A Computational Korean Lexical Access Model Using Artificial Neural Network

摘要

In this paper, we propose a computational Korean lexical access model based on connectionist approach. The model is designed to simulate the behaviors observed in human lexical decision task. The proposed model adopts a simple recurrent neural network architecture which takes a Korean string of 2-syllable length as an input and makes an output as a semantic vector representing semantic of the input. As experimental results, the model shows similar behaviors of human lexical decision task such as frequency effect, lexical status effect, word similarity effect, semantic priming effect, and visual degradation effect.