH-CN translation quality more

widely. Revealing and practically useful answers are found to all research questions,

i.e., what types of error can be found in JA to ZH-CN translations of CSR reports, how

frequently they occur, how far they can be accommodated in an existing error

annotation scheme, how to achieve satisfactory inter-assessor agreement if customizing

a novel scheme becomes necessary, how to identify NTIs, how to establish the

frequency of each NTI, and how to test the effectiveness of eliminating NTIs.

vi

The most noteworthy contribution of this study to the field of TQA is the validation of a revision scheme (the CSR scheme) that accounts for not only TT errors but also problematic features of STs, providing a mechanism for describing what is wrong in the STs and differentiating the root cause. The additional originality is that the revision scheme is tailored for ZH-CN translations of JA hybrid texts. Another important contribution of this study to the field of CLs is that an NTI set consisting of a list of NTIs and their corresponding rewrite/pre-editing rules is formulated for the specific language-pair (JA to ZH-CN) and genre (hybrid texts). Furthermore, the annotated corpora created in Chapters 3 to 6 can be used for machine learning to enable automatic identification of problematic ST features. The corpus of original and rewritten JA texts represents a valuable resource that can be used as parallel/aligned data in machine learning for automatic text simplification.

It is hoped that this study has laid a solid foundation for two previously under- explored areas—JA to ZH-CN commercial translation quality assessment as well as HT-oriented source optimization—which will enable research to flourish in Japan and China.

Keywords: translation quality assessment, error annotation scheme, Japanese to

Chinese translation, source optimization, controlled language

vii

要旨

ソーステクスト（

）の最適化、つまり語彙や文構造を規定するルールセット（制限 言語）に基づくテクスト作成は、母語話者にとっての読みやすさをはじめ、機械翻訳（MT）

及び人手翻訳（

）の品質を向上させるのに有効とされているが、既存の研究の多くは 英語をベースとしたもので、日本語ベースの制限言語についての考察は限定的であるだ けでなく、

の品質を改善するためのものがほとんどである。また、既存の定量的品 質評価モデルの多くは英語およびヨーロッパ言語をベースとし、技術資料の翻訳に主眼 を置き、

に起因する翻訳エラーは存在しないという前提で作られているため、日本 語や中国語の特徴が考慮されておらず、技術的な要素に加えてマーケティング的な機能 も持つハイブリッドテクストの翻訳の評価には応用できず、

の問題まで分類できる 品質評価モデルは調査した範囲では存在しない。

本論文は、日中

の品質を向上させることを主眼とした日本語ベースの制限言語の 考案（第一の目的）と、日本語

の問題に起因する翻訳エラーを考慮し、ハイブリッ ドテクストの日中

に特化した品質評価モデルの構築（第二の目的）を行ったもので ある。

第

章では、研究背景や目的、研究方法を示し、

（企業の社会的責任）報告書を 本論文のデータ分析の対象とした理由を述べた。また、ハイブリッドテクストに着目し て、日中

における

の最適化と適用可能な品質評価モデルに関する考察の重要性 を喚起した。

第

章では、日中産業翻訳、品質評価についての翻訳理論とモデル、及び日本語ベー スの制限言語に関する先行研究を概観した後、正確性と流暢性が翻訳を評価するための 主要基準であることと、日中

における

の最適化および適用可能な品質評価モデ ルが学術的研究の対象とされてこなかったことを明らかにした。

第

章では、本論文の第二の目的に沿って、広く使われている

つの既存品質評価モ

デルをマッピングし、独自のエラーカテゴリを追加することによって、新たな品質評価

モデルを構築した。そのモデルを用い、一般公開されている

報告書の日本語版と

中国語版から作成したバイリンガルデータの品質評価を実施したところ、日本語

字で見られたエラーは

件であった。

viii

第

章では、本論文の第一の目的に沿って、第

章で特定されたエラーと

の言語 的特徴を関連付け、各エラーが

に由来するものかどうかを分析した。その結果、

件中、

件のエラーが

に由来するものと判定された。そして、翻訳エラーをもた らすと考えられる

の言語的特徴とそのリライトルールを組み合わせてルールセット を構築した。

続く第

章と第

章では、コーパスベースの実験とテクストベースの実験を通して、

第

章で構築したルールセットの有効性を検証した。具体的には、第

章で考案した品 質評価モデルを用い、リライト前後の日本語テクストの翻訳結果を比較評価した。この 評価では、文レベルにおける個々のルールの有効性を検証するとともに、文章レベルに おけるルールセット全体の有効性を検証した。その結果、正確性に関しては大幅な改善 が見られた一方で、流暢性に関しては著しい効果は見られなかった。流暢性に関しては リライトルールの改善や他のアプローチも必要と考えられるが、正確性の改善には現状 のルールセットでも一定の効果が現れるものと期待できる。

