A study on neural cellular automata shows that linguistic distance slows consensus and induces mild group divergence without full fragmentation. A collective trained under diverse communication protocols remains robust to mismatch, unlike one trained uniformly, and these results are consistent across ring and 2D grid structures, with parallels to human group dynamics.