Corps de l’article

Throughout the years, a growing body of research in Translation Studies has relied on Corpus Linguistics, either in connection with the latest developments in Machine Translation (MT) or with empirical corpus-based translation research across languages. Apart from that, a number of statistical tools and methods have been added in order to further improve research in the field.

Corpus methodologies explained: an empirical approach to translation studies can successfully be framed within this context as it approaches the field of Corpus Translation Studies (CTS) from distinct methodological perspectives. From corpus-driven to corpus-assisted to corpus-based methodologies, the book gradually develops from a review of machine translation paradigms to contrastive textual genre analysis, also addressing translator style and, ultimately, reflecting on translation universals regarding language, all of this with the support of thorough statistical data.

The publication is a result of the joint effort of four translation scholars from Australia, China, Norway and UK, led by Ji, whose affiliation with the Waseda Institute of Advanced Studies of the Waseda University in Tokyo paved the way for the project here undertaken.

The book is organized into five chapters: Chapter 1 presents a review of the main methods which have been used for machine translation, ranging from the rule-based model to the statistical model; Chapter 2 relies on the contrastive quantitative analysis comprising English, Chinese and translational Chinese in regard to textual complexity and genre shifting; Chapter 3 deals with translation stylistics across different versions of a source text in the language pair Chinese/English; Chapters 4 and 5 address the mutually exclusive “Over-representation of Target-language Specific Features Hypothesis” (Baker 1993, 1996) and the “Unique Items Hypothesis” (Tirkkonen-Condit 2001, 2004), challenging one against the other in the language pairs Norwegian-Spanish and English-Spanish, in an attempt to fully test the “Gravitational Pull Hypothesis” (Halverson 2003, 2007, 2009, 2010), with focus on a specific language item which is the Spanish gerund.

After a brief introduction, which carefully summarizes the main goal and findings of each chapter, Chapter 1, entitled The need for corpora in machine translation and written by Oakes, is outlined so as to offer an overview of the main methodological enterprises into the field of machine translation. Starting from the rule-based system, of which TAUM METEO is a successful example, the author advances towards more complex systems, such as translation memories, the example-based system and the statistical machine translation, of which Europarl is a remarkable example, recognizing the limitations of a system based solely on rules and arguing in favor of more comprehensive machine translation models which can rely on parallel corpora, be more easily updated, be satisfactorily built for less-resourced languages, and, at the same time, be language independent.

Chapter 2 is entitled A multidimensional analysis of the translational Chinese genre system and was written by Ji. It takes a corpus-driven perspective on the systematic differences found in translating certain genres from English into Chinese, which was termed as genre shifting, considering original English genres, target Chinese genres and translational Chinese genres, which are argued to differ from the previous two. In order to explain such differences, the author points out that “while the Chinese genre system is essentially reader-oriented, the English genre system is information-oriented” (Ji 2017: 72). By relying on three different corpora, namely the British National Corpus (BNC) for English, the Lancaster Corpus of Mandarin Chinese (LCMC) for original Chinese and the Zhejiang University Corpus of Translational Chinese (ZJU) for translational Chinese, the multidimensional corpus analysis is divided into two main parts: first, it deals with striking differences between original English genres and original Chinese genres; and second, it analyzes the key differences between original Chinese genres and translational Chinese genres. Several observations are made, especially regarding the translational genre system in the Chinese language, which can be summarized as follows:

[…] the genre system of Chinese translation is a highly dynamic system, as genre shifting not only occurs in the translation of fictitious and literary genres such as fiction and prose, but also occurs in important factual genres such as media and news reporting.

Ji 2017: 93

Chapter 3 focuses on the style adopted by the translators in producing their translations, addressed by several scholars as translator´s subjectivity and visibility (Venuti 1995), voice (Hermans 1996) or style itself (Baker 2000). Written by Defeng, Translator style: a corpus-assisted approach, as the title goes, investigates the translators´ styles in a comparative study of two well-known English translations of Hongloumeng (A <Dream of Red Mansions)– an acclaimed Chinese classical novel. A comparable parallel corpus is built and manually aligned, and data are analyzed in both quantitative and qualitative terms, divided into the following categories: type-token ratios, sentence length, and translators´ background in the two English versions. The author finally argues that

[…] the stylistic differences in the two versions were caused by the translators´ different philosophies of translation and choice of different translation strategies and methods, which in turn was affected by the social, political and ideological milieu in which they lived and worked.

