Action schemata and types of processes Our research focusses on the analysis of knowledge representation structures required in building natural language interfaces, for man/machine dialogue. In particular, for dialogues aimed at performing some "actions" on a computer system, e.g. expert system or data base mainly. Cooperative natural language interfaces must be able to reconstruct the implicit meaning of the user's message on the basis of its linguistic formulation. In this paper we focus on knowledge extraction from verbal expressions. Linguistic and extra-linguistic knowledge involved in the use of verbs have been analysed. Linguistic knowledge is given both by a classification of verbs complements, at the syntactic level, and by a definition of verbs categories, at the semantic level. Three verbs categories have been identified : resultative processes, non-resultative processes and states. Extra-linguistic knowledge underlying the use of the different verbs categories is described in terms of action schemata (AS). An (AS) is a representation structure containing several kinds of informations. Those informations are about : the roles played by the actors surrounding the verb, the aimed-result and the pre-conditions for the existence of the state or for triggering the process to which the verb may refer.