In this paper, we tackle the issue of multiplicity of meaning in derivation using Frame Semantics and eXtensible MetaGrammar (XMG). We use corpus extracted data to identify the range of readings -al derivatives exhibit and identify prominent constraints on the types of situations and entities -al targets. These constraints have the form of type constraints and specify which arguments in the frame of the verbal base are compatible with the referential arguments of the derivative. The introduction of these constraints into the semantics of an affix allows one to predict and generate those readings which are possible for a given derivative and, at the same time, rule out those readings which are not possible. Finally, as a proof of concept, we model these constraints using XMG, and check whether the output resulting of this XMG description is consistent with the range of readings observed in the corpus.