Agents producing and exchanging knowledge are forming as a whole a socio-semantic complex system. Studying such knowledge communities offers theoretical challenges, with the perspective of naturalizing further social sciences, as well as practical challenges, with potential applications enabling agents to know the dynamics of the system they are participating in. The present thesis lies within the framework of this research program. Alongside and more broadly, we address the question of reconstruction in social science. Reconstruction is a reverse problem consisting of two issues: (i) deduce a given high-level observation for a considered system from low-level phenomena; and (ii) reconstruct the evolution of high-level observations from the dynamics of lower-level objects.

In this respect, we argue that several significant aspects of the structure of a knowledge community are primarily produced by the co-evolution between agents and concepts, i.e. the evolution of an epistemic network. In particular, we address the first reconstruction issue by using Galois lattices to rebuild taxonomies of knowledge communities from low-level observation of relationships between agents and concepts; achieving ultimately an historical description (inter alia field progress, decline, specialization, interaction – merging or splitting). We then micro-found various stylized facts regarding this particular structure, by exhibiting processes at the level of agents accounting for the emergence of epistemic community structure. After assessing the empirical interaction and growth processes, and assuming that agents and concepts are co-evolving, we successfully propose a morphogenesis model rebuilding relevant high-level stylized facts. We finally defend a general epistemological point related to the methodology of complex system reconstruction, eventually supporting our choice of a co-evolutionary framework.

Agents producing and exchanging knowledge are forming as a whole a socio-semantic complex system. Studying such knowledge communities offers theoretical challenges, with the perspective of naturalizing further social science, as well as practical challenges, with potential applications enabling agents to know the dynamics of the system they are participating in. The present thesis lies within the framework of this research program. Alongside and more broadly, we address the question of reconstruction in social science. Reconstruction is a reverse problem consisting of two issues: (i) deduce a given high-level observation for a considered system from low-level phenomena; and (ii) reconstruct the evolution of high-level observations from the dynamics of lower-level objects. In this respect, we argue that several significant aspects of the structure of a knowledge community are primarily produced by the co-evolution between agents and concepts. In particular, we address the first reconstruction issue by using Galois lattices to rebuild taxonomies of knowledge communities from low-level observation of relationships between agents and concepts; achieving ultimately an historical description (including inter alia field progress, decline, specialization, interaction -- merging or splitting). Using the framework of a socio-semantic network, or "epistemic network," we then micro-found several stylized facts regarding the empirically observed structure: we exhibit processes at the level of agents accounting for the emergence of epistemic community structure. After assessing empirical interaction and growth processes, and assuming that agents and concepts are co-evolving, we successfully propose a morphogenesis model rebuilding relevant high-level stylized facts. We finally defend a general epistemological point related to the methodology of complex system reconstruction, eventually supporting our choice of a co-evolutionary framework.

Thanks to new conceptual and computational tools, the analysis of kinship and marriage networks has advanced considerably over the past twenty-five years. While in the past, the discussion of empirical marriage practices was often restricted to a casual observation of salient network features, it is now easy to produce a complete census of matrimonial circuits, both between individuals and between groups. However, the abundance of structural features which have thus become accessible raises a new question: to what extent can they be taken as indicators of sociological phenomena (such as marriage preferences or avoidances), rather than as effects of chance or of observer bias?

This paper presents a series of recently developed simulation techniques that deal with this issue. Starting from a new approach to “classical” agent-based modeling of kinship and alliance (group) networks (Section 2), we then present an automatic model discovery technique which, instead of constructing alliance networks from given matrimonial rules, reconstructs plausible matrimonial rules underlying given alliance networks (Section 3). While these techniques apply to “objective” representations of kinship and alliance networks, we also present two methods that take into account the generally lacunar and biased character of empirical kinship datasets. The first method we propose to deal with this problem (Section 4) is a generalized version of White’s (1999) “reshuffling” approach, which consists in redistributing marriage or descent links between individuals or groups while keeping the numbers of links constant. (For alliance networks, the question can be dealt with analytically by straightforward calculation of expected marriage circuit frequencies.) The second method (Section 5) consists in simulating the processes of network exploration by a virtual fieldworker navigating through kinship or alliance networks according to given behavioral constraints.