Over- and Under-generalization in Derivational Morphology

Abstract

Surfeit of the stimulus experiments (Moreton, 2008; Hayes et al., 2010; Becker et al., 2011), in which data contain patterns which are apparent to the analyst, but which human learners fail to learn, shed light on the nature of the human language learning faculty by demonstrating where and how it fails. This paper presents a natural language surfeit of the stimulus experiment, in which speakers mislearn a pattern apparent in the lexicon, overgeneralizing one output form at the expense of others. I present a model in which this overgeneralization is produced by incorporating both phonological knowledge and knowledge of the lexical type frequencies of each output form. I argue that adult speakers must know and employ lexical type frequencies in addition to phonological knowledge to produce novel morphological forms.

Documents:

Paper from the Proceedings of NELS 42
Handout from the 9th Old Word Conference in Phonology
Poster from a workshop on "How the mind/brain accommodates linguistic variability", held at the LSA institute in Ann Arbor, MI in July of 2013.

Data and scripts:

Production results (tab-delimited text)
Corpus of denominal verbs (PDF | tab-delimited text)
MaxEnt learner input file (tab-delimited text)

    The Flume, near Bass Lake CA.