To acquire a new word, learners need to create its representation in phonological short-term memory (STM) and then encode it in their long-term memory. Two strategies can enable word representation in STM: universal segmentation and phonological mapping. Universal segmentation is language universal and thus should predict word learning in any language, while phonological mapping is language specific. This study investigates the mechanisms of vocabulary learning through a comparison of vocabulary learning task results in multiple languages. We tested 44 Polish third graders learning English on phonological STM, phonological awareness in Polish and in English, and on three tasks, which involved learning novel word forms in Polish (first language), in English (second language), and in a language that did not resemble any language known to participants (an unknown language). Participants’ English proficiency was controlled through a vocabulary task. The results suggest that word learning engages different mechanisms for familiar and unfamiliar languages. Phonological awareness in English predicted learning second language and unknown language words, and phonological STM predicted learning words of the unknown language. We propose that universal segmentation facilitates word learning only in an unfamiliar language, while in familiar languages speakers use phonological mapping in order to learn new words.