Statistical Learning of Conjunctive Probabilities

AbstractMost statistical learning studies focus on the learning of transitional probabilities between adjacent elements in a sequence, however, other statistical regularities may underpin different aspects of processing language and regularities in other domains. Here, we investigate how conjunctive statistical regularities (of the form A and B together predict C) can be learned, and how this learning is impacted by similarity in representations analogous to that in unambiguous words, homonyms with multiple unrelated meanings, and polysemes with multiple related meanings. We observed that provided the stimulus structure is relatively simple, participants are readily able to learn conjunctive probabilities and display sensitivity to relatedness among representations. These results open new theoretical possibilities for exploring the domain-generality of how the learning and processing systems merge conjunctive information in simple laboratory tasks and in natural language.

Return to previous page