It’s time to do the math: Computation and retrieval in phrase production

Research suggests that the lexicon stores high-frequent combinations of words (e.g., Arnon & Snider, 2010), thereby optimizing the balance between computation and retrieval during language comprehension. Here, we studied whether the production of multi-word expressions is optimized in a similar way. First, we measured speech onset latencies for Dutch clock-time expressions. Second, we developed a statistical model of these latencies, using two types of predictors: (1) the speech-onset latencies for arithmetic problems involved in Dutch clock time naming, and (2) the expressions’ Google frequencies. The resulting model explains 94% of the variance in our naming study. We conclude that phrase production shows the same frequency-driven balance between online computation and lexicon-related retrieval found in phrase comprehension.