The problem of frame selection in parsing is discussed, with the focus on the selection of a frame for texts which contain highly ambiguous or vague words. An approach to frame selection is presented which involves the use of a small number of general inference rules, in conjunction with a hierarchically-organized conceptual memory. This is in contrast to various other methods, which rely either on disambiguation rules stored in the dictionary definitions of ambiguous words, or on previously-activated frames to guide the selection of new frames. The selection of frames for vague or ambiguous words using these previous methods is shown to be problematic. The method presented here does not suffer from these same problems because of the hierarchical organization of memory and the general inference rules used.