The CoFI Reader: A Continuous Flow of Information approach to modeling reading

I present a novel cognitive model of reading based on a continuous flow of information approach, the CoFI reader, where partial information from different levels of representation is continuously being made available to next levels. In an example application, I implement the model in a hierarchical Bayesian
framework and fit it to self-paced reading times data: a reading
task where one word is presented at a time and the presentation time is controlled by the experimental subject. The results show that the model provides a reasonable fit to word-level reading times, and can account for two previously observed findings: (i) reading times are much shorter than the minimum time required for all cognitive processes that should take place, and (ii) the processing difficulty of a word affects the reading times of subsequent words (i.e., spillover or lag effects). Computational models have explained these findings through
parafoveal preview, that is, the partial processing of upcoming words during reading before they are directly fixated by the eyes. The CoFI reader model provides an explanation for these findings that is relevant for natural reading, but also, crucially, for self-paced reading, where parafoveal preview is not
possible.