Modeling harmony biases in learning exceptions to vowel harmony

Artificial language learning experiments typically show non-categorical results after training on categorical data. This is generally due to incomplete learning, but these can also reveal biases. One example that participants trained a vowel harmony with alternating and non-alternating affixes prefer the affix in harmonic contexts (Finley 2021). In this paper, I (i) preference for items replicates remote (online) data collection, (ii) effect be modeled MaxEnt Harmonic Grammar. Grammar, score of each candidate determines its grammaticality, probability surfacing. Because satisfy have higher scores than violate harmony, candidates will more likely surface disharmonic candidates, even when both types at levels greater expected by chance. The theoretical methodological implications are discussed