Oliver Čulo’s research while affiliated with Johannes Gutenberg University Mainz and other places

Post-editing is a rather new mode of translation production increasingly being studied from various angles. In this chapter, we contrast post-editing and human translation along the dimension of term translation within the domain of Languages for Specific Purposes. We make use of the perplexity coefficient to measure terminological variation in term translation from English into German. Our findings reveal levels of variation on the terminological level in the post-edited texts close, but not identical, to those of the machine translation outcomes. They thus indicate a shining through of the machine translations in the post-editing products, motivating further research into the properties of post-edited texts within corpus-based translation studies. On the basis of our observations, we discuss potential effects of this shining through, such as diminished quality of machine translation if post-edited texts are used for retraining, and we critically examine the applicability of the perplexity coefficient as a quality measure for term translation.

Translation can generally be seen as a task in which the meaning of the original should be preserved as far as possible. This paper formulates the preservation of meaning in terms of the primacy of the frame hypothesis: ideally, the frame of the original is matched by the frame of the translation. I investigate one factor overriding this principle in translations between English and German through the examination of two grammatical constructions, one in English, one in German, which are not commonly available in the other language. Picking a construction comparable in function in the target language leads to frame shifts. In addition to highlighting the interplay between construction and frame choice, the paper explores how frame-to-frame relations can be used to describe the semantic relatedness of original and translation in cases of frame divergences. Theoretical and methodological questions and implications of the cross-lingual application of frame relations are discussed at the end.

This volume deals with different aspects of the creation and use of multilingual corpora. The term 'multilingual corpus' is understood in a comprehensive sense, meaning any systematic collection of empirical language data enabling linguists to carry out analyses of multilingual individuals, multilingual societies or multilingual communication. The individual contributions are thus concerned with a variety of spoken and written corpora ranging from learner and attrition corpora, language contact corpora and interpreting corpora to comparable and parallel corpora. The overarching aim of the volume is first to take stock of the variety of existing multilingual corpora, documenting possible corpus designs and uses, second to discuss methodological and technological challenges in the creation and analysis of multilingual corpora, and third to provide examples of linguistic analyses that were carried out on the basis of multilingual corpora.

Translation shifts can be informative in various ways. Amongst other things, they can point to typological differences between languages or be indicators of properties of translated text like e.g. explicitation or normalisation. Detecting translation shifts in parallel corpora is thus a major task from the viewpoint of translation studies. This paper presents an analysis of translation shifts in a parallel corpus (English-German). It offers an operationalisation of queries which can exploit multi-layer annotation and alignment in order to detect various kinds of translation shifts across category boundary lines and empty alignment links. The paper furthermore discusses the shifts and links them to certain translation properties.

In this article, we address the task of comparing and combining different semantic verb classifications within one language. We present a methodology for the manual analysis of individual resources on the level of semantic features. The resulting representations can be aligned across resources, and allow a contrastive analysis of these resources. In a case study on the Manner of Motion domain across four German verb classifications, we find that some features are used in all resources, while others reflect individual emphases on specific meaning aspects. We also provide evidence that feature representations can ultimately provide the basis for linking verb classes themselves across resources, which allows us to combine their coverage and descriptive detail.

This paper presents results from empirical studies on language contrasts, translation shifts and translation strategies gained by exploiting the CroCo Corpus. The aim of this paper is to show that the insights from investigating the comparable parts of the corpus can be complemented by additionally exploiting the parallel parts of the corpus using the examples of word order peculiarities and diverging part-of-speech frequencies in English and German. The exploitation of the corpus proceeds in two steps. First, contrastive differences are identified in the comparable parts of the corpus. In the second step, the solutions chosen by human translators to deal with the contrastive differences are identified. These can be used to decide between different possible translation strategies and can serve as templates for translation strategies to be adopted in the development of MT systems.