Mental scanning of visual images generated from verbal descriptions: towards a model of image accuracy

Recent mental scanning experiments have shown that subjects are not only able to construct mental images from verbal descriptions, but that these images have structural properties similar to those of images derived from perception. In addition, the specific sequencing of a description can affect the internal structure of images of described objects, in particular their metric properties. Discontinuous descriptions require additional exposure to achieve the structural coherence of images constructed from continuous descriptions. Thus, the capacity of images to reflect accurately the objects they refer to is not an all-or-nothing property, but rather results from stepwise elaboration. This study describes a quantitative model designed to account for the gradual process of image elaboration and the progressive increase in image accuracy. The model posits that the location of a landmark mentioned in a description is not represented as a sharp point in the mental image, but is instead associated with a region around this point. Learning the description essentially consists of progressively narrowing each 'region of uncertainty' associated with a landmark to its exact location. Additional experimental data were collected to provide a more finegrained understanding of image elaboration. Computations of the regions of uncertainty associated with each landmark were used to develop a computer program simulating the whole mental scanning protocol, which provided support for our account of image accuracy.