In this thesis, I investigate whether learners’ avoidance of alternation and neutralization, as well as learners’ exposure to their native language (L1), affect how they learn new morpho-phonological patterns. While the effect of individual factors on morpho-phonological learning has been widely studied, whether these factors have a collective effect on learning and interact with the frequency of variants in the input has been understudied. To explore whether there are any interactive effects of these factors, I modify the type of alternations, learners’ native languages, and relative frequency of variants across several repetitions of an experiment. I exposed adult English speakers to an artificial language in which plural forms were probabilistically marked by one of two prefixes. One of the prefixes triggered either a non- neutralizing or neutralizing alternation that could create homophony. I found that English speakers generally matched the relative input frequency to their output. However, learners avoided the construction that resulted in a phonological alternation, but only when it was infrequent. This finding suggests that though there is a tendency to avoid alternations, it depends on how frequent the relative variants are in the input. Moreover, English speakers were poorer at learning the neutralizing alternation than the non-neutralizing alternation, showing their bias against neutralization that can create homophony. Additionally, I replicated the same experiments with Korean speakers because there is abundant exposure to neutralization in their L1. I found that Korean speakers were successful at learning both neutralizing and non-neutralizing alternations, suggesting that having abundant exposure to neutralization can make new neutralization easier to learn. Finally, I argue for a model which implements the avoidance effect as a discounting of observations that trigger homophony in the training data, rather than requiring a special constraint penalizing neutralization in the grammar. This Discount model correctly predicts the different
learning results between English and Korean speakers and provides a straightforward explanation for learners’ bias against neutralization and homophony. This approach places the locus of the bias in the learning process rather than in the grammar.