This paper asks if language learners show a preference or bias for syllable-based generalizations in morphophonology. In particular, we address the patterns of infixation with syllable as a pivot in an artificial language learning experiment. Our study found no evidence for a learning bias towards syllable-based infixation patterns, in line with the existing results on typology of infixation pivots (Yu 2007; Samuels 2010; Wilson 2014). On the other hand, our results are consistent with perceptibility-based accounts of cluster splittability where learners expect that infixes will more likely break up CC clusters that are perceptually similar to CVC sequences (Fleischhacker 2005; Zuraw 2007; Yun 2016). Our results thus suggest a role for grammatically-encoded perceptual similarity (P-Map, Steriade 2009) in learning novel infixation patterns.