Almost two decades of research has demonstrated that labels facilitate infants’ categorization of novel objects. Some interpret this as evidence of an early link between infants’ linguistic and conceptual systems. Others suggest that these effects stem exclusively from lower-level processing mechanisms in cross-modal perception, and that words promote categorization only because they are more familiar to infants than non-linguistic acoustic stimuli and therefore easier to process. Here we address these discrepant interpretations using a novel approach. We expose infants to unfamiliar non-linguistic stimuli (sine-wave tone sequences), manipulating the exposure conditions. For 6-month-olds, if the novel acoustic stimuli were embedded within a communicative episode, they subsequently facilitated categorization (Experiment 1), but if they were presented in a non-communicative episode, they had no such effect (Experiment 2). We propose a developmental model that takes infants’ burgeoning perceptual and conceptual capacities into account in identifying how communication and words are linked to concepts.