This thesis proposes a practical model making it possible to implement a case-based reasoning system that adapts processes represented as natural language text. The use of natural language simplifies both the modelling and the execution by avoiding the need for the users to use special formalisms such as workflows to represent processes. In answer to a query describing a goal, the system shall be able to present the user with a consistent set of instructions enabling them to achieve that goal, expressed using natural language.In order to make inferences possible, a formal representation of a process ought to be attached to the text describing it. We use classical methods from natural language processing, a custom anaphora resolution mechanism and a set of annotation rules to extract events and objects from instruction texts, as well as temporal constraints represented using a qualitative interval algebra.During the adaptation stage, substitutions are performed in the source solution in such a way that it becomes a solution to the target problem. Temporal constraints are modified using a belief revision operator in order to maintain consistency with the application domain knowledge. We define two belief revision operators applicable on qualitative algebras: the first, using a best-first search algorithm, is consistent with the Alchourrón, Gärdenfors and Makinson (1985) postulates. The second is a repair propagation algorithm based on Vilain and Kautz (1986). It is faster, but may not obey all the postulates. It is shown that the reasoning process applied to processes can also be applied to different problems, such as farming problems, that can be represented using a qualitative algebra.Finally, the annotation rules are applied inversely with respect to temporal constraint changes, in a text regeneration stage. This has the effect of making minimal modifications to the text that make it consistent with the new temporal constraints. Strategies are used to maintain global consistency and anaphoric cohesion.The proposed model was applied to cooking problems, and implemented as a Facebook application, named Craqpot. Comparative tests were run, in which our solution was compared to a retrieval-only solution and a solution performing a more superficial adaptation. Our in-depth adaptation model produced texts of the same quality as the more superficial solution, but the recipes themselves were judged slightly better. The quality of the adapted recipes and texts were expectedly not as good as that of unmodified recipes and texts from the case base. Overall though, the users were as much satisfied with the deeply adapted recipes as with the original ones, and were much less satisfied with the superficially adapted recipes.

EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Main/Exploration

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000912 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Main/Exploration/biblio.hfd -nk 000912 | SxmlIndent | more

{{Explor lien
   |wiki=    Wicri/Lorraine
   |area=    InforLorV4
   |flux=    Main
   |étape=   Exploration
   |type=    RBID
   |clé=     Hal:tel-01098087
   |texte=   Reasoning with Qualitative Spatial and Temporal Textual Cases
}}