In speech production, previously spoken and upcoming words can impinge on the word currently being said, resulting in perseverations (e.g., "beef needle soup") and anticipations (e.g., "cuff of coffee"). These errors reveal the extent to which the language-production system is focused on the past, the present, and the future and therefore are informative about how the system deals with serial order. This article offers a functional analysis of serial order in language and develops a general formal model. The centerpiece of the model is a prediction that the fraction of serial-order errors that are anticipatory, as opposed to perseveratory, can be closely predicted by overall error rate. The lower the error rate, the more anticipatory the errors are, regardless of the factors influencing error rate. The model is successfully applied to experimental and natural error data dealing with the effects of practice, speech rate, individual differences, age, and brain damage.