本論文の意義は以下の

点にある。すなわち、第一に、これまで学術的研究の対象と

されてこなかった日中産業翻訳に特化した品質評価モデルを構築したこと、第二に、日

中

の品質を向上させることを目的とした日本語ベースの制限言語はこれまでに存在

していなかったが、それを新たに考案したこと、第三に、他の研究者の参考となる具体

的な研究方法を示したこと、第四に、翻訳理論の観点から業界における品質評価の基準

を基礎付けたことである。

ix

Table of Contents

Abstract ... i

要旨 ... vii

Table of Contents ... ix

List of Tables and Figures... xii

Author’s Declaration ... xiv

Acknowledgments ... xv

List of Acronyms ... xvii

Chapter 1 Introduction ... 1

1.1 Background and motivation ... 1

1.1.1 Awareness of the scope of quality assessment in the translation industry .. 4

1.1.2 Source optimization as a means to achieve better translation outcomes ... 9

1.1.3 Integration of machine translation and human translation ... 10

1.1.4 Translation competences ... 14

1.2 Corporate social responsibility reports ... 20

1.3 Research questions, hypotheses, and methods ... 25

1.4 Chapter organization ... 29

Chapter 2 Prior Related Work ... 31

2.1 Studies into JA to ZH-CN translation ... 31

2.2 Translation quality assessment (TQA) ... 32

2.2.1 Accuracy and fluency as translation quality criteria ... 33

2.2.2 Translation theories on quality assessment ... 34

2.2.3 Academic-based TQA models ... 39

2.2.4 Industry-based TQA models ... 42

2.3 Controlled Languages and Negative Translatability Indicators ... 50

2.3.1 Internationalization/Source optimization ... 50

2.3.2 Controlled languages (CLs) ... 55

2.3.3 Japanese-based CLs ... 56

2.3.4 Negative Translatability Indicators (NTIs) ... 57

x

2.4 Summary ... 59

Chapter 3 Developing a Revision Scheme that Accounts for Both TTs and STs ... 61

3.1 Methods ... 62

3.2 ATA and DQF-MQM error schemes ... 64

3.3 Mapping of the ATA and DQF-MQM error schemes ... 66

3.4 Validating and updating the novel revision scheme ... 74

3.4.1 Experimental setup: task design and implementation ... 75

3.4.2 Results ... 77

3.5 Summary ... 91

Chapter 4 NTI Identification and Rewrite/Pre-editing Rule Formulation ... 94

4.1 Methods ... 94

4.2 Formulation of NTIs and rewrite rules ... 97

4.3 Comparison with Acrolinx ... 120

4.4 Summary ... 121

Chapter 5 Corpus-Based Rewriting/Pre-editing Experiment ... 123

5.1 Methods ... 123

5.2 Experimental setup: task design and implementation ... 130

5.3 Results and analysis ... 135

5.4 Summary ... 146

Chapter 6 Text-Based Rewriting/Pre-editing Experiment ... 148

6.1 Methods ... 148

6.2 Experimental setup: task design and implementation ... 150

6.3 Results and analysis ... 154

6.4 Summary ... 166

Chapter 7 Conclusions ... 168

7.1 Review of research objectives, approach, and findings ... 168

7.2 Contributions ... 174

7.3 Limitations and future research ... 175

Appendices ... 180

Appendix 1: O’Brien’s (2003) CL rule categories and explanation... 180

xi

Appendix 2: NTIs with associated issues and rewrite rules ... 182

Appendix 3: CQL/Regular expressions and frequency of search result ... 210

Appendix 4: Accuracy evaluation results of the corpus-based experiment ... 215

Appendix 5: Fluency evaluation results of the corpus-based experiment ... 218

References ... 221

xii

List of Tables and Figures

List of Tables

Table 1.1 Example of Google Translate from JA into EN and ZH-CN ... 12

Table 1.2 Chapter organization with research questions ... 30

Table 3.1 Mapping of ATA and DQF-MQM ... 68

Table 3.2 ST categories in the CSR scheme ... 74

Table 3.3 Comparison between the initial and final CSR schemes (TT) ... 78

Table 3.4 Improvement of IAA level with training over time ... 84

Table 3.5 Type and frequency of agreed TT errors ... 87

Table 3.6 Error distribution in each extract ... 88

Table 3.7 Type and frequency of agreed ST issues ... 89

Table 3.8 An example of ST- Incomprehensibility... 90

Table 4.1 NTI identification example ... 95

Table 4.2 O’Brien’s (2003) CL rule classification framework ... 96

Table 4.3 NTIs allocated based on O’Brien’s CL rule classification ... 98

Table 4.4 NTIs not accommodated by O’Brien’s typology ... 100

Table 4.5 Example of NTIs and the rewrite rule (syntactic) ... 102

Table 4.6 Example of the NTI “Modifier is too long” ... 104

Table 5.1 Sample of CQL/regex and frequency of search result ... 127

Table 5.2 38 NTIs tested in this experiment ... 131

Table 5.3 Accuracy evaluation interface ... 134

Table 5.4 Fluency ranking evaluation interface ... 135

Table 5.5 Accuracy evaluation result of NTIs No. 1 to No. 3 ... 137

Table 5.6 Accuracy evaluation results summary (all 38 NTIs) ... 137

Table 5.7 Fluency evaluation results of NTIs No.1 to No.3 ... 138

Table 5.8 Positive, negative, and neutral impacts of rewriting each NTI ... 139

Table 5.9 Ratio of positive, neutral, and negative results ... 142

Table 5.10 Ineffective NTIs in the accuracy and fluency evaluation results ... 143

Table 5.11 Inter-assessor agreement (IAA) in fluency evaluation ... 144

xiii

Table 5.12 Accuracy evaluation results of NTIs No.3, No. 29, and No. 30 ... 144

Table 6.1 A sample of working files used for rewriting/pre-editing ... 151

Table 6.2 Translation assignment in the text-based experiment ... 151

Table 6.3 Accuracy evaluation example in the text-based experiment ... 152

Table 6.4 Fluency ranking example in the text-based experiment ... 154

Table 6.5 Accuracy evaluation results of TT1-o and TT1-r ... 156

Table 6.6 Fluency ranking results of TT1-o and TT1-r ... 158

Table 6.7 Accuracy evaluation results of TT2-o and TT2-r ... 160

Table 6.8 Fluency ranking results of TT2-o and TT2-r ... 161

Table 6.9 Normalized score of the accuracy and fluency evaluation results .... 162

