In his contribution to this issue, Martin Hilpert documents a shift over time in the collocational preferences of the modal auxiliary may. Hilpert’s paper is interesting beyond the empirical tools he brings to bear on this topic, as he also pleads for another shift, a theoretical one. In his treatment of may’s changing collocates, Hilpert (this issue, p. 82) proposes that we move away from conceiving constructions as “schemas with slots” towards seeing them as “networks of connections”: “a constructional view needs to pay close attention to the mutual associations between modal auxiliaries and the lexical elements with which they occur” (this issue, p. 67). As Hilpert argues, the topic of modal verb constructions is a challenging one for Construction Grammar, since it is not immediately obvious that such constructions exist at all. Modal verbs do not seem to enter into configurations that are semantically non-compositional, exhibit unpredictable formal properties or are subject to unexpected constraints. The only clear motivation for a constructional view is the observation that a modal verb combines with some infinitives much more frequently than could be expected (and with other infinitives much less frequently than expected). We agree with Hilpert that a structure such as [can + infinitive] is not semantically or formally abnormal and has no special restrictions. This is also the case for an inverted structure such as [should + subject NP + infinitive], which falls naturally from the general grammatical pattern common to all core modals. Nevertheless, if we add more elements to such kernel structures, extending them to the left, right or in both directions, we may find modal verb idioms which do have idiosyncratic semantic or formal properties and constraints. In our own contribution (Cappelle and Depraetere, this issue), we have discussed some examples of such longer sequences, such as the idiom Not if I can help it, which is routinely interpreted as a closed conditional (‘I can help it’), is relatively fixed in form (?This won’t happen if I can help it) and does not permit standard permutations (*If I can help it, then not). In this response to Hilpert’s paper, we will both support and extend his main claim. On the one hand, we will argue that Hilpert is right in stressing the importance of collocations in describing speakers’ knowledge of language. We will do so by taking the modal verb must as a case, although we will not compare the collocational preferences of present-day must with those of must as used in previous periods. Rather, we will only use contemporary data. On the other hand, we will argue that Hilpert’s emphasis on a modal’s collocational preferences need not – and in fact should not – be restricted to the following lexical infinitive. In short, we aim to adopt Hilpert’s focus on the collocational profile of a modal verb, showing that we can obtain further descriptive benefits by widening this focus so as to include several collocating elements at a time.