Linguistic Preconfiguration of Computational Narration

Since the boom of generative artificial intelligence, an increasing amount of people use computation to automate and enhance their narration. Some see this development as a democratization of professional media creation. Technically this might be plausible, but epistemologically it is a reinforced reproduction of certain, often dominant, perspectives about the world. Language is inherent to such reproduction. The most common interface to these systems is a text box. In the broadest sense most training data are linguistic representations. Language is often considered as a thought-structuring interface between humans and their world. It represents and forms relationships with all living and non-living systems. Language produces, preserves and reproduces worldviews, frames perceptions and creates differentiations. Hence, language preconfigures everything that generative models produce. The narratives they generate are always a result of linguistic preconfiguration. In reverse narratives also preconfigure current vocabularies. This leads to a socio-technical reproduction loop conditioned through language, that reinforces and stabilizes the stuckness within existing and often problematic worldviews. A lot of research has focused on problematizing algorithmic bias. As important as such analyses are, I would argue that biases are a symptom of specific linguistic categorizations. Language produces biases and biases manifest through language. They are inaccuracies within existing worldviews. But when we acknowledge that worldviews themselves are and will always remain inaccurate projections onto reality, we might need approaches that focus on linguistic foundations in order to effectively engage with the reproduction loops of generative artificial intelligence. This raises the question if and how linguistic re-preconfiguration could enable new computational narratives? I think what these systems need are linguistic interventions that attempt to change the language that preconfigures them. New terms and new meanings that reflect contemporary discourses are needed to shift computational narratives into new directions. Furthermore, I believe training data have to be more thoughtfully curated and interfaces for prompt engineering need to have feedback mechanisms that raise critical questions about the usage of certain words in prompts. Generative artificial intelligence mostly stabilizes existing worldviews and thereby also problematic conditions. But it could also contribute to worldview shifts by mediating individual reflection and enable the distribution of alternative perspectives and narratives.

Apr 30, 2025