A Dataset for Probing Translationese Preferences in English-to-Swedish Translation

2026-03-09

AI Summary

This study addresses the tendency of machine translation models to produce English-influenced “translationese” rather than idiomatic Swedish in English-to-Swedish translation. To investigate this issue, the authors construct the first publicly available English–Swedish parallel dataset comprising translationese sentences alongside human-rewritten, fluent alternatives, annotated with error types. Through controlled experiments and evaluations using both multilingual and Swedish-specific small language models, the study finds that models consistently favor translationese outputs. Even when the source English sentence is removed—thereby eliminating direct priming—the preference for human-rated fluent translations increases only modestly, and model generations still frequently deviate from natural Swedish usage. This dataset establishes a new benchmark and analytical framework for probing stylistic preferences in translation and improving the fluency of non-English text generation.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Learning Human Values and Preferences

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Translations often carry traces of the source language, a phenomenon known as translationese. We introduce the first freely available English-to-Swedish dataset contrasting translationese sentences with idiomatic alternatives, designed to probe intrinsic preferences of language models. It includes error tags and descriptions of the problems in the original translations. In experiments evaluating smaller Swedish and multilingual LLMs with our dataset, we find that they often favor the translationese phrasing. Human alternatives are chosen more often when the English source sentence is omitted, indicating that exposure to the source biases models toward literal translations, although even without context models often prefer the translationese variant. Our dataset and findings provide a resource and benchmark for developing models that produce more natural, idiomatic output in non-English languages.

Problem

• translationese
• idiomaticity
• English-to-Swedish translation
• language models
• naturalness

Innovation

• translationese
• idiomaticity
• English-to-Swedish dataset
• language model bias
• naturalness in translation