Morpheme segmentation from distributional information

Morpheme segmentation is a process in language learning where the learner must parse morphologically complex words into their component parts (e.g., stems and affixes). Morpheme segmentation involves extracting the rule-like processes that can be abstracted or generalized from morphologically complex words. Because the bulk of research on learning morphological regularities has focused on form-meaning associations, it is unclear whether learners are able to extract statistical regularities on the formation of words without reference to meaning. The present study uses an artificial language learning experiment to test whether learners can decompose words into stems and affixes in the absence of semantic information, and whether this ability is greater for some types of affixation (prefixes and suffixes) over others (infixation). Participants were trained on CVCV stems with a CV affix that was always a suffix (Suffix condition), prefix (Prefix condition), or infix (Infix condition). Participants were able to successfully extract the regularities for both suffixes and prefixes, but not infixes. We discuss these results in light of prior language experience and general cognitive biases for word edges and local dependencies.