How semantic seeds shape what can be learned from the input: Evidence for semantic generalization over distributional categories

Early in language learning, learners may know the meanings of only a small subset of the words they encounter. We refer to these meaningful words as semantic seeds. In three experiments, we asked how this sparse semantic knowledge shapes what learners extract from distributional experience and whether the resulting semantic expectations facilitate word learning. Adults were first introduced to Seed pseudowords. Half of participants learned meanings for these Seeds, whereas the other half were only familiarized with Seed forms. All participants were then exposed to an artificial language with Subject-Object-Verb-like structure, containing Seed and additional, novel Test pseudowords. In Experiments 1a and 1b, learners with meaningful Seeds developed category-consistent semantic expectations for previously meaningless Test words based on their distributional histories. These expectations depended on systematic alignment between Seed meanings and distributional categories: disrupting that alignment eliminated the corresponding semantic preferences. However, these expectations did not reliably accelerate subsequent cross-situational learning of specific word-referent mappings. Experiment 2 showed that Seed meanings generalized to Test words that had never directly co-occurred with Seeds, indicating that semantic properties became associated with distributionally defined categories and were propagated throughout the language. While meaningful Seeds affected the semantic interpretation of distributional categories, they did not affect the acquisition of distributional structure itself; distributional category learning occurred even when Seeds were meaningless. Together, these findings suggest that although learners can acquire distributional structure without semantic knowledge, sparse word meanings can reveal semantic properties of that structure and support meaning hypotheses for otherwise unfamiliar words.