Approaches to creativity commonly distinguish between F-creativity (rule-compliant use) and E-creativity (rule-breaking use). This dichotomy in part stems from a focus on grammatical constructions (‘nodes’) at the relative expense of their connections (‘links’). We approach creativity and productivity from a link-based perspective in Usage-Based Construction Grammar, and assume that productivity pertains to a unit’s inventory of links, while creativity pertains to the creation and maintenance of links. These assumptions are showcased using the into-causative (He talked me into going, They scared us into working harder). The construction is productive because it hosts a large inventory of verbal slot-fillers (talk, scare). Conversely, these slot-fillers themselves are creative because they can establish and maintain links with a construction that is not their primary host. This property is not linear: we assume that the slot-fillers’ ability to occur in unusual constructional environments reflects their general ‘creative potential’ to form and maintain (new) links within the network. In data from the Corpus of Contemporary American English (COCA), we find weak, but consistent correlations between verbs’ association with the into-causative and (i) their semantic and syntactic compatibility with the construction, and, crucially, (ii) their general flexibility and ability to establish and maintain links.

How do we know that would rather and may well are more idiomatic than would well or will really? Can this intuition be measured systematically in usage data? Traditionally, modal idioms such had/’d better, would/’d rather or might (as) well are seen as distinct from more compositional collocations, which may be modally harmonic (could possibly, will probably) or not (could also, might even). Yet the collocation of modal auxiliaries + adverbs (mod + adv) is more complex than suggested by a binary classification into idioms and non-idioms. This article uses data from COCA and the method of collostructional analysis to show that the difference between qualitatively distinct types of mod + adv is a matter of degree. Modal idiomaticity should be seen as gradient along a continuum from strong association (would rather) to strong dissociation (would well). The results support assumptions that statistical information about the collocational behavior of modal auxiliaries is a cue for the scope of adverbial modification and is thus an important aspect of speakers’ knowledge of modal meaning. The study contributes to recent approaches to modality from a ‘combinatorial’ perspective, which recognizes the importance of the lexical environment in core areas of grammar.