Principles in the computational modelling of eye-movements in reading

We take computational models of eye-movements in reading and consider the philosophical status of their components, specifically the visual input—the models’ eyes. We distinguish between an ‘abstract universal’, defined as that which is common across a range of instances, and a ‘concrete universal’, defined as an abstract, but still objective aspect of the process of reading. We argue that models composed exclusively of abstract universals cannot be the basis for a sustainable, scientific investigation of reading. In contrast, the concrete universal in a model mediates the whole of the domain, interacting in necessary ways with the other components of the domain. Unlike the abstract universal, the concrete universal cannot be defeated by new data. The modelling enterprise becomes one in which the goal is completeness rather than simplicity, and explanation is found in the double-movement of analysis down to the abstraction of the concrete universal and synthesis up to the concrete of the full complexity of the reading system. We argue for the space-text-and-reader (STAR) principle: the precise spatial relationship (x, y, and z coordinates) between the plane of the text and the reader is the ‘concrete universal’ in reading.