A substantive inductive bias for *NonFinalR in an artificial grammar learning replication experiment

Substantive inductive biases are preferences for acquiring phonological regularities motivated by articulatory or perceptual naturalness. Findings from various experimental studies yielded mixed results, leading to debates over whether substantive biases exist. In one Web-based artificial grammar learning experiment, we revisited the substantive bias toward learning a constraint against non-phrasal-final rising tones (*NonFinalR) in [Chen, T.-Y. (2020). An inductive learning bias toward phonetically driven tonal phonotactics. Language Acquisition, 27(3), 331–361. https://doi.org/10.1080/10489223.2020.1769630]. We compared the learnability of *NonFinalR and substance-free tonal constraints and analyzed learners’ performance with mixed-effects regression modelling and a machine-learning classification algorithm (RIPPER). The quantitative analyses suggested that all target tonal constraints were learned equally well, and learners seemed equally (un)aware of acquired tonal knowledge. Partial support for the biased learning of *NonFinalR was found only in the judgments of true accidental and systematic gaps and in classification rules induced by RIPPER. Our findings suggest a weak tonal substantive bias and further underscore the importance of replication studies and diverse analytical approaches.