Beyond tier-based bigrams: an artificial grammar learning study

Some of recently proposed phonotactic learners are tier-based bigram learners that restrict their hypothesis space to patterns between two segments that are adjacent at the tier level. This assumption is understandable considering that typologically frequent nonadjacent sound patterns are predominantly those that hold between two tier-adjacent segments. However, it is not clear whether the assumption is psychologically justified, i.e., whether speakers are indeed exclusively attentive to patterns between two tier-adjacent segments when it comes to learning nonadjacent sound patterns. In general, many recent studies suggest that learnable sound patterns are not limited to typologically observed sound patterns. Specifically, Koo and Callahan (2012) argue that adult speakers in laboratory settings have no trouble learning artificial patterns that cannot be explained by tier-based bigram learners. In this paper, we replicate their results in a more carefully controlled setting and argue that the assumption of tier-based bigram learning must be relaxed in order to properly explain human performance.

► We test the limits of human phonotactic learning with artificial grammar learning experiments.
► Human speakers can learn a wider variety of phonotactics than tier-based bigram learners predict.
► Our results are from more rigorously controlled experiments than a similar previous study.