How what you know changes what you learn: how existing knowledge impacts informativity in statistical learning

Arnon, I. & Siegelman, N.

Hebrew University

Why do adults rarely reach native-like proficiency in a second language? Existing accounts highlight the role of cognitive and neural differences between children and adults. We focus on the effect of adults' existing knowledge on the linguistic units they learn from, and how this impacts distributional learning. We suggest that adults prior knowledge of words leads them to rely less on multiword units, thereby hindering learning of certain grammatical relations between words by changing the input's information structure. If multiword units facilitate learning, then having adults learn from larger units should lead to better learning outcomes. Recent findings supports this prediction: adults showed better learning of grammatical gender in an artificial language when exposed first to sentences (multiword units) and only then to individual words [Arnon-Ramscar-2012]. Here, we extend this finding by showing (1)that exposure to unsegmented speech leads adult learners to extract more multiword building blocks, resulting in better learning of grammatical gender; and (2)that larger units do not impact learning outcomes when the ?grammatical? classes are semantically-conditioned (animate vs. inanimate). The findings highlight the impact of early building blocks on subsequent learning and illustrate how prior knowledge changes the information conveyed by the same distributional input.