This thesis investigates how well machine learned classifiers can identify translated text, and the effect translationese may have in Statistical Machine Translation -- all in a Swedish-to-English, and reverse, context. Translationese is a term used to describe the dialect of a target language that is produced when a source text is translated. The systems trained for this thesis are SVM-based classifiers for identifying translationese, as well as translation and language models for Statistical Machine Translation. The classifiers successfully identified translationese in relation to non-translated text, and to some extent, also what source language the texts were translated from. In the SMT experiments, variation of the translation model was whataffected the results the most in the BLEU evaluation. Systems configured with non-translated source text and translationese target text performed better than their reversed counter parts. The language model experiments showed that those trained on known translationese and classified translationese performed better than known non-translated text, though classified translationese did not perform as well as the known translationese. Ultimately, the thesis shows that translationese can be identified by machine learned classifiers and may affect the results of SMT systems.

Mit Versprechungen von teils enormen Produktivitätssteigerungen bei mindestens gleicher Qualität im Vergleich zu Humanübersetzungen hat Post-Editing über die Jahre zunehmend an Relevanz gewonnen und ist heute aus der Übersetzungslandschaft nicht mehr wegzudenken. Neben zahllosen Qualitäts- und Produktivitätsstudien gab es vereinzelt Bemühungen, mögliche linguistische Unterschiede zwischen posteditierten und humanübersetzten Texten („Post-Editese“ nach Daems et al 2017) zu untersuchen. So stellte Toral (2019) fest, dass posteditierte Texte bestimmte Merkmale übersetzter Texte („Translationese“ nach Gellerstam, 1986), nicht nur aufweisen, sondern dass diese signifikant häufiger in posteditierten Texten vorzufinden seien. Insofern sei Post-Editese eine verschärfte Form der Translationese. Allerdings hat das Forschungsfeld der Post-Editese vergleichsweise wenig Aufmerksamkeit erfahren und in den meisten Korpusanalysen wurden Zeitungsartikel oder Texte aus dem Bereich general language untersucht, obwohl sich ein Großteil des Übersetzungsbedarfs im Bereich der Fachtexte und Language for Specific Purposes verorten lässt. Um dieses Desiderat zu bearbeiten, wird an die Vorarbeit von Toral angeknüpft und anhand eines Fachtextkorpus untersucht, ob sich seine These für die Translationese-Merkmale Simplification und Interference anhand von Ausschnitten aus technischen und medizinischen Fachtexten und den dazugehörigen Übersetzungen bestätigen lässt. Das Korpus besteht aus 75 Korpustexten des Sprachenpaars Englisch-Deutsch, darunter 6 Ausgangstexte und 69 Übersetzungen der Modi MÜ, LPE, FPE und HÜ. Die Methodik von Toral (2019) und Lapshinova-Koltunski (2013) wird in adaptierter Form und unter Zuhilfenahme der Sprachanalyse-Software Sketch Engine angewendet. Simplification soll anhand der lexikalischen Dichte und lexikalischen Diversität (Type-Token-Ratio, TTR) nachgewiesen werden. Das Verhältnis Nominalität/Verbalität und das Satzlängenverhältnis zwischen AT und ZT werden zum Nachweis von Interference herangezogen. Die Ergebnisse lassen weder zu Simplification noch zu Interference eindeutige Schlüsse zu: Während der TTR in posteditierten Texten wie erwartet geringer ist als in humanübersetzten Texten, d. h. die posteditierten Texte weniger lexikalisch divers sind als die humanübersetzten, ist die lexikalische Dichte in den posteditierten Texten entgegen der Hypothese höher als in den humanübersetzten Texten. Die Satzlängenverhältnisse zwischen AT und ZT sind in den posteditierten Texten tatsächlich ähnlicher als bei den Humanübersetzungen, was entsprechend der aufgestellten Hypothese auf mehr Interference des AT in den posteditierten Texten hindeutet. Die Ergebnisse zu den Verhältnissen von Nominalität/Verbalität jedoch konnten aufgrund von mutmaßlichen Verzerrungen durch Sprachsystemunterschiede im Bereich der Kompositabildung nicht sinnvoll ausgewertet werden. Durch die geringe Korpusgröße und die geringe Länge der Korpustexte wirken sich bereits einzelne idiosynkratische Entscheidungen auf Einzeltextebene, etwa bei der Satzsegmentierung oder bei der Terminologie, stark auf das Gesamtergebnis aus. Insofern zeigt sich, dass Besonderheiten auf Einzeltextebene in künftigen quantitativen Korpusanalysen, insbesondere in solchen mit größeren Korpora, stärker berücksichtigt werden sollten. Torals These, Post-Editese sei eine verschärfte Form der Translationese, konnte im Rahmen dieser Arbeit folglich nicht eindeutig bestätigt werden.:1 Einleitung 1.1 Herleitung 1.2 Ziel der Arbeit 1.3 Aufbau der Arbeit 2 Hintergrund 2.1 Post-Editing 2.1.1 Definition 2.1.2 Arten von Post-Editing 2.1.3 Produktivitäts- und Kostenvorteile 2.1.4 Qualitätsanalysen 2.1.5 Einflussfaktoren auf die Qualität des MÜ-Outputs 2.1.6 Präferenz- und Akzeptanzstudien 2.1.7 Einstellungen von Übersetzern, Sprachdienstleistern und Kunden 2.1.8 Weitere Forschungsrichtungen 2.2 Translationese 2.2.1 Definition 2.2.2 Übersetzungsuniversalien nach Baker 2.2.3 Interference nach Toury bzw. Teich 2.2.4 Forschungsstand 2.2.5 Neuere Forschungsrichtungen 2.3 Post-Editese 2.3.1 Definition 2.3.2 Forschungsstand 2.4 Zwischenfazit und Forschungsdesiderat 3 Daten