Wright J F, Ahmad K
Department of Mathematical and Computing Sciences, University of Surrey, Guildford, United Kingdom.
Brain Lang. 1997 Sep;59(2):367-89. doi: 10.1006/brln.1997.1821.
The simulation of language disorders using interactive activation (IA) networks and connectionist systems is discussed. An existing IA account of aphasic naming is described, in which two network parameters (decay rate and connection strength) are varied to fit the error production of an aphasic patient. Fairly similar results can be obtained through modification of additional parameters, including the so-called "shared weight increase factor" linking lexical and semantic units. This leads us to consider simulation of aphasic naming using connectionist networks which do not require explicit variation of network parameters. A modular connectionist architecture is presented, in which semantic-lexical and phonological knowledge are instantiated using self-organizing Kohonen maps, while connections between them are implemented using Hebbian networks; a linear connectionist network (Madaline) is used to simulate nonword repetition. The Hebbian connections are lesioned in order to reproduce the patient's naming errors.
本文讨论了使用交互式激活（IA）网络和联结主义系统对语言障碍进行模拟的问题。文中描述了一种现有的关于失语症命名的IA模型，该模型通过改变两个网络参数（衰减率和连接强度）来拟合失语症患者的错误产生情况。通过修改包括连接词汇和语义单元的所谓“共享权重增加因子”在内的其他参数，也可以获得相当相似的结果。这使我们考虑使用不需要明确改变网络参数的联结主义网络来模拟失语症命名。本文提出了一种模块化的联结主义架构，其中语义-词汇和语音知识通过自组织的科霍宁映射来实例化，而它们之间的连接则使用赫布网络来实现；使用线性联结主义网络（玛德琳）来模拟非词重复。通过损伤赫布连接来重现患者的命名错误。
