It returns to the distinction between differences in pronunciation and mistakes in reading, and shows that it is not possible to decide the status of a given possible error in most cases. A probabilistic solution to the problem is obtaining by studying the semantic shadows cast by possible errors in raising or lowering the frequency of followiing errors. Results differ for clear errors, pand correct readings, while various types of possible errors tend to behave like one or the other. Striking differences among language/ethnic groups appear which may be linked to varying knowledge of underlying forms. From these results, it is possible to determine which phonological and grammatical patterns most require instruction. .pdf version