und Methoden 3.1 Korpus 3.1.1 Beschreibung der Korpustexte 3.1.2 Erstellung des Korpus 3.1.3 Begründung der Auswahl des Korpus 3.1.4 Aufbereitung der Korpustexte für die maschinelle Auswertung 3.2 Analysemethoden 3.2.1 Simplification 3.2.2 Interference 4 Ergebnisse der eigenen Korpusanalyse und Einordnung der Ergebnisse 4.1 Simplification 4.1.1 Lexikalische Diversität 4.1.2 Lexikalische Dichte 4.1.3 Einordnung der Ergebnisse zu Simplification 4.2 Interference 4.2.1 Verhältnis der Verhältnisse von nominalen und verbalen Wortarten zwischen AT und ZT 4.2.2 Verhältnis der Satzlängen zwischen AT und ZT 4.2.3 Einordnung der Ergebnisse zu Interference 5 Diskussion 6 Fazit und Ausblick 7 Literaturverzeichnis / Promising moderate to sharp increases in productivity while achieving quality that is at least on par with human translation, post-editing has increasingly gained in importance in recent years and has become an integral part of the translation landscape. Aside from countless studies on productivity and quality, there have been isolated efforts to examine possible linguistic differences between postedited texts and human translations (“post-editese” according to Daems et al, 2017). Toral (2019) observed that postedited texts not only exhibit certain characteristics of translated texts (“translationese” according to Gellerstam, 1986), but that they exhibit them to a significantly higher degree compared with human translations. This prompted Toral to describe post-editese as an “exacerbated translationese”. However, research into post-editese has received relatively little attention and most corpus studies have focused on newspaper articles or so-called general language, despite the fact that most of professional translation takes place in the domain of Language for Specific Purposes. In order to address this desideratum, this work builds on Toral’s research. A corpus of technical texts is used to examine whether his thesis can be confirmed for simplification and interference, two well-known translationese characteristics, when relying on text excerpts from the technical and medical domain and their corresponding translations. The corpus consists of 75 texts in total, made up of 6 English source texts and 69 corresponding German translations produced via MT, LPE, FPE and HT. The analysis relies on an adapted form of Toral’s (2019) and Lapshinova-Koltunski’s (2013) methods and makes use of the text analysis software Sketch Engine. The parameters lexical density and lexical variety (type-token ratio, TTR) are used to identify simplification, whereas the nominality/verbality ratio and sentence length ratio between source and target text are interpreted as evidence for interference. The results are inconclusive for both simplification and interference. While the TTR is, as per the hypothesis, lower in the postedited texts, i. e. while the postedited texts are less lexically varied compared with the human translations, the lexical density of the postedited texts is higher than that of the human translations, which contradicts the hypothesis. The sentence length ratios between source and target texts are more similar in the postedited texts compared with human translations, confirming the hypothesis that postedited texts show more interference from their source text. The results for the nominality/verbality ratio, however, could not be interpreted in any meaningful way, as distortions due to differences in the language systems with respect to compounding likely play a role. Given the small corpus size and the short length of the corpus texts, even isolated, idiosyncratic decisions at the individual text level, e.g. in regards to sentence segmentation and terminology seem to have a major impact on the total result. This shows that peculiarities at the individual text level need to be considered more carefully in future quantitative corpus studies, in particular when dealing with larger corpora. Consequently, Toral’s thesis of post-editese being an exacerbated form of translationese could not definitively be confirmed within this work.:1 Einleitung 1.1 Herleitung 1.2 Ziel der Arbeit 1.3 Aufbau der Arbeit 2 Hintergrund 2.1 Post-Editing 2.1.1 Definition 2.1.2 Arten von Post-Editing 2.1.3 Produktivitäts- und Kostenvorteile 2.1.4 Qualitätsanalysen 2.1.5 Einflussfaktoren auf die Qualität des MÜ-Outputs 2.1.6 Präferenz- und Akzeptanzstudien 2.1.7 Einstellungen von Übersetzern, Sprachdienstleistern und Kunden 2.1.8 Weitere Forschungsrichtungen 2.2 Translationese 2.2.1 Definition 2.2.2 Übersetzungsuniversalien nach Baker 2.2.3 Interference nach Toury bzw. Teich 2.2.4 Forschungsstand 2.2.5 Neuere Forschungsrichtungen 2.3 Post-Editese 2.3.1 Definition 2.3.2 Forschungsstand 2.4 Zwischenfazit und Forschungsdesiderat 3 Daten und Methoden 3.1 Korpus 3.1.1 Beschreibung der Korpustexte 3.1.2 Erstellung des Korpus 3.1.3 Begründung der Auswahl des Korpus 3.1.4 Aufbereitung der Korpustexte für die maschinelle Auswertung 3.2 Analysemethoden 3.2.1 Simplification 3.2.2 Interference 4 Ergebnisse der eigenen Korpusanalyse und Einordnung der Ergebnisse 4.1 Simplification 4.1.1 Lexikalische Diversität 4.1.2 Lexikalische Dichte 4.1.3 Einordnung der Ergebnisse zu Simplification 4.2 Interference 4.2.1 Verhältnis der Verhältnisse von nominalen und verbalen Wortarten zwischen AT und ZT 4.2.2 Verhältnis der Satzlängen zwischen AT und ZT 4.2.3 Einordnung der Ergebnisse zu Interference 5 Diskussion 6 Fazit und Ausblick 7 Literaturverzeichnis