Table 6.10 Normalized comparative scores for accuracy and fluency (Text 1) . 162 Table 6.11 Normalized comparative scores for accuracy and fluency (Text 2) . 164 Table 6.12 Number and percentage of positive, neutral, and negative results ... 165

Table 6.13 Average percentage of positive, neutral, and negative results ... 166

List of Figures Figure 1.1 Content journey from content authoring to final translation delivery .. 6

Figure 1.2 Incorporation of pre-editing in the translation workflow ... 28

Figure 2.1 Movement from summative to formative TQA ... 49

Figure 3.1 Example of parallel texts in Excel for quality assessment ... 63

Figure 3.2 Example of explanation of error categories ... 64

Figure 3.3 Decision tree ... 82

Figure 3.4 Inter-assessor agreement (Cohen’s kappa) ... 85

Figure 5.1 Search result for the Japanese word “ ^イメージ ” ... 126

Figure 5.2 Relative effectiveness of each NTI on both accuracy and fluency ... 140

xiv

Author’s Declaration

I, Beibei He, declare that the thesis entitled “Identification and Rewriting of Negative Translatability Indicators in Human Translation: Optimizing Japanese Hybrid Texts for Translation into Chinese” and the work presented in it are my own.

I confirm that:

・this work was done wholly or mainly while in candidature for a research degree at this University;

・ where any part of this thesis has previously been submitted for a degree or any other qualification at this University or any other institution, this has been clearly stated;

・ where I have consulted the published work of others, this is always clearly attributed;

・ where I have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work;

・ I have acknowledged all main sources of help;

・where the thesis is based on work done by myself jointly with others, I have made clear exactly what was done by others and what I have contributed myself;

・parts of this work have been published as:

He, B. (2018). Developing and validating a revision scheme to account for both source and target texts: The case of Japanese to Chinese translation of hybrid texts. T&I Review, 8, 87-118. Ewha Research Institute for Translation Studies, Seoul, Sourth Korea.

He, B. (2019). Identification and rewriting of negative translatability indicators in human translation: Pre-editing Japanese hybrid texts for translation into Chinese.

Intercultural Communication Review, 17, 19-39. The Graduate School of Intercultural Communication, Rikkyo University, Tokyo, Japan.

Signed:

Date:

xv

Acknowledgments

Throughout this long journey, I have been guided and supported by so many people, without whom this dissertation would not have been possible.

Firstly, I would like to extend my deepest gratitude to my thesis advisor, Professor Kayoko Takeda. I would have not started the PhD study if it were not for her encouragement to me when I was a masters student. It has been a slow progress for me in the first three years because life can be quite difficult sometimes, and a pursuit can become a second priority. However, she has always been patient and supportive in reassuring me that my research was meaningful and that I should never give up. Her guidance on how to conduct research scientifically and write a thesis properly has made me understand the importance of rigor. Her wealth of knowledge and expertise has always been a valuable resource for me in seeking advice and inspiration.

My profound appreciation goes to Professor Anthony Hartley for the continuous support of my PhD, for his patience, inspiration, and immense knowledge. His guidance helped me throughout the duration of research and writing of this thesis. Everything I have learned from him during the process has empowered me to achieve success in my profession as a business consultant in the translation industry. His passion in teaching and his generosity in advising will always be the benchmark for me when I am given an opportunity to share my knowledge and experience with others.

My sincere thanks also go to all the professors, especially Professor Yoko

Nadamitsu, in the Graduate School of Intercultural Communication of Rikkyo

University for their precious comments in the interim reporting meetings held every

year where all PhD students share their progress with professors from other domains

such as Intercultural Communication. I am also deeply obliged to Associate Professor

Kayo Matsushita and Associate Professor Yuko Shinohara for their valuable advice and

feedback during the preliminary review. Many thanks also to Professor Masaru Yamada

xvi

for all the support and encouragement he gave me, especially at the beginning of this journey.

I cannot thank enough the two accuracy evaluators, four translators, three fluency evaluators, and one rewriter who participated in the experiments for their commitment despite the fact that they all have a busy day job, and for all the fun we have had during the experiments. I am aslo grateful to my husband, a professional English to Japanese translator for 20 years, whose input on how to analyze Japanese was a great help for me to understand and articulate why it is difficult for translators to perceive and translate Japanese in specific scenarios.

I also wish to express my sincere gratitude to my colleagues at SDL, who have been generously sharing their experience and knowledge with me so that I could be certain that this research is grounded in facts perceived by a wider community and would be useful for the industry.

Finally yet importantly, I would like to thank my nine-year old daughter, Aira,

who has been extremely patient and supportive. It was at the time when she was born

that I started the journey from my masters study at Rikkyo. All I remember now are the

images of her playing or drawing quietly next to me while I was working or researching,

and her giving me a little massage to my shoulders with her little fingers for the

sleepless nights. Her advice on health maintenance and never-give-up spirit was

without doubt priceless to me.

xvii

List of Acronyms

ATA American Translators Association CL Controlled Language

CAT Computer-Assisted Translation CSR Corporate Social Responsibility CQL Corpus Query Language

EMT European Masters in Translation

EN English

DQF Dynamic Quality Framework

GMX-Q Globalization Metrics Exchange-Quality HOCL Human-Oriented Controlled Language HT Human Translation

JA Japanese

JTF Japan Translation Federation IAA Inter-Assessor Agreement

ISO International Organization for Standardization LISA Localization Industry Standards Association LSP Language Service Provider

MOCL Machine-Oriented Controlled Language MQM Multidimensional Quality Metrics MT Machine Translation

