Say it again, you'll be faster: Or on how the language production system keeps track of co-occurrences

The linguistic system allows for an almost infinite number of combinations, but speakers tend to use some word combinations more often than others. It would be intuitive to assume that these probabilities are registered and used to make the selection of upcoming words easier. The present study was designed to test this assumption. Participants named pictures using an adjective and a noun before and after a series of picture naming training sessions. During training, half of the presented pictures remained the same, while the other half were altered to maintain the same number of adjective and noun occurrences in novel combinations. The decrease in naming latencies after training is about 40 ms larger for repeated phrases. Experiment 2 replicates these results and shows that they are still observed after a day without additional exposure. Experiment 3 shows that these effects cannot be reduced to mere picture exposure. These findings challenge important assumptions of dominant psycholinguistic models. They suggest that the language production system registers direct word co-occurrences and stores the information for some time. We discuss an account of these effects in which connection strength between word representations are based on direct co-occurrence. Notably, these results open the way for detailed investigations and modeling of the mechanisms that drive probabilistic effects in language production.