Résumé
Based on the assumption that innovation is a social construction which can only result from a simple scientific injunction or proposition, this article emphasizes the significance of organizational innovations in relation to sustainable development. Innovation requires a structuring framework for organizing actor participation, i.e. the interactions between multiple stakeholders including researchers. The methodological protocol described and employed for the co-construction of indicators concerns the implementation of sustainable development indicators in fish farming. After recalling the epistemological foundations of their approach to innovation, the authors describe the structure and interest of the methodological protocol that they have developed in the second section. This protocol is used for facilitating the appropriation of a new reference system by integrating actors' representations. Furthermore, it is organized according to an interrelation pattern: principles, criteria and indicators. This approach enables indicators to be constructed based on criteria, which themselves are defined by principles. These principles account for actors' representations and issues. The application of this approach involves an individual selection phase of principles and criteria and then the collective approval by stakeholders. Its application to pilot sites has enabled to draw up a reference check list including 17 principles, which are then broken down into criteria and indicators. It can then be recomposed and adapted infinitely according to the characteristic needs and expectations of each application. The implementation of this check list therefore represents, for users, an intermediate object facilitating co-construction. The third section proposes an illustration of the procedures employed for its implementation in diversified aquaculture systems. The authors detail the results for Mediterranean fishfarming by showing that the approach can be used for comparing viewpoints and for generating the emergence of compromises around common principles and criteria. Lastly, the authors list the simple and double loop learning processes which were used during the experimentation of the approach.