This paper proposes new results in decentralized adaptive stabilization for a class of large-scale non-linear systems with output measurements. Like previous work, the parametric uncertainty does not satisfy any matching conditions and the uncertain interconnections may be highly non-linear. Unlike most previous results in the literature of decentralized control, the development of our systematic approach does not rely on the crucial assumption that the (unmeasured) zero-dynamics of each individual local system are linear and enter the system additively and linearly. Simulation results based on a practical example of two inverted pendulums on carts demonstrate the effectiveness of the proposed decentralized adaptive stabilization methodology.