Defeng 2017: 131

Chapter 4 is entitled The translation of formal source-language lacunas: an empirical study of the Over-representation of Target-language Specific Features and the Unique Items Hypotheses and was written by Hareide. It discusses the validity of two mutually exclusive hypotheses in corpus-based translation studies, namely the Over-representation of Target-language Specific Features Hypothesis, as defended by Baker (1993, 1995, 1996) and the Unique Items Hypothesis, later proposed by Tirkkonen-Condit (2004). The former relies on the assumption that “translations over-represent features of their host environment in order to make up for the fact that they were not meant to function in that environment” (Baker 1993 apud Hareide 2017: 137). The latter, on the other hand, is presented on the grounds that “target-language specific features are in fact under-represented in translations, because there are no structures in the source language that will trigger the use of these unique language structures.” (Hareide 2017: 138)

Having the language pair Norwegian/Spanish in mind and relying on both the Norwegian-Spanish Parallel Corpus (NSPC) and the Corpus de Referencia del Español Actual (CREA), Hareide (2017) tests the two hypotheses, at first favoring Tirkkonen-Condit´s in regard to the inexistence of a specific language item in Norwegian which could translate into the Spanish gerund, and thus classifying the Spanish structure as a unique item. However, with basis on the reference corpus (CREA), the Spanish gerund proved to be over-represented in the translations from Norwegian into Spanish, since the original Spanish corpus did not produce such a high number of gerunds. As the author points out, “Since the two hypotheses are mutually exclusive, the rejection of the Unique Items Hypothesis implies that evidence is presented in favour of the Over-representation of Target-language Specific Features Hypothesis.” (Hareide 2017: 179)

Chapter 5, whose title is Testing the Gravitational Pull Hypothesis, naturally develops from the findings in Chapter 4, as the Gravitational Pull Hypothesis (Halverson 2003, 2007, 2009, 2010)

[…] aims to predict and explain how the two outcomes of the translation process proposed by Baker´s Over-representation of Target-language Specific Features Hypothesis (Baker 1993, 1996) and Tirkkonen-Condit´s Unique Items Hypothesis (Tirkkonen-Condit 2001, 2004) can be expected in different situations.

Hareide 2017: 190

Hareide then notes that there are three factors that can lead to over- or under-representation, namely “patterns of prototypicality in the target language, conceptual structures or the representation of the source language item and patterns of connectivity” (Halverson 2010: 356 apud Hareide 2017). By using both the P-ACTRES English-Spanish Parallel Corpus and the CREA Reference Corpus, the author tests the Spanish gerund against a number of English corresponding structures. She then calculates the relative frequencies of the Spanish gerunds from the P-ACTRES, the CREA and the NSPC corpora, which are of different sizes, using the log-likelihood test. The author concludes her chapter by stating that

[…] All three factors of the Gravitational Pull Hypothesis, prototipicality and/or frequency, salience, or prototypicality in some part of the source language network and linkage between the related concepts in the bilingual´s mental lexicon, have pulled towards over-representation in this study.

Hareide 2017: 225

As a whole, the book reviewed here –Corpus methodologies explained: an empirical approach to translation studies – is carefully organized into a gradient of corpus methodology approaches, while at the same time it touches relevant and potential areas in translation studies – from sound machine translation methodologies to outstanding research into translation universals. Although the chapters are very well organized, and methodologies carefully explained, this book is not aimed at a novice reader in the subject. From the very beginning, the authors assume some familiarity with both the methods and theories covered in this volume. Nonetheless, both students and researchers will find insightful ways of exploring the ideas developed throughout the book.