Their findings have now been published in the prestigious journal *PLoS Computational Biology*.
The article focuses on cognitive science research at the intersection of language and thought. Specifically, one of the questions is: How is it that natural languages offer us multiple ways to refer to the same object using different terms—for example, Dalmatian, dog, animal? “Using computer-based modeling, we show that efficient languages develop from the interplay of the needs of conversation partners in communicative interaction,” explains Kobrock. While speakers want to use short, concise, and readily available terms, listeners prefer informative messages. “We investigate the constant balancing of these conflicting desires using agent-based modeling. Small AI systems learn to communicate with one another and develop their own language for this purpose, which we interpret and examine for features of natural languages,” adds the postdoctoral researcher.
The researchers’ findings can be summarized as follows: First, communication systems that are efficient and similar to natural languages emerge from the interplay of communicative needs during interaction. “Second, we identify the factors relevant to this process: Communication within a shared context—that is, a situation where conversation partners share knowledge—is crucial for the evolution of efficient languages,” says Kobrock. If speakers also consider how their messages will be interpreted, this further improves the efficiency of language systems—but only if the languages have already developed within a shared context.
Third, the modeling study contributes to a theory of language evolution that can explain the co-development of linguistic terms and mental categories arising from communicative needs in concrete interactions: The results support the idea that words and concepts emerge simultaneously. Both develop from the desire for balanced communication—simple enough for speakers, yet informative enough for listeners. “Agent-based modeling allows us to gain groundbreaking insights into communication because we have complete control over all factors. We could not achieve this to the same extent with human experiments. I look forward to further exciting research with this excellent research team,” adds Prof. Nicole Gotzner.
Kobrock K, Ohmer X, Bruni E, Gotzner N (2026) The role of pragmatic mechanisms in referential communication and categorization: An emergent communication model. PLoS Computational Biology 22(5): e1014326. https://doi.org/10.1371/journal.pcbi.1014326
Further information for the media:
Dr. Kristina Kobrock, University of Osnabrück
, Institute for Cognitive Science
kristina.kobrock@uos.de
