Prior research had documented that semantically-similar labels that co-occur in child-directed speech promote generalization in young children. The present study examined whether co-occurrence probability – in the absence of semantic similarity – can influence children’s inferences. Four- and five-year-old children were exposed to an auditory speech stream consisting of trisyllabic nonsense words (e.g. “golabu”) that were concatenated into a continuous speech stream. After listening to the stream, children were given a label extension task where the first two syllables of a nonsense word were assigned to a novel target object (e.g. “gola”); children were asked to choose which of the three test items should be referred to by the remaining syllable of this nonsense word (e.g., “bu”; Experimental) or by a syllable from a different nonsense word (e.g., “ti”; Control). Children’s generalization performance in this task was similar to results of previous research that used natural rather than artificial language.