Real-world time-series data can show substantial short-term
variability as well as underlying long-term trends. Verbal
descriptions from a pilot study, in which participants
interpreted a real-world line graph about climate change,
revealed that trend interpretation might be problematic
(Experiment 1). The effect of providing a graph interpretation
strategy, via a linguistic warning, on the encoding of longterm
trends was then tested using eye tracking (Experiment
2). The linguistic warning was found to direct visual attention
to task-relevant information thus enabling more detailed
internal representations of the data to be formed. Language
may therefore be an effective tool to support users in making
appropriate spatial inferences about data