NTI Negative Translatability Indicator POS Part of Speech

RQ Research Question

SAE Society of Automotive Engineers

ST Source Text

xviii

STJ Simplified Technical Japanese

TAUS Translation Automation User Society TQA Translation Quality Assessment

TSP Translation Service Provider TM Translation Memory

TT Target Text

TTQA Total Translation Quality Assessment

ZH-CN Simplified Chinese

1

Chapter 1 Introduction

This study addresses limitations in optimizing source texts (STs) for human translation (HT) and in translation quality assessment (TQA) within Japan’s translation industry, focusing on Japanese (JA) to Simplified Chinese (ZH-CN) translation of hybrid texts. This introductory chapter first describes the background to this study, including the scope of quality assessment in the translation industry, the notion of source optimization, integration of machine translation (MT) with HT, and translation competences. Section 1.2 introduces the corporate social responsibility (CSR) reports which form the basic data for the study, and Section 1.3 sets out the research questions.

Section 1.4 explains the chapter organization of the thesis.

1.1 Background and motivation

In Japan’s translation industry, ZH-CN has become the second biggest (in volume) target language after English (EN) for translation from JA. According to the Translation Industry Questionnaire Survey for 2017 conducted by the Japan Translation Federation (JTF, 2018), 67.8% of translation agencies provide JA to ZH-CN translation services (sample size=276).

However, based on the present author’s experience as a JA to ZH-CN translator and linguistic solution consultant working in the industry, the quality of draft ZH-CN translations of JA tends to entail a time-consuming revision process that involves client reviewers; tellingly, the number of client revisions and comments is often considered as a major indicator of translation quality.

The quality issue becomes even more obvious when it comes to hybrid texts

which have both technical content and a marketing function, a trending genre in the

translation industry. Texts that have more than one function are referred to as hybrid

texts in both Translation Studies (Munday, 2016, pp. 114-117) and in the translation

2

industry. In its April 2018 journal, Amelia, a well-known translator network in Japan, highlighted this trending genre by featuring interviews with language service providers (LSPs) in Japan, such as SDL and Lionbridge

, on translation as content marketing.

According to Amelia, the need for translating content for promoting product sales has been increasing over the past few years, creating a new and promising domain with an under-supply of competent translators, as noted by JTF. The volume of translated promotional material is relatively limited compared to that of manuals, but clients tend to have high expectations towards translation quality for promotional materials because they have a direct influence on sales performance. According to the Amelia interviews, typical hybrid source types are articles and videos for presenting products, application examples, advertisements, and E-learning content.

This study considers the text type of corporate social responsibility (CSR) reports as hybrid, as justified further in Section 1.2. Moravcikova, Stefanikova, and Rypakova (2015, p. 338) point out that customers are interested in such CSR activities that reach out to the community, business partners, employees, and the environment, and customers prefer buying a product from a company that is socially responsible. This means that in providing information on the actions implemented by companies regarding their contribution to the principles of sustainable development, a CSR report affects general perceptions of a company’s brand, products, and social identity. In other words, a CSR report has a marketing function. Consequently, LSPs are expected to deliver high quality translations of CSR reports that fulfill not only an informative function but also expressive and operative functions (see Section 1.2 for a discussion of functions of language and hybridity). Moreover, HT remains the preferred approach to providing optimal quality despite the fact that MT followed by post-editing has

1 According to CSA Research—Common Sense Advisory, an independent market research firm with a focus exclusively on the business of globalization, localization, interpreting, and translation—

Lionbridge is the world’s second largest LSP and SDL the fourth based on verified and validated 2018 revenues. See https://csa-research.com/More/Featured-Content/Global-Market-Study/Top-100- LSPs for details.

3

become mainstream in the localization industry (see Section 2.3.1 for the definition of localization).

This study presents the analysis of 13 publicly available ZH-CN translations of JA CSR reports through translation quality assessment performed by three independent, trained assessors. The result reveals a high incidence of translation errors across all the ZH-CN CSR reports published by different companies, demonstrating that there is a systemic quality issue in JA to ZH-CN translation, which implies that the average translation competency (see a discussion of the PACTE competence model in Section 1.1.4) of ZH-CN translators is far from satisfactory. What also emerges is that JA to ZH-CN translation quality issues in this hybrid genre originate to a significant degree in problematic features of STs, indicating there is a systemic issue in JA authoring.

Motivated by the presence of errors in the target texts (TTs) and problematic features in the STs, the first objective of this study is to develop a methodology for identifying problematic source features that inhibit translation quality through analyzing the cause of translation errors. For cases in which translation errors are found to be induced by problematic source features, this study aims to establish a framework for eliminating such source features for the purpose of reducing translation errors, i.e., source optimization for improved translatability. This study terms the relation between translation errors and problematic source features overt when the ST is deficient even from the viewpoint of JA native speakers, and latent when the ST is difficult to translate but understandable to JA native speakers. The focus of this study is on the latter.

As a precursor to identifying latent ST features that correlate with errors in the

translations, this study develops a novel revision scheme that accounts for both

translation errors and overt problematic source features. Mossop (2014, p. 1) defines

revising (or revision) as “reading a translation in order to spot problematic passages,

and making any needed corrections or improvements.” This study adopts Mossop’s

definition of revising that covers both corrections and improvements and thus prefers

4

the label revision scheme to the more common error scheme or error typology. It refers to problematic text spans (a continuous/discontinuous sequence of words/characters) in TTs as translation errors and to problematic text spans in STs generally as features, more specifically as NTIs (negative translatability indicators, explained in Section 2.3.4). Both errors and NTIs involve “checking linguistic correctness as well as the suitability of a text’s style to its future readers and to the use they will make of it”

(Mossop, 2014, p. 1). Therefore, categories in the customized revision scheme cover not only errors in TTs but also overt NTIs in STs, while latent NTIs are to be identified through translation errors in TTs.

