The combination of language processing, social interaction and the emergence of distributed human systems is ubiquitous, but has been largely unexplored. Cognitive architectures allow us to investigate the problem space, make predictions and explain experimental results. My models of adaptation in dialogue have been extended to the case of teams, which often have to develop a common language to carry out their tasks. A model of small-scale domain language evolution combines accounts of learning, empirical distributions of nouns and small world networks. Cue-based memory retrieval and learning describe how we can create novel expressions for concepts (Reitter 2010). A noisy initial knowledge base was sampled from noun occurrences in archives of the Wall Street Journal. A network graph connected several instances of the cognitive models. Over time, the team of models establishes agreement on a communication system and increases its task success. The simulations explain earlier empirical results by Fay et al. There metaphorical and iconic drawings are established in the community to express given meanings in a series of Pictionary games. Thus, empirical data come not from natural language, but from different modalities. % Future work will simulate cultural language evolution at a large scale with my model. Network science provides useful mathematical tools and simulations to describe structural properties of such communities. The structure of the communities in the domain language evolution experiment turned out to be important. The model predicts that the structure of the communication network determines whether the evolved languages are robust against structural change in the community. The very common small-world networks lead to more reliable convergence toward a common language than do typical organizational hierarchies (trees) (Reitter 2010b). This prediction is now being tested empirically at large scale. In ongoing work, I have therefore developed the Geo Game, an experimental paradigm to study the behavior of larger human groups under controlled conditions. A first study with groups of about 20 participants suggested the usefulness of targeting one's communications in human networks, compared to making information broadly available. The near future will see the use of the experimental platform with remote participant groups large enough to demonstrate the interaction of network structure and individual cognitive processes. The Geo Game platform gives participants a naturalistic communication channel (natural language chat) and a richer user interface than previous experimental designs from cognitive economics or language evolution studies. % Besides the obvious increase in ecological validity, it gives us new avenues to study multi-modal information transmission in human networks.