Phonotactic constraints on how sounds can be combined to form words vary by language, but certain patterns are more commonly attested across the languages of the world. For instance, a large number of typologically different languages restrict transvocalic consonants with the same major place of articulation (e.g., *gVk, *fVp), whereas the reverse pattern (i.e., long-distance agreement in major consonantal place) is rare. One explanation of these typological asymmetries in phonotactics is that humans are equipped with cognitive dispositions to learn certain types of phonological generalisations over others. However, this explanation is met with empirical challenges as studies have not consistently demonstrated that adults preferentially learn novel phonotactic rules that align with typological observations. Additionally, the theory that humans possess an inductive bias for these generalisations does not adequately explain why the linguistic input to learners may drift in a particular direction in the first place.

In this talk, I explore an alternative account of how these typological skews in phonotactics emerge. The key insight of this account is that phonotactics are generalisations derived from the lexicon, and it is the lexicon that is shaped over time as word forms that are easier to learn and better at preserving distinctiveness tend to be retained or innovated during transmission across generations of speakers. I will discuss a series of studies examining this lexicon-driven model and its capacity to account for the observed typological asymmetries in phonotactic distributions.