We conceive of time as a sequential order of real-world events, one ev
ent following another from past to present to future, This conception
colours the way we speak of time (''we look forward to the time'') and
, as we show here, the way we process written statements referring to
the temporal order of events, in real time. Terms such as 'before' and
'after' give us the linguistic freedom to express a series of events
(real or imaginary) in any order. However, sentences that present even
ts out of chronological order require additional discourse-level compu
tation. Here we examine how and when these computations are carried ou
t by contrasting brain potentials across two sentence types that diffe
r only in their initial word ('After' X, Y versus 'Before' X,Y). At si
tes on the left frontal scalp, the responses to 'before' and 'after' s
entences diverge within 300 ms; the size of this difference increases
over the course of the sentences and is correlated with individual wor
king-memory spans. Thus, we show that there are immediate and lasting
consequences for neural processing of the discourse implications of a
single word on sentence comprehension.