During the nineteenth and twentieth centuries observational seismologists recorded primarily the earthquake-induced translational wave field, while the rotational motion still remains poorly observed and investigated. We aim to further understand the rotational ground motion and its relation to the translational wave field, with a special emphasis on the near field, a few wavelengths away from the hypocenter, where damage related to rotational motion might need to be considered. A broad picture of the available values of rotational amplitudes and their variability is obtained by gathering most of the published data on strong rotational motion. To obtain a more detailed picture we perform a large scale 3D numerical study of a strike-slip event in the Grenoble valley where a combination of topographic, source, and site effects produces a realistic wave field. We analyzed the synthetic dataset in terms of the rotational and translational peak amplitudes and their dependence on two effects: nonlinear soil behavior and source directivity. On a soft soil deposit, we observe peak ground rotation of 1 mrad and the peak ground rotation rate of 10 mrad/sec, for an Mw 6.0 event. Those values show a strong dependence on the hypocenter location, the local site conditions, and the topographical features, inducing a variability of almost one order of magnitude in a range of distances of 20 km. Finally, we compare our numerical results in terms of peak ground velocity (PGV) versus peak ground rotation (PGω) with field data obtained at similar scenarios (e.g., Parkfield) by array techniques to investigate the relation between translational and rotational amplitudes expected in the near field for shallow, medium-sized earthquakes. Results of our numerical simulation fit reasonably well with those observed in past studies. Furthermore, the spatial variations of the PGV/PGω ratio show a trend, which is correlated with the velocity structure of the model under study.