Purpose: The purpose of this study was to determine if the Vocabulary Acquisition and Usage for Late Talkers (VAULT) intervention could be efficaciously applied to a new treatment target: words a child neither understood nor said. We also assessed whether the type of context variability used to encourage semantic learning (i.e., action or object) would affect learning outcomes. Method: Nineteen primarily English-speaking late-talking toddlers received 8 weeks of VAULT intervention. They were quasirandomly assigned to a condition that highlighted either object or action variability. Individual effect sizes were calculated for target (treated) and control (not treated) words for each child. These were combined to assess group-level comparisons of treatment efficacy and treatment conditions. Generalization of the word-learning ability was assessed by comparing rates of learning on a vocabulary checklist prior to and during intervention. Bayesian statistics (e.g., t tests, analysis of variance) were used for the analyses. Results: There was strong evidence for a treatment effect showing that children produced more target than control words and moderate evidence that they understood more target than control words. There was strong evidence for generalization. Children learned an average of 6.8 words per week during treatment. There was anecdotal evidence for no difference between treatment conditions. Conclusions: VAULT, with a focus on context variability, can be used efficaciously to teach children to say words they do not understand at the start of treatment. The effects were most pronounced in the generalization data. Additionally, children were able to learn later-acquired words.