The present author brings to this study her experience as a professional JA to ZH- CN translator (i.e., 12-year preservation of knowledge of problematic JA source features that inhibit translation into ZH-CN), a language solution consultant (i.e., knowledge of the impact of ST quality on the end-to-end localization process), and a researcher (i.e., knowledge of research methodologies, frameworks, and meta- language). In order to explain the background relevant to this study and justify the approach taken to improve JA to ZH-CN translation of high-profile hybrid texts, Section 1.1.1 describes, to the best of the present author’s knowledge, the main application of quality assessment in the translation industry. Section 1.1.2 introduces source optimization, which is an established industry approach to achieving better translation quality. Section 1.1.3 explains the recent integration of MT and HT into translation workflows, while Section 1.1.4 discusses translation competences, with special reference to the circumstances of translators in China working into ZH-CN.

1.1.1 Awareness of the scope of quality assessment in the translation industry

As Lommel (2016, p. 1) has asserted, translation quality is “notoriously” difficult

to define and can cause debate and dispute between LSPs and buyers. According to

Lommel, when LSPs and buyers talk about quality, they may be referring to different

5

concepts of quality. As a result, if LSPs’ deliverables do not match what the customers want—which may not have been made clear in the beginning—it may result in wasted effort, lost profits, or end of the partnership. Unfortunately, in Japan LSPs and buyers often do not explicitly agree on quality expectations or how to measure the quality of the ST before starting translation. Based on the present author’s observations in the translation industry, translation buyers in Japan tend overwhelmingly to assume it is translators’ job to produce a satisfactory translation regardless of whether or not the provided STs are problematic, such that there is little room for negotiation on compromised quality or increased time and cost.

However, through experience, both LSPs and buyers have started to realize that they need to sell or buy translation services based on specific quality levels (e.g., one cannot expect to buy a quality news release translation at the rate applied for translating daily internal emails). What is also observed is the growing awareness of the need to agree on a common definition of quality in order to avoid debates and disputes that arise from the common case that quality assessment relies either on individual client reviewers who know their content but are not necessarily familiar with translation or on third party assessors. So, what are the conditions for achieving agreement on this issue?

Sager (1983, p. 121), an academic writing from an operational perspective,

emphasizes that it is necessary to consider the process of translation as an industrial

process when discussing quality as time and effort need to be considered when

measuring a translator’s work. Figure 1.1 presents a simplified but typical workflow of

a document localization process. The wheel indicates that: (1) quality management

starts from content creation, even though it is observed that little attention has been paid

to problematic ST features when authoring JA source content; and (2) if issues in STs

are identified in the later stages through problems found in TTs, it is necessary to go

back to the beginning and re-do everything, which entails increased time and cost.

6

Figure 1.1 Content journey from content authoring to final translation delivery

Moreover, Geoffrey, Fields, Hague, Lommel, and Melby (2014, p. 416) contend that a quality translation should deliver the accuracy and fluency required for a given audience and purpose. It should also comply with all other specifications negotiated between the requester and provider, taking into account end-user needs. This contention is hardly new. Sager (1983, p. 121) argued that “there are no absolute standards of translation quality but only more or less appropriate translations for the purpose for which they are intended.” But according to Lommel (2016, p. 4), this definition has gained support from both sides only in recent years. Thus, it is increasingly accepted that there is no absolute definition of quality, and how to measure the “relative excellence” of a translation is subject to “human terms” such as the prevailing constraints on production (e.g., time, cost, and technology) and the agreed purpose of the translated text (Pym, 2019, p. 437).

Quality assessment in a commercial setting as explained by Lommel (2016, p. 4)

refers to the after-the-fact measurement of adherence to a specification. A typical way

of assessing quality is to use a scorecard to examine a translation and generate a score

7

for how well it meets the quality requirements. The score is generated based on error type, error count, and error weight. Errors are labeled for type and weighted for severity of impact based on a pre-defined error typology specified on the scorecard. How such error typologies are used in an HT process (based on the present author’s experience) is summarized as follows.

First, many LSPs and buyers have started to tie a quality level to an acceptable percentage of errors in the total word count based on an agreed error typology. Cited by Pielmeier and Lawrence (2014), Gengo (a wholly owned subsidiary of Lionbridge as of December 2018) has adopted this approach to defining the different quality levels they are offering—using numerical scores to distinguish quality levels. This allows buyers who are not familiar with translation errors to understand the difference in a quantitative manner. This appears to be the industry’s preferred way of quantifying quality levels at present. However, which error typology to use for a particular language pair and content type, how to decide the appropriate error rate for a specific quality level, how to train linguists to apply the error typology, and how to ensure reliable inter- assessor agreement all remain open questions.

Second, more and more LSPs have started to use error typologies to assess how good a translator is in translating different types of content (or genres) (e.g., legal, marketing, IT, and so forth) through trial translations. A score is needed to label the competence of the translators for objectivity and comparison when LSPs possess a big pool of translators. Scores are calculated separately for each content type, and LSPs will decide which content type(s) a translator who took the test is good at, and enter that information into their supplier database.

