We use Bayesian inference methods to provide fresh insights into the sub-nanosecond dynamics of glycerol, a prototypical glass-forming liquid. To this end, quasielastic neutron scattering data as a function of temperature have been analyzed using a minimal set of underlying physical assumptions. On the basis of this analysis, we establish the unambiguous presence of three distinct dynamical processes in glycerol, namely, translational diffusion of the molecular centre of mass and two additional localized and temperature-independent modes. The neutron data also provide access to the characteristic length scales associated with these motions in a model-independent manner, from which we conclude that the faster (slower) localized motions probe longer (shorter) length scales. Careful Bayesian analysis of the entire scattering law favors a heterogeneous scenario for the microscopic dynamics of glycerol, where molecules undergo either the faster and longer or the slower and shorter localized motions.