Learning Morphological Productivity as Meaning-Form Mappings

Jordan Kodner, Sarah Payne, Charles Yang

Abstract: Child language acquisition is famously accurate despite the sparsity of linguistic input. In this paper, we introduce a cognitively motivated method for morphological acquisition with a special focus on verbal inflections. Using UniMorph annotations as an approximation of children’s semantic representation of verbal inflection, we use the Tolerance Principle to explicitly identify the formal processes of segmentation and mutation that productively encode the semantic relations (eg, past tense) between stems and inflected forms. Using a child-directed corpus of verbal inflection forms, our model acquires the verbal inflection morphemes of Spanish and English as a list of explicit and linguistically interpretable rules of suffixation and stem change corresponding to sets of semantic features.