We conducted a psycholinguistic experiment in which native speakers of Korean orally produced sentences using a subject, a temporal adjunct, a locative adjunct, an accusative object, and a verb, presented in a pseudo-randomized way on the computer screen before each trial. Results of the experiment indicate that Korean speakers produce the temporal adjunct earlier than the locative one irrespective of the manipulated length difference between the adjuncts. This ‘time-before-place’ preference is argued to reflect the production strategy to put the sentence constituents describing the event structure as closely as possible, thereby minimizing the cost for communicating the propositional contents of the predicate. The underlying assumption of this interpretation is that locative adjuncts belong to the necessary part of the event description, whereas the temporal adjuncts do not. We conclude that even in free word order languages like Korean, word order might well be constrained by various processing strategies operating at the performance level for a more efficient communication.