The Interaction of Bayesian Pragmatics and Lexical Semantics in Linguistic Interpretation: Using Event-related Potentials to Investigate Probabilistic Predictions of Hearers

话语语境如何影响由词汇意义构成句子意义的方式？我们对比了如何解释上下文影响的两种相互竞争的观点。语义相似观认为，语篇语境对句子意义的影响主要是由于语篇语境中的词与目标句子中的词之间的语义相似（如语义启动，Otten & Van Berkum, 2008）。与此相反，自由语用观则为一种有争议的观点进行了辩护，即语篇语境的语用方面，除了单纯的指示物和类比的解析之外，还可以直接影响句子的意义构成，从而允许在句子意义构成的任何阶段自由地丰富语用（Recanati, 2010）。我们的项目旨在对这些相互矛盾的观点进行评判。为此，我们引入了一个预测完成任务，在这个任务中，听者在交际情境中的每一刻都必须对说话者所说的句子/话语如何继续进行概率预测。语义相似观和自由语用观对听者如何完成这一任务做出了不同的预测。这些预测可以定量确定，一方面，通过使用潜在语义分析（Landauer & Dumais, 1997）获得语义相似值，另一方面，通过采用贝叶斯语用学框架（Frank & Goodman, 2012）计算语用影响。我们使用Pustejovsky（1995）在生成词典方法中引入的所谓的质质结构作为词汇结构模型，并关注话语语境如何在具体名词的词汇条目中与Telic、agent和Formal成分相互作用。我们将测试EEG中语义相似度和自由语用观点的预测，使用经验证实的观察结果，即给定前一上下文的单词的条件概率与其N400分量的振幅呈负相关。根据实验数据，我们将形式语义学的工具与贝叶斯语用学的工具结合起来，建立一个话语语境如何通过自由语用丰富影响句子意义组成的模型。我们期望该项目的结果对语义极简主义和真理条件语用学之间的语言哲学辩论产生重大影响，特别是关于是否应该以或多或少严格的方式理解组合性。我们的目标是以一种与实验结果一致的方式重新调整组合原则。该项目还可以深入了解词汇和句子意义的心理表征结构，以及词汇检索和句子意义构成背后的神经认知过程。

How do discourse contexts influence the way sentence meaning is composed from lexical meaning? We contrast two competing views of how contextual influence can be explained. The Semantic Similarity View maintains that discourse context affects sentence meaning mainly because of the semantic similarity between the words in the discourse context and the words in the target sentence (e.g., as in semantic priming, Otten & Van Berkum, 2008). The Free Pragmatic View, in contrast, defends the controversial claim that also pragmatic aspects of the discourse context, other than the mere resolution of indexicals and anaphors, can immediately affect sentence meaning composition, allowing for a free pragmatic enrichment at any stage of sentence meaning composition (Recanati, 2010). Our project aims at adjudicating between these competing views. To this aim, we introduce a Predictive Completion Task in which the hearer at every moment in a communicative situation has to generate a probabilistic prediction about how a sentence/discourse being uttered by the speaker will be continued.The Semantic Similarity View and the Free Pragmatic View make different predictions as to how this task will be solved by the hearer. These predictions can be quantitatively determined, on the one hand, by using Latent Semantic Analysis (Landauer & Dumais, 1997) to obtain semantic similarity values and, on the other hand, by adopting the framework of Bayesian Pragmatics (Frank & Goodman, 2012) to calculate the pragmatic influence. We use the so-called Qualia Structure introduced in the Generative Lexicon approach by Pustejovsky (1995) as a model of lexical structure and focus on how discourse contexts interact with the Telic, the Agentive and the Formal components in the lexical entry of concrete nouns. We will test the predictions of the Semantic Similarity and the Free Pragmatic views in EEG using the empirically well-established observation that the conditional probability of a word given a preceding context is negatively correlated with the amplitude of its N400 component.In the light of the experimental data we will combine tools of formal semantics with those of Bayesian Pragmatics to develop a model of how discourse context affects the composition of sentence meanings by means of free pragmatic enrichment. We expect the results of the project to have a major impact on the debate in the philosophy of language between Semantic Minimalism and Truth-Conditional Pragmatics, and especially on whether compositionality should be understood in a more or less rigorous way. We aim at readjusting the principle of compositionality in a way that is consistent with the experimental findings. The project may also gain insight into the structure of mentally represented lexical and sentence meanings and the neuro-cognitive processes underlying lexical retrieval and sentence meaning composition.