A quantum mechanical relaxation theory is developed to enable approximate computation o electronic absorption line shapes of condensed phase systems where nonadiabatic coupling effects are important. At the simplest level, these computations require a time kernel (termed a memory kernel) which can be obtained from a sequence of wave packet propagations, each carried out on a single Born-Oppenheimer potential surface. Complications associated with the need to evolve wave packets on several nonadiabatically coupled surfaces are thereby avoided. Moreover, for many condensed phase problems the memory kernel can be computed via semiclassical techniques which rely on classical trajectories and simple Monte Carlo methods. The promise of the theory is demonstrated by numerical applications to the spectroscopic spin boson model [R. D. Coalson, J. Chem. Phys. 86, 995 (1987)], a nontrivial multimode model of electronic absorption lineshapes involving two nonadiabatically coupled excited state surfaces. The relevant quantum dynamics for the spectroscopic spin boson model can be computed exactly via path integration techniques. In this way, the accuracy of the proposed relaxation theory can be benchmarked, and the applicability of various semiclassical prescriptions for computing the memory kernel ascertained.