Predicting Syntactic Equivalence in Translation
Bram Vanroy, Arda Tezcan and Lieve Macken

Many of the difficulties that translators are faced with can be ascribed to equivalence or the lack thereof (Sun, 2015). In translation studies, equivalence is a concept that indicates how a source text and its translation can be compared to each other. It works on low-level language features such as morphology, lexicon, and syntax, as well as on general text properties such as semantic, pragmatic, and cultural planes (Baker, 2011). In the current study, we focus on syntactic equivalence. When this equivalence is high, no or a few re-ordering steps are needed to transform the source text’s syntactic structure into the target structure. When it is low, many re-ordering steps are required.

In light of the PreDicT project (Predicting Difficulty in Translation) [1], this study aims to estimate a sentence’s syntactic equivalence without the need of a translation. To that end, we use the wealth of data available in aligned source and target sentences of parallel corpora. By making use of a source sentence that is maximally similar to the input, we can use that sentence pair’s syntactic equivalence as an approximation of the equivalence of the input sentence. We use similarity metrics to see in how far they can be used to select the sentence that is syntactically most equivalent to the input sentence.

[1]: http://research.flw.ugent.be/en/projects/predict


References

Baker, M. (2011). In Other Words: A Coursebook on Translation (2nd ed.). Abingdon, UK: Routledge.
Sun, S. (2015). Measuring Translation Difficulty: Theoretical and Methodological Considerations. Across Languages and Cultures, 16(1), 29–54.