Orthographic skills were showed to require a long and widely implicit learning. For example, an early sensitivity to the orthographic system redundancy was demonstrated in children without having been explicitly taught (e.g., Cassar & Treiman, 1997). As in the field of oral language acquisition (verbal morphology), two opposite interpretations were proposed to account for these implicit learning mechanisms: the rule abstraction hypothesis proposed that a rule would be abstracted during the learning phase (Reber, 1967; 1993) whereas the fragmentarist hypothesis assumed that the learning would be based upon similarity between new and learned items (Brooks & Vokey, 1991).
The present experiments used some written language regularities in French that can be described in rules format and that are not explicitly taught at school. In the first experiment, children and teenagers (from grade 3 to 12) were submitted to a word and pseudo-word dictation task with items ending in ``*asïõ'' and ``*j'' (almost 100% regular in written French). The results showed that before five years of exposure to written language, performances on pseudo-words were equivalent to those obtained on familiar and regular words. These results are in accordance with the rule abstraction hypothesis. However, the alternative interpretation could not be completely excluded:the equilibration in the performances on both types of items could also be interpreted as reflecting the similarity degree between the previously learned items (words from the lexicon) and the new ones (pseudo-words).
In order to test the role of similarity, in a second experiment, adults participants were submitted to a computerised pseudo-word judgment task recording reaction times in which the similarity degree was manipulated on the items used in experiment 1 so as on items used by Pacton & al. (2001) testing sensitivity to the legal position of double consonants. Participants were presented with pairs of pseudo-words (ex. édication (legal)/édicacion (illegal) versus Dissimilar (ex. alication/alicacion) and were asked to choose the one they thought looked most like a word. The similarity degree was manipulated by using either pseudo-words with any phonological or orthographical neighbour (dissimilar) or pseudo-words with a frequent neighbour (similar). If similarity was a critical factor, it was expected that the subjects would be more accurate and fast on similar items than on dissimilar items, on the other hand if a rule has been abstracted no difference between similar and dissimilar items would be expected. The results are being currently analysed.