The Role of Working Memory in Statistical Word Learning

Statistical word learning (SWL) has been proposed and tested as a powerful mechanism for word learning under referential ambiguity. Learners are adept at resolving word-referent ambiguity by calculating the co-occurrences between words and referents across ambiguous scenes. Despite the generalizability of such capacity, it is less clear which underlying factors may play a role in SWL, such as learners’ language experience and individual differences of working memory. The current study therefore asked two questions: 1) How do learners of different language experience (monolinguals and bilinguals) approach SWL of different mapping types–when each referent has one name (1:1 mapping) or two names (2:1 mapping)? and 2) How do working memory capacities (spatial and phonological) play a role in SWL by mapping type? In this pre-registered study (OSF: https://osf.io/mte8s/), 69 English monolinguals and 88 bilinguals completed two SWL tasks (1:1 and 2:1 mapping), a symmetry span task indexing spatial working memory, and a listening span task indexing phonological working memory. Results showed no differences between monolinguals and bilinguals in SWL of both mapping types. However, spatial and phonological working memory positively predicted SWL regardless of language experience, but only in 1:1 mapping. The findings show a dissociation of working memory’s role in SWL of different mapping types. The study proposes a novel insight into a theoretical debate underlying statistical learning mechanisms: learners may adopt more explicit processes (i.e. hypothesis-testing) during 1:1 mapping but implicit processes (i.e. associative learning) during 2:1 mapping. Future studies can locate memory-related brain areas during SWL to test out the proposal.