Online supplement for

Evert, Stefan & Neumann, Stella (2017). The impact of translation direction on characteristics of translated texts: A multivariate analysis for English and German.

 

Evert, Stefan and Neumann, Stella (2017). The impact of translation direction on characteristics of translated texts. A multivariate analysis for English and German. In G. De Sutter, M.-A. Lefer, and I Delaere (eds.), Empirical Translation Studies. New Theoretical and Methodological Traditions, number 300 in TiLSM. add page number Mouton de Gruyter, Berlin.

manuscript (PDF)

This paper investigates the influence of the source and target language on translations in a selection of 150 translation pairs from a bidirectional parallel corpus of English and German texts, applying a combination of multivariate analysis, visualisation and minimally supervised machine learning. Based on a procedure developed by Diwersy, Evert and Neumann (2014), it investigates the way in which translations differ from comparable original texts depending on the translation direction and other factors. The multivariate approach enables us to detect patterns of feature combinations that cannot be observed in conventional frequency-based analyses, providing new evidence for the validity of interference or shining through in translation. We report a clear shining through effect that is more pronounced for translations from English into German than for the opposite translation direction, pointing towards a prestige effect in this language pair.
   

Figure 1
Scatterplot matrix showing the first four PCA dimensions

download MP4 (7.0 MiB)
open interactive viewer (on supported browsers)

Figure 1: Scatterplot matrix showing the first four PCA dimensions. Circles are German texts, crosses are English texts. Translation status is indicated by colour.

The animation rotates through the top-left, top-centre and centre panels of the scatterplot matrix. German texts are shown as balls, English texts as tetrahedra (pyramids).

download MP4 (7.1 MiB)
open interactive viewer (on supported browsers)

Scatterplot matrix and animation corresponding to Figure 1, but colour-coded for register (not shown in paper).
 

Figure 2
Distribution of texts along the second PCA dimension

 

Figure 3
LDA discriminant for German vs. English originals

download MP4 (15 MiB)
open interactive viewer (on supported browsers)
interactive viewer with arrows linking originals to translations (7.8 MiB)

Figure 3: LDA discriminant for German vs. English originals (vertical axis of top row) with additional PCA dimensions from the orthogonal complement space.

Black points are originals, red points are translations.

download MP4 (6.9 MiB)
open interactive viewer (on supported browsers)

Scatterplot matrix and animation corresponding to Figure 3, but colour-coded for register (not shown in paper). Small points are translations, large points originals.
 

Figure 4
Distribution of texts along the LDA discriminant for German vs. English originals

   

Figure 7
Feature weights contributing to the LDA discriminant (normalized for orthogonal projection)

 

Figure 8
Boxplots showing the contribution of each feature to the position of German and English originals on the LDA discriminant

 

Figure 9
Distribution of texts along a simplified discriminant that represents characteristic patterns in the realization of themes