Norris Dennis
Medical Research Council Cognition and Brain Sciences Unit, Cambridge, United Kingdom.
Psychol Rev. 2009 Jan;116(1):207-19. doi: 10.1037/a0014259.
R. Ratcliff, P. Gomez, and G. McKoon (2004) suggested much of what goes on in lexical decision is attributable to decision processes and may not be particularly informative about word recognition. They proposed that lexical decision should be characterized by a decision process, taking the form of a drift-diffusion model (R. Ratcliff, 1978), that operates on the output of lexical model. The present article argues that the distinction between perception and decision making is unnecessary and that it is possible to give a unified account of both lexical processing and decision making. This claim is supported by formal arguments and reinforced by simulations showing how the Bayesian Reader model (D. Norris, 2006) can be extended to fit the data on reaction time distributions collected by Ratcliff, Gomez, and McKoon simply by adding extra sources of noise. The Bayesian Reader gives an integrated explanation of both word recognition and decision making, using fewer parameters than the diffusion model. It can be thought of as a Bayesian diffusion model, which subsumes Ratcliff's drift-diffusion model as a special case.
R. 拉特克利夫、P. 戈麦斯和G. 麦昆（2004年）指出，词汇判断过程中发生的许多情况都可归因于决策过程，可能对单词识别并无特别的指导意义。他们提出，词汇判断应以一种决策过程为特征，采用漂移扩散模型（R. 拉特克利夫，1978年）的形式，该模型作用于词汇模型的输出。本文认为，区分感知和决策是不必要的，并且有可能对词汇处理和决策做出统一的解释。这一观点得到了形式论证的支持，并通过模拟得到了加强，模拟展示了贝叶斯阅读者模型（D. 诺里斯，2006年）如何只需添加额外的噪声源就能扩展以拟合拉特克利夫、戈麦斯和麦昆收集的反应时间分布数据。贝叶斯阅读者对单词识别和决策都给出了综合解释，使用的参数比扩散模型少。它可以被看作是一个贝叶斯扩散模型，拉特克利夫的漂移扩散模型是其特殊情况。
