Human languages are reliably learned and passed down through generations, yet the mechanisms
driving their evolution remain poorly understood. The emergence of a new sign language in Nicaragua in the mid-1970s provides a naturalistic case for tracing language development from inception. Drawing on an archival video corpus spanning three decades, we analyze signing by members of the original generation of language users through to the present day language community. Leveraging recent advances in computer vision, we generated 3D skeletal models from 2D video recordings of signed narratives. From these skeletal models, we extracted continuous metrics of sign size, velocity, and articulatory effort across hundreds of signing samples. Our analyses reveal two distinct adaptive pathways. With repeated use over time, signing became slower and less effortful, reflecting optimization for articulatory economy. In contrast, as the language was transmitted successively to new learners, signing shifted inward toward the face, increasing visual salience despite increased production cost, consistent with adaptation for visual perception. These findings offer quantified, empirical evidence in a natural language of distinct selective pressures during language use and learning: language use made the language easier to produce, while language learning made it easier to perceive.