🤖 AI Summary
This study addresses the challenge that current language models face in adapting to the symbolic systems and narrative conventions of diverse cultural audiences during open-ended text generation. It introduces cultural adaptability into the task of art description generation for the first time, proposing a pragmatics-informed speaker model that incorporates cultural sensitivity. The work also establishes an automatic evaluation framework centered on culturally grounded question answering. By integrating pragmatic modeling with cultural understanding, this paradigm significantly enhances cross-cultural communication effectiveness, yielding an 8.2% improvement in simulated listener comprehension and an 8.0% increase in human-rated scores for “helpfulness in understanding.”
📝 Abstract
Language models are known to exhibit various forms of cultural bias in decision-making tasks, yet much less is known about their degree of cultural familiarity in open-ended text generation tasks. In this paper, we introduce the task of culturally-adapted art description generation, where models describe artworks for audiences from different cultural groups who vary in their familiarity with the cultural symbols and narratives embedded in the artwork. To evaluate cultural competence in this pragmatic generation task, we propose a framework based on culturally grounded question answering. We find that base models are only marginally adequate for this task, but, through a pragmatic speaker model, we can improve simulated listener comprehension by up to 8.2%. A human study further confirms that the model with higher pragmatic competence is rated as more helpful for comprehension by 8.0%.