Third, quantifying errors identified by a client reviewer is a way for LSPs to

defend their quality as client feedback (i.e., client reviewers’ revision details and the

volume of their revisions) is often the means by which a client rates an LSP. As pointed

out by Pym (2019, p. 445), “[i]n principle, clients are the first recipients of translations,

8

since they are the ones who have to be satisfied in order to pay for them and to keep paying for more. Their perception of quality is thus of key importance.” Since client reviewers are often not linguists, it is inevitable that LSPs receive a significant number of revision comments on their deliverables. Due to limitations of their own language skills or their lack of understanding of the source language, client reviewers may contradict their own revisions, violate style guides and glossaries, change a 100%

translation memory (TM) match

with little knowledge of how a TM works, introduce spelling errors, or even change a good-enough translation into an awkward expression.

Taking such situations into account, it is considered best practice in the industry to ask a third-party translation vendor or a translator other than the one who did the job to objectively label each client revision based on a pre-defined error typology. The reason for doing so is to allow LSPs to negotiate with clients based on facts, that is, numbers of errors and non-errors (changes due to personal stylistic preferences of the reviewer or to deficiencies in the ST), error types, error severity, and an overall score, which can be compared to industry standards. If the result is indeed below industry standards, it can help both clients and buyers to understand where the issue is and to think about what preventive actions can be taken to improve the situation. In sum, a quantified evaluation can help all stakeholders to make informed decisions, but the approach itself is time-consuming and costly. More importantly, it relies on the reliability of assessors, which can be an issue because translators are not necessarily trained assessors.

Finally, monitoring error rates is a means of quality management which requires constant attention from both buyers and LSPs. Mature

buyers who have integrated translation into the overall production process tend to perform periodic quality

2 Translation Memory (TM) is a database which stores translated information as the translator works and allows a translator to re-use any of the text segments that have been translated fully or partially before. A 100% match is identified when there is a segment in a new document that is identical to a segment stored in the translation memory.

3 See Ray and DePalma (2017) for the Localization Maturity Model from Common Sense Advisory, Inc.

9

evaluations on sample projects in order to rate their LSP(s). Such evaluation results can guide buyers to decide how to distribute their translation needs, how to manage their vendors, and how to take corrective/preventive actions.

The wide range of application of error typologies just described demonstrates that error typologies are used extensively in the translation industry at various stages and for various purposes. This is indicative of a need to develop practically applicable error typologies for different contexts. What the present study aims to develop is an error typology that (1) is applicable for JA to ZH-CN translations of hybrid texts, and (2) accounts for both source and target texts.

1.1.2 Source optimization as a means to achieve better translation outcomes

According to Lommel (2016, p. 5), there are six areas to which buyers and LSPs should pay attention in order to ensure quality: (1) terminology; (2) style guide; (3) source content optimization; (4) specifications and project requirements; (5) continuous improvement and source quality; and (6) metrics. What is noteworthy is his emphasis on the source side in four of the six areas. To be specific, for terminology he suggests that buyers need to manage and control terminology in the source; for style, he recommends buyers to develop clear and actionable style guides and to enforce adoption in the source; for source content optimization, buyers need to implement continuous improvement practices to improve the source and invest in technologies to control the source. Moreover, he advises LSPs to provide feedback about problems that come from the source, such as inconsistencies, terminology problems, or lack of clarity.

For continuous improvement and source quality, he recommends that buyers develop a

process to take LSP feedback into account and make changes, while suggesting to LSPs

to report specific and actionable feedback about the source content and any problems

it may cause.

10

Based on the present author’s experience, what Lommel (2016) itemizes are precisely those areas where both buyers and LSPs are struggling and looking for solutions and best practices, given that source quality and its translatability impact the end-to-end translation process. If a source error or a translation error (multiplied by the number of target languages, potentially a large number) occurring due to some ambiguity in the source is detected in the subsequent processes, both buyers and LSPs will be forced to go back to square one to repeat the whole cycle, which often includes content update, translation and review, publishing (several rounds), layout check, client review, and release (see Figure 1.1). It is no longer an option for buyers to expect translators to deliver perfect translations without considering whether the source content is translation-friendly. The reason is that buyers often have a fixed release deadline and so they cannot afford increased time even if they can bear increased costs.

What needs to be stressed is that translated content is not in itself the goal of the buyer.

The goal is launching a product or releasing information on time and within budget.

1.1.3 Integration of machine translation and human translation

Machine translation quality has reached a higher plateau in recent years due to the emergence of neural machine translation (NMT) since it entered the public domain in 2016. DePalma and Lommel (2018, p. 7) summarize the development of NMT as follows. (For details regarding traditional MT models and MT in general, see Arnold, Balkan, Lee Humphreys, Meijer, & Sadler, 1994; Hutchins & Somers, 1992; Koehn, 2010; Nirenburg, Somers, & Wilks, 2003; Sin-wai, 2015.)

Machine translation based on neural networks is the newest entrant to the field.

These systems use machine-learning techniques that emulate the behavior and

function of biological neurons. The theory is that this approach more closely

reflects how human translators perform their tasks. These techniques advanced

11

rapidly in the last two years, and now dominate MT development and commercial deployments. (DePalma & Lommel, 2018, p. 7)

From a linguistic point of view, what makes NMT most promising is that it can produce more natural-sounding translations. Compared to Statistical MT (SMT), NMT rates much higher in terms of fluency, which increases user satisfaction with its output (DePalma & Lommel, 2018, p. 7). The shortcoming of NMT, on the other hand, is that debugging and troubleshooting is almost impossible, and NMT technology remains unproven, thus still unreliable.

Nevertheless, the integration of NMT and HT is a growing trend in the translation industry. There are two main scenarios: (1) an MT system translates the source content and humans post-edit (PE) the MT output (MT works as translators, humans as the revisers, a combination known as MTPE); and (2) humans translate source content referring to MT suggestions (humans work both as translators and revisers). But whichever the scenario, what matters the most to human translators is the unit price per source word or character.

