Title: Response bias modulates illusions of (un)grammaticality in subject-verb agreement: A diffusion model account

Abstract: Over the past decade, the comprehension of subject-verb agreement has been argued to best be modeled via the dynamics of direct-access, cue-based, memory retrieval. The central area of study has focused on when and how illusions in comprehension occur. For example, readers who are asked to judge sentences often fail to notice the ungrammaticality of sentences like The key to the cabinets are on the table. In sentences of this form, the correct form of the verb should be singular (i.e. is), as the true controller of agreement (key) is singular. However, the presence of a plural distractor noun (cabinets), which correctly covaries with the plural form of the verb (are), results in these sentences being erroneously judged as grammatical. Crucially, it has been shown that there is an asymmetry such that readers systematically judge ungrammatical sentences to be grammatical (i.e. there is an illusion of grammaticality), but not vice versa (there is no illusion of ungrammaticality). This asymmetry has been the empirical keystone in supporting memory-based accounts. In three experiments conducted jointly with Brian Dillon (UMass Linguistics) and Adrian Staub (UMass Psychology), I show that this asymmetry only appears when participants are biased towards making grammatical responses. When response bias is neutralized, both illusions of grammaticality and ungrammaticality appear. This finding calls into question the claim that the dynamics of memory retrieval lead to grammatical illusions in agreement comprehension. I instead model the results using a drift diffusion process, implemented using fast-dm (Voss & Voss, 2007), where grammatical illusions are argued to be a function of the rate at which evidence about the grammaticality of the sentence accumulates during the judgement decision process, and the asymmetry is a function of pre-decision response bias.