认知草案的连通主义模型

A Connectionist Approach to Cognitive Script

认知语言学对草案的传统研究采用的是符号模型 ,这对研究静态草案是有效的 ,但忽略了草案的动态性 ,且难以解释诸如草案的来源、表征、激活规则及其推理等问题。Miikkulainen以分布式连通主义理论为依据提出的“等级特征映射系统”(HirarchicalFeatureMapSystem)有效地弥补了符号研究在解释草案问题方面的不足。该系统能从特殊草案示例中提取草案、路径和角色连接的等级分类。等级分类为每个激活的草案设定了特定的记忆位置。这样 ,不完整的草案及其空缺事件也得以自动填补。

Symbolic approaches to accounting for scripts are successful with static scripts, but do not address such questions as where the scripts come from, how the scripts are represented and what are their instantiation rules and inferences. Miikkulainen's hierarchical feature map system, as a distributed connectionist model can be effective in this respect. The system can extract the hierarchical taxomomy of scripts, tracks and role-bindings from examples of script-based input text. The script taxonomy assigns a unique memory location for a particular script. Incomplete story representations are automatically filled in.