Researchers have identified an 'evaluability gap' in AI translation, where even when the source text is shown, it's hard to determine what aspects of the original content are preserved. A study published on arXiv examined how different rendering styles and source texts affect perceived quality and trust in AI-generated translations. The results suggest that displaying the source text doesn't guarantee a high-quality rating, especially for complex or nuanced content. This raises questions about the reliability of AI translation tools and the importance of transparency in their use. --- Why it matters: This study matters to researchers and engineers working on AI translation because it highlights the need for more accurate evaluation methods and greater transparency in AI decision-making processes. The findings also have implications for the development of more trustworthy AI systems that can handle complex and nuanced content. Source: https://arxiv.org/abs/2608.19083