Informations and abstract
Keywords: morphological complexity, discriminative learning, recurrent neural networks, self-organisation, Russian verb inection.
The main focus of this paper is to investigate how aspects of morphological regularity may have an impact on early stages of word processing, prior to full lexical access. Here I explore the interaction of regularity and morphological structure by using a computational simulation of the process of learning Russian verb forms, without any morpho-syntactic or morphosemantic additional information. With a recurrent variant of self-organising memories, namely a Temporal Self-Organising Map, or TSOM, experimental results allow an investigation of the impact of incremental learning and online processing principles on paradigm organisation, by assessing the dierential impact of several aspects of regularity, ranging from formal transparency and predictability to allomorphy, on the processing/learning behaviour in a connectionist framework. The proposed analysis suggests a performance-oriented account of inectional regularity in morphology, whereby perception of morphological structure is not the by-product of the design of the human word processor, with rules separated from exceptions, but rather an emergent property of the dynamic self-organisation of stored lexical representations, dependent on the adaptive processing history of inected word forms, intrinsically graded and probabilistic.