Previous studies investigating effects of language comprehension on spatial processing have used existing words with pre-existing spatial associations. Here participants learnt novel words and novel objects with spatial associations. Following training, participants had to judge whether a visual object matched a word. Objects could match in identity or in spatial location. In Experiment 1, participants learnt just novel words and objects; Experiment 2 compared performance with existing objects with pre-existing spatial associations. We found mismatching (but task irrelevant) spatial information interfered with judgements of object identity, but only for novel words. In Experiment 3, we altered correspondence between visual targets and semantics using a target discrimination task, where the target had no relationship to the verbal cue. We found the opposite results to the previous two studies, as responses to spatially matching targets were slower than spatially mismatching targets. We discuss implications for embodied and non-embodied accounts of these findings.