Download: pdf (25 pages, 226 Kb)
Abstract: A distributed connectionist framework for phonological development is proposed in which phonological representations are not predefined but emerge under the pressure of mediating among acoustic, semantic, and articulatory representations in the service of both comprehension and production. In this way, the approach sidesteps the perennial question of what are the specific ``units'' of phonological representation. Within the framework, articulatory feedback during speech production is derived from the acoustic consequences of the system's own articulations via a learned forward model of the physical mapping relating articulation to acoustics. An implementation of the framework, in the form of a discrete-time simple recurrent network, learned to comprehend, imitate, and intentionally name a corpus of 400 monosyllabic words, and its errors in development showed similar tendencies as those of young children. Although only a first step, the results provide support that the approach may ultimately form the basis for a comprehensive account of phonological development.
Copyright Notice: The documents distributed here have been provided as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.