Morphology Learning as Paradigm Learning: Developmental and Computational Evidence from Romance Languages

In a comprehensive comparison of the developmental stages in the acquisition of inflection in nearly two dozen languages (in the Indo-European, Ugro-Finnic and Semitic families plus Turkish), Bittner et al. (2003) arrive at the conclusion that the transition from lexical processing to morphological patterning is not the automatic outcome of rote lexical storage, but rather the result of an active construction of the child, crucially conditioned by typological factors such as richness, uniformity and transparency of inflectional paradigms. In the present talk I intend to assess this hypothesis by observing the dynamics of a purely morphological acquisition of Romance verb paradigms through a family of Artificial Neural Networks known as Self- Organizing Maps (Kohonen 2002). I shall show that the interplay between built-in principles of acquisition of time-coded sequences and morphology-specific principles of organization of inflectional paradigms can go a long way in accounting for the typological trends highlighted in Bittner et al. (2003). Reported results allow us to draw some general conclusions concerning the process of morphology acquisition as paradigm-based learning and lead to a reappraisal of the traditional one-route vs. dual-route debate in morphology processing and learning.