Generating new kinds of knowledge is a key concern of knowledge clusters in which cooperative relationships generally outweigh competitive ones. The knowledge in these kinds of clusters has the character of a flow, or to use a simple metaphor, a stream. The idea of a knowledge capital and how to measure it are very difficult to apply in practice. In France, the distinctive feature of knowledge clusters, called ‘pôles de compétitivité’, is that they are made up of very different entities (businesses, research groups, public or mixed institutions, and so on). The heterogeneous and extended nature of these kinds of organization raises the question of knowledge boundaries and of misunderstandings among members belonging to different entities. In this context, different actors who speak the same mother tongue, French for instance, do not share the same language. To understand one another it is not enough to speak the same tongue. This observation is quite common, however, the importance of its consequences has not been much taken into account. Indeed, in day-to-day work, speakers express themselves in their usual way, in conformity with the codes of the community to which they belong. The presence of several communities brings with it problems of misunderstanding. These arise sporadically, then retreat into the background, and so remain dormant. We demonstrate this by studying the case of a collaborative project named NP that we followed for 2 years, taking part in meetings of the consortium, and transcribing verbatim discussions arising at moments of obvious tension which we were able to observe through our involvement in the project. More particularly we analyze a moment of tension emerging during a consortium meeting between communities of physicists and chemists. Reading this moment in the light of Harry Collins’ theoretical framework we will show that the collective tacit knowledge of each community gives rise to different interpretations of the same subject. These are strongly linked to the usual methods and practices within each community or profession. Following Yrjö Engeström, we show that such moments of tension reveal the dynamics of an activity system and the obstacles to mutual understanding; they also provide a way for actors to transcend the language boundaries. Thus, contradictions and tensions by no means impair the project: on the contrary they should be considered as an opportunity to make progress, generate new knowledge and innovate. The Peer-Mediation Process, by transcending rather than simply overcoming those tensions, offers new perspectives: it becomes a constituent part of knowledge management.