When we talk, we are constantly referring to time. Our daily language is full of expressions that refer to the past or the future (‘let’s have lunch tomorrow’). Moreover, every verb provides a temporal anchor for the situation it describes, think of is-was-has been. The fact that languages of the world vary in their expression of time complicates the picture, though. We use translations to tackle this problem: in translation, the form changes, but the meaning in context remains stable. We show how we detect patterns of temporal reference across languages with the help of a computational approach to parallel corpora.