One of the main aims of this thesis is to examine the translational situation of popular fiction in post-industrial Japan. Specifically, the goal is to uncover two main aspects surrounding the phenomenon of translationese, the language used in translation. One aspect to be investigated is the characteristic features of Japanese translationese, and the other is readers’ attitudes toward translationese. This research is conducted within the framework of Descriptive Translation Studies (Toury, 1995). The literature review includes a background of how translationese has been approached previously and how methods from different fields (e.g., corpus linguistics, sociolinguistics) can be used in the research of translation. Through the review of the historical background of Japanese translationese and the development of Japanese writing styles, it is revealed that the translation norm in Japan had been very closely oriented toward the original text. In the text analysis, the corpora consist of translations from English and non-translations (i.e., originally written in Japanese) in the genre of popular fiction. The goal of the text analysis is to determine whether the features of translationese are actually characteristics of translationese. The features selected for this examination include the following: 1) overt personal pronouns; 2) more frequent loanwords; 3) female specific language; 4) abstract nouns as grammatical subjects of transitive verbs; and 5) longer paragraphs. Two features (third person pronouns and longer paragraphs) are shown to be characteristic of translationese, while others were proven otherwise or questionable (loan words, female language, abstract nouns as subjects of transitive verbs). Findings from the investigation of readers’ attitudes can help identify what constitutes the “norms” of translation (Toury, 1995, 1999) in Japanese society. Readers appear to be able to tell the difference between translation and non-translation. However, readers’ attitudes toward both translationese and non-translationese are more or less neutral or slightly positive. This may indicate that Japanese translationese has become integrated into the contemporary Japanese writing system and that readers do not regard translationese as overtly negative. This study shows that the major translation norm is becoming more domesticated translation in popular fiction, with the focus on making translations easier for the readers. / Translation Studies

No description available.

This paper aims to study translationese in the Japanese language, and to study if there are any differences between Swedish to Japanese translations compared with English to Japanese translations– which are studied more. Claimed features and characteristics of Japanese translationese include the increase use of personal pronouns, loanwords and paragraph length. However in this study the usage of formal language and gender language in translationese will also be included to in order to distinguish differences between translationese and non-translationese. The method used to analyze the material is corpus-based translations studies which is a rather newly developed method to study translation and especially translationese. This method will be used in this thesis to compare translated and non-translated texts. As material two novels are used – one originally written in Japanese, and one originally written in Swedish but translated into Japanese. Due to the lack of translation corpora in Japanese, the novels had to be digitalized in order for the author to browse in the novels. By comparing two novels of the same genre it will be clear what the translated novel lack compared to the non-translated novel as translationese is considered broken or bad translation. The analysis and the comparison will be done with the theories about the features of Japanese translationese. The study found that some of the features of translationese were adoptable on Swedish to Japanese translated texts, but there were few significant differences regarding formal language and gender language.

