Attempts at predicting syntactic behavior from semantic or other generalizations are often unsatisfactory. It is argued that the notion of competition as used by Goldberg (2019) can serve as an explanation for unpredictability in language because established formulations that are preferred over others automatically distort the collocational profiles of verbs in argument structure constructions. As a consequence of this, an approach of seeing items as items-in-constructions (and not as elements attracted to them) is argued for. It is then shown how this items-in-constructions view can be applied to designing models of reference constructicons and mental constructicons.