Generative knowledge and item-specific knowledge trade off as a function of frequency in multiword expression processing

How do generative linguistic knowledge and knowledge of item-specific usage preferences jointly contribute to language processing? We investigated this using the test case of binomial expression ordering preferences (e.g. bread and butter vs. butter and bread), which are driven both by a) generative constraints referencing phonological, semantic, or other lexical properties of the elements in a binomial, e.g. a short-beforelong constraint; and b) item-specific experience in which an order is preferred purely because it has been experienced more often. In a forced-choice task looking at items that span the frequency continuum from very low to very high frequency, we asked how these two factors jointly determine binomial ordering preferences, and how their contributions vary based on expression frequency. We found that generative knowledge exerts a consistent role across the frequency continuum, while relative frequency plays an increasingly large role with increased expression frequency. We also found substantial reliance on item-specific knowledge at much lower frequencies than has been suggested previously. Our findings bring a quantitative perspective to the debate over the joint roles of generative and item-specific knowledge in language processing. We additionally connect our findings to models of language change, providing a test case for a unified account of how speakers use both generative and item-specific knowledge in processing to accommodate the existing structure of the language, and how these processing strategies in turn maintain the balance of generative structure versus idiosyncrasies in the language over time.