Abstract
Understanding how children process ambiguous words is a challenge because sense disambiguation depends on sentence context bottom-up and top-down aspects. Here, we seek in- sight into this phenomenon by investigating how such a com- petence might arise in large distributional learners (Transform- ers) that purport to acquire sense representations from lan- guage input in a largely unsupervised fashion. We investigated how sense disambiguation might be achieved using model rep- resentations derived from naturalistic child-directed speech. We tested a large pool of Transformer models, varying in their pretraining input size/nature as well as the size of their param- eter space. Tested across three behavioral experiments from the developmental literature, we found that these models cap- ture some essential properties of child sense disambiguation, although most still struggle in the more challenging tasks with contrastive cues. We discuss implications for both theories of word learning and for using Transformers to capture child lan- guage processing.