The purpose of this communication will be to show how implicit morphological competence resists the progress of Alzheimer’s disease. We will illustrate our point with the results of two studies to study the semantic abilities of patients, through understanding and producing morphological brands. We will seek to address categorisation not on the basis of a taxonomic model (Rosch), but through the information conveyed by morphology. We will rely on studies showing that the properties of language signs in particular determine the structuring of the categories (Dubois and Poitou). We will show that patients with Alheimer’s disease use morphological information (lexical and derivative) to perform a categorical inclusion task. To this end, we have offered a categorical inclusion task to 30 MA patients and 30 controls. We created an experimental protocol consisting of items involving 3 variables: (1) whether or not they belong to the category ‘Arbre’ or ‘Metier’, (2) with or without derivative morphdema ‘-ier’ and (3) lexical morphem linked (or pseudo-linked) or unrelated to the fruit/function. The aim was to highlight the respective effects of the lexical and derivative morphdema. Analysis of the results shows that the combined information conveyed by lexical morphdema and the derivative morphem ‘-ier’ is effective, for both patients and controls. Patients use language information more than witnesses. There is therefore a robustness of morphology, in implicit memory, which allows subjects to respond to the task of categorical inclusion. Thus, patients’ lexico-semantic difficulties do not arise solely in terms of difficulty in accessing or losing semantic representations, as is assumed by cognitive psychology, since patients and witnesses can treat items as linguistic signs. The second study that we will present focuses on the production of neoforms, based on the corpus analysis of a patient carrying Alzheimer’s disease at the severe stage. We will try to show how these manifestations, the number of which increase with the advance of the disease, are to be seen not as a sign of irreversible disintegration of language, but as genuine compensation mechanisms, which have the morpholexical and morphosyntactical characteristics of the language code. Both studies tend to show that apparently meaningless productions can in fact be considered as idiolects. These idiolects remain dependent, even at the final stage of the development of the disease, on the rules of the language code, implicitly built by each subject. Language disorders, in the context of neurodegenerative diseases, cannot therefore be regarded as ‘loss of language’ or even ‘language dislearning’.