Fig. 4: Validation, PCA, and bias control in the temporal analysis of cohesion and deviation in imaginary worlds.

A Example of validity check: graphical representation of GPT-annotated Sizes and Exploration for various types of imaginary worlds, as established by Wolf (2013). Our analysis reveals a congruence between the size classifications annotated by GPT and the expected hierarchy of Wolf’s categories. Specifically, the GPT-annotated Size of imaginary worlds exhibits a trend that aligns with intuitive size distinctions among the various world types—for instance, an island being typically smaller than a country, which in turn is smaller than a continent, and so forth. Conversely, Exploration was equally high across all types. B Visualization of the principal component analysis (PCA) on the variables contributing to the construction of imaginary worlds. C A scatter plot mapped according to their scores on Cohesion (Dimension 1) and Deviation (Dimension 2). We label a selected sample of imaginary worlds. D The weighted evolution of Cohesion and Deviation. E The comparison of Cohesion and Deviation before and after 1800. F Bias control, estimating how many works from pre-1800 distribution of Cohesion should be added to the post-1800 distribution of Cohesion for the statistical difference to disappear (for p being above 0.05). The blue line is the average of all simulations.