We derive empirical constraints on the nucleosynthetic yields of nitrogen by incorporating N enrichment into our previously developed and empirically tuned multi-zone galactic chemical evolution model. We adopt a metallicity-independent ("primary") N yield from massive stars and a metallicity-dependent ("secondary") N yield from AGB stars. In our model, galactic radial zones do not evolve along the observed [N/O]-[O/H] relation, but first increase in [O/H] at roughly constant [N/O], then move upward in [N/O] via secondary N production. By $t\approx5$ Gyr, the model approaches an equilibrium [N/O]-[O/H] relation, which traces the radial oxygen gradient. We find good agreement with the [N/O]-[O/H] trend observed in extra-galactic systems if we adopt an IMF-averaged massive star yield $y_\text{N}^\text{CC}=3.6\times10^{-4}$, consistent with predictions for rapidly rotating progenitors, and a fractional AGB yield that is linear in mass and metallicity $y_\text{N}^\text{AGB}=(9\times10^{-4})(M_*/M_\odot)(Z_*/Z_\odot)$. This model reproduces the [N/O]-[O/H] relation found for Milky Way stars in the APOGEE survey, and it reproduces (though imperfectly) the trends of stellar [N/O] with age and [O/Fe]. The metallicity-dependent yield plays the dominant role in shaping the gas-phase [N/O]-[O/H] relation, but the AGB time-delay is required to match the APOGEE stellar age and [O/Fe] trends. If we add $\sim$40\% oscillations to the star formation rate, the model reproduces the scatter in gas-phase [N/O] vs. [O/H] observed in external galaxies by MaNGA. We also construct models using published AGB yields and examine their empirical successes and shortcomings. For all AGB yields we consider, simple stellar populations release half their N after only $\sim$250 Myr.
Comment: 21 pages, 10 figures. See Figs. 5, 6, and 9 for key results. Submitted to MNRAS