Lexical selection in language production

Language production models typically characterize lexical selection as choosing the single best word to describe a concept, using competition- based algorithms to implement that goal. Name agreement effects in simple picture naming have provided crucial evidence that competition governs selection even in the absence of strong task demands: slow responses are linked to measures of response dispersion by assuming that coactivated alternatives delay dominant name selection. This paper proposes a strictly non-competitive parallel activation model of lexical selection, whereby speakers efficiently choose stochastically appropriate words by simply selecting the first word to reach an absolute threshold. Two simulations demonstrate that it can explain all major aspects of name agreement effects and generate a novel prediction linking strong secondary names with faster dominant name production. Two experiments confirm this prediction, also showing that it holds for speakers who demonstrably use both names and is not detectably modulated by relationships between co-activated responses. Thus, while speakers can clearly exert some control over the end point of word production when a task demands it, the broader computational goal of lexical selection is better characterized as finding a ‘good enough’ word than as finding the best word available.