In the first scenario, there are mainly two ways for LSPs to decide the translation

rate: (1) some LSPs might request translators to perform a sample check or test on MT

output and establish if translators are willing to accept the discount rate proposed by

the LSP. If not, either a revised discount rate can be agreed through discussion or the

translators will not work for a particular project/account; and (2) some LSPs simply

present the discount rate to the translators, and no negotiation is possible. Translators

will have to decide whether to sell their services at a lower rate irrespective of whether

MT can improve their productivity to a reasonable level. As for how a discount rate is

decided, LSPs often need to evaluate the content to be translated and the language pairs

first to decide if MT is suitable for a particular job. If the evaluation result turns out

12

positive, LSPs will further perform productivity tests with their language teams in order to set a discount rate by language pair.

It is widely believed in the translation industry that very few NMT systems support JA to ZH-CN translation to a PE threshold level at the moment. Based on the present author’s experience, when JA to ZH-CN is the language pair and the quality expectation from the buyer is high, the MTPE process will not be recommended by LSPs as it is unlikely to be able to improve productivity, nor are LSPs willing to promise

“perfect” quality at a discounted price. (See Koponen (2016) for the general trend of MTPE research, and Yamada (2019) for productivity gains and cognitive effort associated with PE of NMT.)

As for the second scenario where humans work both as translators and revisers with MT output as a reference, when JA is the source and ZH-CN the target and quality translation is expected for content such as CSR reports, the MT result is, again, often of no use, especially when the source JA sentences are long or ill-structured. Table 1.1 shows the EN and ZH-CN Google Translate results of a sample sentence from a JA CSR report.

Table 1.1 Example of Google Translate from JA into EN and ZH-CN a. JA sentence from a CSR report

XXX

は、写真分野で培ったコラーゲン技術を基盤に開発した、細胞培養に必要な細 胞外マトリックス（リコンビナントペプチド：RCP）「cellnest」を研究用試薬として 発売しているほか、

年当時、国内で唯一、自家培養表皮や培養軟骨等の再生医療 製品を上市していた

を連結子会社化。

b. Google translation of a to EN (as of March 21, 2020)

XXX has released cellnest, an extracellular matrix (recombinant peptide: RCP)

required for cell culture, developed as a research reagent, based on collagen technology

13

cultivated in the field of photography. Only YYY, which marketed regenerative medicine products such as self-cultured epidermis and cultured cartilage, became a consolidated subsidiary.

c. Google translation of a to ZH-CN (as of March 21, 2020)

XXX 发布了 cellnest，这是细胞培养所需的细胞外基质（重组肽：RCP），它是根据摄 影领域培养的胶原技术开发的研究试剂。只有销售自养表皮和软骨的再生医学产品的 YYY 成为合并子公司。

d. Google translation of c to EN for comparison with b (as of March 21, 2020) XXX released cellnest, an extracellular matrix (recombinant peptide: RCP) required for cell culture, which is a research reagent developed based on collagen technology cultivated in the field of photography. Only YYY, which sells regenerative medicine products of autotrophic epidermis and cartilage, becomes a consolidated subsidiary.

The Google NMT output of a JA sentence from a CSR report shown in Table 1.1

demonstrates that: (1) it would be less confusing and straightforward for translators to

translate from scratch without struggling to work upon a bad ZH-CN NMT result; and

(2) the ill-structured JA has a bad impact on EN translation as well. In any case, it will

take translators considerable time to read such a long source sentence and make sense

of the complicated structure. This shows the utility of the suggestions (already

summarized in Section 1.1.2) that Lommel (2016) makes to buyers on source content

optimization to achieve better outcome, even if they are general suggestions that are

not specific to either HT or MT. By definition, post-hoc solutions are contingency

measures which cannot improve the quality of translation from JA systematically. JA

source optimization through a joint effort of buyers and LSPs (possibly, academics as

well) is the solution calculated to achieve long-term success. This is echoed by Kelly

(2010, p. 1) who observes that “translation and localization professionals often point

out that quality starts at the source.”

14

Back to the integration of MT and HT which is predominating the workflow in the translation industry day by day, Yamada (2018, para. 1) has pointed out that the assumption behind the MT+PE model is that the competences of the machine and those of the human are complementary, but this assumption is no longer valid. The reason is that technological advances have created NMT systems that more closely mimic HT, including human-like errors, “thus increasing the effort required for post-editors to detect and correct errors” (para. 1). Yamada (2019, p. 87) also concludes that NMT+PE does not contribute to reducing the post-editors’ cognitive effort or error correction rate and that NMT+PE requires almost the same competence as ordinary HT. Although Yamada’s (2019) experiment involved college students who might arguably be less familiar with the task, the present author observes that the NMT+PE model poses the same challenges in a professional setting.

1.1.4 Translation competences

PACTE has been a competitive research group since 1997 and a consolidated

group (as recognized by the Generalitat de Catalunya) since 2002. The PACTE group

has been carrying out empirical-experimental research into translation competence and

its acquisition in written translation since October 1997. Analyzing data obtained from

two experimental groups (35 expert translators and 24 foreign-language teachers),

PACTE (2009) defines translation competence as the underlying system of knowledge

required to translate. According to the PACTE competence model, translation

competence: (a) is expert knowledge; (b) is predominantly procedural; (c) comprises

different inter-related sub-competences; and (d) includes a strategic component which

is of particular importance. In addition, the degree of expertise in translation is believed

to be reflected in both the process and the product of translation, and the definition of

expertise is based on: (a) years of experience as a translator; (b) translation as a main

source of income; and (c) experience in translating a wide range of texts (PACTE, 2009,

15

pp. 208-209). As its indicator of translation competence, PACTE adopted acceptability to define translation product quality based on whether or not a translation effectively communicates (a) the meaning of the ST; (b) the function of the translation; and (c) makes use of appropriate language (PACTE, 2009, pp. 210-217).

