Jotaro Tasaki, Ryosuke Yamanishi: A semantic analysis of video game reviews on a multilingual feature space, Proc. 25th International Conference on Entertainment Computing, 2026年10月 (to appear)

As video games reach a worldwide audience, linguistic translation alone may not be sufficient to make titles truly accessible. To truly localize a title, it should be essential to capture the diverse cultural interpretations and sensibilities of players from different regions. This study investigates cross-cultural differences in player feedback by analyzing Steam reviews of the game Apex Legends from the United States, Japan, and France. To evaluate semantic relationships across different languages, we implemented a modified BERTopic pipeline that maps multi-lingual reviews into a shared 5-dimensional semantic space. We observed cases where similarly labeled topics were isolated, indicating that shared labels do not necessarily imply semantic similarity. Also, upvotes and downvotes clustered by shared game mechanics, suggesting that different cultures may attach different sentiment values to the same gameplay experiences. Such alignments revealed potential cultural-linguistic paradoxes, notably instances in which Western frustration vocabulary was used as praise in Japan. These structural and contextual divergences support the claim that relying only on literal translation is insufficient, and investigating cultural differences should be important.