Researchers investigated whether multilingual large language models (MLLMs) generate translationese - a distinct style of writing that is characteristic of translated text. They compared MLLM-generated text to non-translated and human-written baselines in five languages, using established indicators of translated text. The study found that MLLM-generated text does contain translationese features, but with some differences from direct translation. The researchers used high-accuracy classification models and human annotations to analyze the results. --- Why it matters: This research matters because it sheds light on the limitations of multilingual large language models in generating high-quality text. Understanding how these models produce translationese can help developers improve their performance and reduce biases in generated text. Source: https://arxiv.org/abs/2608.17399