Spatial coding and discourse models during text reading

Abstract Text comprehension involves frequent backtracking through previously read material, either accompanied by regressive saccades or not, in order to find the elements necessary for the interpretation of words currently being inspected. An obvious example is anaphor processing, where reference to an earlier element in the text, and thus the necessity to backtrack, is marked linguistically (by a pronoun, a definite article, a preposition, etc.). Little is known, however, concerning the exact nature of the text representation(s) used by readers when performing such backtracking. In the present paper, two potential levels of text representation will be discussed: a representation of the text content (see, for example, the notion of a discourse model proposed by Garnham, 1981) and a representation of the surface form of the text, including the spatial locations of words on the screen.