Before newer tools automated the process by tracking motion of a target, folks wouldn't have to adjust them frame by frame per se but would instead annotate the subtitle script with keyframe times and positions. The video player would know to interpolate between those annotations.