Frost, R.
Department of Psychology,The Hebrew University
Computational models of visual word recognition are major tools in developing theories that have descriptive and explanatory adequacy with regard to the fundamental phenomena of reading. Hence, two main constraints should be considered for assessing their contribution: first, they should be universal, in the sense that they should reflect the common cognitive operations involved in treating printed language across different writing systems. Second, they must have a linguistic validity, in the sense that they should consider the full linguistic environment in which the cognitive operations take place. By drawing on evidence from morphological processing in Hebrew I will argue that most if not all recent reading models that focus on flexible letter-position coding, are severely impoverished and lack linguistic validity. These models focus on computing an orthographic code by mapping an input structure of letters to an output structure of word units, while disregarding the contribution of phonological, semantic, and especially morphological factors to the process. More importantly, they account for findings in reading that are not fundamental, reflect idiosyncratic properties of relatively morphologically-impoverished languages, and are therefore non-universal by definition. Languages differ with respect to the internal structure of words, which is determined by the systematic statistical co-occurrences of orthographic and phonological sublinguistic units (not necessarily contiguous), and by their specific patterns of correlation with semantic meaning. Native speakers and L2 learners detect these correlations and this shapes their processing routines. Therefore, only models that can “pick-up” and learn the statistical properties of the language can be both linguistic and universal.