How do stories change as they cross linguistic and cultural boundaries? Many theories of cultural evolution emphasize the influence of repeated transmission, but in real-world cross-cultural dissemination, transmission is typically confounded with large-scale changes in cultural context. This makes it difficult to identify mechanisms of cultural stability and change. Here, to unravel this theoretical tangle, we leverage an unusual translation cycle: a story first published in English was translated into Russian by novelist Leo Tolstoy, misattributed to Tolstoy, translated into French, and finally translated back into English --- with the first and final versions published in the same language, city (Chicago, USA), and time period (mid-1890s). This unusual 'translation cycle' offers a unique opportunity to separate the cumulative effects of iterated transmission from the influences of the broader cultural context. Using a novel computational method that represents stories as high-dimensional semantic trajectories, we show that each translation introduced cumulative distortions. Strikingly, when the story was translated back into its original cultural context, its semantic trajectory shifted toward the original version, despite the final translator having no knowledge of the original, consistent with a regularizing influence of the shared cultural context. Our findings reveal how the broader cultural context can act as a stabilizing force in cultural evolution, constraining variation introduced by repeated transmission.