This is a somewhat delayed announcement that my paper exploring the relationship between translation task difficulty and translationese will be presented at WMT, co-located with EMNLP 2026 in Budapest.

This paper is my most significant contribution to the B7 project Translation as Rational Communication, part of SFB 1102.

I am happy that I managed to push this work through and obtain convincing results by combining information-theoretic measures of cognitive load with more traditional indicators of source-text and cross-lingual transfer difficulty. The key question was:

To what extent does translation task difficulty explain the amount of translationese in professional written and spoken translation?

I investigate this question using parliamentary data for English ↔ German.

The answer? At best, R² = 21%. So a lot of variance remains unexplained. But I have seen papers on similar topics where a model explaining 5% of the variance was described as a strong model, so 21% is certainly not negligible.

More interestingly, digging into the individual difficulty indicators revealed a sharp divide between translation and interpreting in how incoming difficulty is handled.

Written translation tends to reproduce difficulty: more demanding source texts result in more translationese, broadly in line with what we would expect.

In interpreting, however, structural complexity seems to be dispensed with altogether, producing a paradoxical inverse relationship: the more structurally demanding the source text, the less translationese we observe in the interpreted output.

This contrast, I think, is one of the most interesting findings of the paper.