In the world of Descriptive Translation Studies, translationese refers to the specific traits that characterise the language used in translations. While translationese has been often investigated to illustrate that translational language is different from non-translational language, scholars have also proposed a set of hypotheses which may characterise such di erences. In the quest for the validation of these hypotheses, embracing corpus-based techniques had a well-known impact in the domain, leading to several advances in the past twenty years. Despite extensive research, however, there are no universally recognised characteristics of translational language, nor universally recognised patterns likely to occur within translational language. This thesis addresses these issues, with a less used approach in the eld of Descriptive Translation Studies, by investigating the nature of translational language from a machine learning perspective. While the main focus is on analysing translationese, this thesis investigates two related sub-hypotheses: simplication and explicitation. To this end, a multilingual learning framework is designed and implemented for the identification of translational language. The framework is modelled as a categorisation task, the learning techniques having the major goal to automatically learn to distinguish between translated and non-translated texts. The second and third major goals of this research are the retrieval of the recurring patterns that are revealed in the process of solving the task of categorisation, as well as the ranking of the most in uential characteristics used to accomplish the learning task. These aims are ful lled by implementing a system that adopts the machine learning methodology proposed in this research. The learning framework proves to be an adaptable multilingual framework for the investigation of the nature of translational language, its adaptability being illustrated in this thesis by applying it to the investigation of two languages: Spanish and Romanian. In this thesis, di erent research scenarios and learning models are experimented with in order to assess to what extent translated texts can be diff erentiated from non-translated texts in certain contexts. The findings show that machine learning algorithms, aggregating a large set of potentially discriminative characteristics for translational language, are able to diff erentiate translated texts from non-translated ones with high scores. The evaluation experiments report performance values such as accuracy, precision, recall, and F-measure on two datasets. The present research is situated at the con uence of three areas, more precisely: Descriptive Translation Studies, Machine Learning and Natural Language Processing, justifying the need to combine these elds for the investigation of translationese and translational hypotheses.

The central theme of the thesis is an in-depth translation criticism of the Austrian novel Die Liebhaberinnen by Elfriede Jelinek in the translation of Jitka Jílková. The criticism is based on the theoretical model of Katharina Reiss which is explored in her work Möglichkeiten und Grenzen der Übersetzungskritik. This model is expanded by the role of the editor and translationese style. At the end of the thesis Jitka Jílková is briefly presented, the translation analysis is summarised and evaluated with regard to an interview with the translator about her work.

Den här studien behandlar fenomenet tolkspråk – tanken om att tolkat språk skiljer sig från icke-tolkat språk. Översättningsvetenskapen och dess motsvarighet översättarspråk har utgjort en stor inspirationskälla till arbetet, då forskningen kommit längre där. Ett forskningsområde behandlar så kallade översättningsuniversalier – universella regler eller lagar för hur översatt språk ser ut. En av dessa lagar kallas för normalisering. I studien undersöks om normalisering är applicerbart även på teckenspråkstolkning, med utgångspunkt i påståendet att normalisering kan vara synligt i översatt text som en överrepresentation av typiska målspråksdrag. För att undersöka detta har två jämförbara korpusar använts, dels Svensk teckenspråkskorpus (SSLC) med icke-tolkade teckenspråkstexter, dels Korpus för simultantolkade teckenspråkstexter (KST) med tolkade teckenspråkstexter. Det typiska, teckenspråkiga drag som valts för undersökningen är det tecken som glosas KOPPLA. Förekomsten av tecknet i de båda korpusarna har undersökts för att kunna upptäcka en eventuell överrepresentation i KST. Resultaten visar att KOPPLA mycket riktigt är överrepresenterat i KST, men att det är svårt att generalisera på grund av flera begränsande faktorer. / This essay is concerned with the subject of interpretese – the idea that interpreted language differs from non-interpreted language. Within translation studies, the corresponding term is translationese, and this essay draws upon much of the research in this field, as it is more developed. One particular area of research into translationese revolves around so called translation universals, or universal features of translation. They could be described as rules or laws that define translated language. One of those universals is called normalisation. This essay seeks to answer if normalisation also exists in Sign Language interpreting, with reference to exaggeration of typical target language patterns. Two comparable corpora were used, Swedish Sign Language Corpus (SSLC) with non-interpreted Sign Language texts, and Korpus för simultantolkade teckenspråkstexter (KST) with interpreted Sign Language texts. The typical target language pattern that was chosen for the investigation is the Swedish sign KOPPLA. Instances of the sign were investigated in both corpora, to spot any exaggeration in KST. The results show that KOPPLA is in fact overrepresented in KST, but that one should be careful to generalize, as several limiting factors were at play.