Thus, the acceptability of the translations produced by the college students employed by Yamada (2019) was expected to be low compared to those of professional translators as they were thought to need more training, experience, and self-learning to gain the competences just described. Based on PACTE’s model of translation competence and translation competence acquisition, Castillo (2015) conducted an experiment to observe how the acceptability of translations evolves throughout the process of acquiring translation competence, employing translation students and recent graduates as participants with a focus on EN to Spanish translation. The preliminary results showed no firm conclusions. Employing professional translators for experimental research in an academic setting is considered to be difficult as it entails greater cost even if the result approximates better to reality.

The EMT expert group was set up by the Directorate-General for Translation in April 2007. Its main task is to make specific proposals with a view to implementing a European reference framework for a Master's in translation (European Master's in Translation – EMT) throughout the European Union. EMT (2009) defines competence as the combination of aptitudes, knowledge, behavior, and know-how necessary to carry out a given task under given conditions. Moreover, this combination needs to be recognized and legitimized by a responsible authority such as an institution or an expert.

EMT (2009) proposed six types of competence: translation service provision

competence, language competence, intercultural competence, information mining

competence, thematic competence, and technological competence. Knowing the

possibilities and limits of the MT+PE model (as discussed in Section 1.1.3) is

considered as part of the technological competence (EMT, 2009, p. 7).

16

ISO 17100:2015 is an internationally recognized standard for translation services and was created to respond to the market need for a universally accepted benchmark. It sets standards for achieving quality by following the translation process steps it defines.

According to the standard, translators should have six types of professional competences: (a) translation competence; (b) linguistic and textual competence in the source language and the target language; (c) competence in research, information acquisition, and processing; (d) cultural competence; (e) technical competence; and (f) domain competence. As for translator qualifications, the same standard requires that translators should meet at least one of the following criteria by having:

a) a recognized graduate qualification in translation from an institution of higher education;

b) a recognized graduate qualification in any other field from an institution of higher education plus two years of full-time professional experience in translating;

c) five years of full-time professional experience in translating.

The present study conducted three experiments that employed Chinese

professional translators for translation and quality assessment tasks. All of these

translators met either criterion b) or c) stated above. Below is a summary of the social

background to the China translation industry revealed in a seminar organized by Japan

Translation Federation (JTF, 2012), which attributes the difficulties of quality

management in China’s translation industry to translators’ lack of training, social status,

and remuneration. These shortcomings have a detrimental impact on Japan’s translation

industry since translation companies in Japan often work with translation companies or

translators based in China for the purpose of cost reduction. To the present author’s

17

knowledge, China branches of many Japanese enterprises also employ local translation vendors or in-house translators for their translation needs.

a) Lack of social status: Translation Studies is yet to be established as an academic discipline in China, and the social status of translators is extremely low. The importance of translation/interpreting is not acknowledged by society. Translation is deemed to be a practice of replacing STs with TTs on a word level, and anyone can translate if there is a dictionary at hand.

b) Lack of remuneration: translation as a profession is yet to be established in China. As a profession, there is no clear line between translation and interpreting. The same word ^翻译 can be used to refer to both translators and interpreters. Most importantly, one cannot make a living if working only as a translator. In most cases, part-time translators have other professions as their daily job.

c) Lack of management: most translation agencies are small scale ones. The average number of translators they employ is around ten, and the people who run the company often do not understand what translation is. This suggests that translators cannot acquire proper training or management in the process of exercising translation.

d) Lack of training: most students who major in Japanese language become translators after learning JA in universities for only 3.5 years. Translators who are skilled in JA and equipped with specialized knowledge are scarce.

Thus, because of the difference in social, industrial, educational, and institutional

environments, the translational competence of a professional translator in China who

can meet one of the three criteria specified by ISO 17100:2015 is very likely to be

markedly different from that of a professional translator in the European Union. With

18

such a difference as a given, LSPs and buyers can be more understanding of generally low JA to ZH-CN translation quality, and they should think about what measures can be taken to improve the current situation.

The same observation can be seen in TAUS (2017a, p. 10) which points out that

“many people in the human translation (HT) space say translators are not good enough to tackle today’s challenges, and more and more errors are detected than in the past,”

implying that ZH-CN may not be an exception.

All of the observations just reported strengthen the rationale for source optimization. As suggested by Lommel (2016), optimized source content can help translators produce better quality. In addition, it will lower the risk of poor outcomes and help buyers achieve their ultimate goal. Despite the growing conviction that optimized source will help both translators and buyers, precisely how to optimize source content remains an area under-researched in the academic domain and a topic with which both LSPs and buyers are struggling, especially when it comes to JA.

Kelly (2010, p. 1) asserts that “poorly constructed text is notoriously difficult to translate, but the definition of good writing may vary from one language to another.”

While general guidelines for specific languages (see Section 2.4) are available on the

market, Kelly (2010) suggests that having a non-native speaker review source language

texts can make the texts more translation-friendly as the non-native speaker can identify

what might be difficult to express in another language. His reasoning is that the more

input authors can obtain from individuals who are able to read the content from the

standpoint of a non-native speaker, the better suited the content will be for global

audiences. Moreover, writing text that is more “translatable” from the start will enable

more people in more languages to understand the message authors originally intended

to convey. On this basis, translators who are Chinese-speakers are considered to be

suitable candidates for identifying problematic features in JA STs.