Variations on a Bayesian Theme: Comparing Bayesian Models of Referential Reasoning
Variations on a Bayesian Theme: Comparing Bayesian Models of Referential Reasoning
复制标题
贝叶斯主题的变体：比较参照推理的贝叶斯模型
DOI：
--
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
M. Franke
中科院分区：
文献类型：
--
作者：
Ciyang Qing;M. Franke
贝叶斯实验语用学的最新发展受到了极大的关注。理性言语行为(RSA)模型定量地形式化了传统语用理论的核心概念，并做出了与经验数据相吻合的预测。本文通过阐述RSA模型的信念、目标和动作，分析了RSA模型及其与博弈论方法之间的关系。我们介绍了一些源于博弈论传统的备选方案，并系统地将这些备选方案与原始RSA模型进行了比较，采用贝叶斯模型比较的方法，比较了它们对相关经验数据的预测能力。结果表明，RSA模型可以进行调整和扩展，以提高其预测能力，特别是通过考虑说话人的偏好。
Recent developments in Bayesian experimental pragmatics have received much attention. The Rational Speech Act (RSA) model formalizes core concepts of traditional pragmatic theories quantitatively and makes predictions that fit empirical data nicely. In this paper, we analyze the RSA model and its relation to closely related game theoretic approaches, by spelling out its belief, goal and action components. We introduce some alternatives motivated from the game theoretic tradition and compare models incorporating these alternatives systematically to the original RSA model, using Bayesian model comparison, in terms of their ability to predict relevant empirical data. The result suggests that the RSA model could be adapted and extended to improve its predictive power, in particular by taking speaker preferences into account.