A multidimensional approach to aligned sentences in translated text

Abstract

Using unsupervised clustering techniques this study explores sentence alignment patterns in a parallel corpus of Norwegian source texts and Spanish translations, the NSPC (Hareide and Hofland 2012). The results show that three strategies with respect to sentence alignment dominate: one to one correspondence, merging two sentences into one, and removing sentences altogether (omission). The strategies are intricately correlated with the variables translator, author, and genre. However, we show how visualization techniques for cluster analyses offer a possibility for teasing apart these interactions as well as their relative importance. Our results indicate that non-fiction texts allow translators more freedom with respect to the treatment of sentences than do texts that are written by professional authors of fiction. The style of the author appears to play only a secondary role, but is especially important in fiction.

Keywords: corpus based translation, cluster analysis, parallel corpora, corpus alignment, unidirectional bilingual corpus