Explanation Generation: Questions Direct Exploration in Semantic Space

In everyday reasoning and scientific inquiry, explanations help people make sense of their experiences and observations. We investigate how people generate new possible explanations. Specifically, we study how the phrasing of “why” questions shapes exploration of the space of possible explanations. We hypothesized that semantic exploration is guided by words that appear earlier, are more surprising, and are more familiar. Participants in our experiment generated explanations in response to “why” questions comparing two words, with the order of these two comparison words randomized. Using a large language model, we measured how closely each explanation aligned with the comparison words. Results supported two of three hypotheses: explanations aligned more strongly with earlier and more familiar words. These findings highlight how subtle phrasing of a question can shape explanation generation, demonstrating context-dependence in how people explore a conceptual space.