Modelling the Use of Quantifiers in Typical and Atypical Speakers Probabilistically

自亚里士多德的工作以来，哲学家、语言学家、逻辑学家和心理学家一直在研究表示量化的词(例如，一些、大多数)。目前形式语义学中的共识是，许多或大多数量词表示泛化的量词，即集合之间的关系，具有明确定义但通常逻辑上较弱的含义。然而，来自关于量词使用和解释的心理学实验的证据表明，情况截然不同。例如，实验参与者会从一些A是B中推断出大约四分之一的A是B，而不是像通常的语义含义所暗示的那样，至少有一个A是B。我们把在实验数据中观察到的这些窄范围称为‘焦点范围’。此外，实验数据证明，量词的焦点范围通常不是精确描述的，而是可能有一个模糊和难以捉摸的周长。尽管如此，量词使用和解释的规律性仍然存在，因此本项目的主要目标是在形式语义和量词使用和解释的心理数据之间建立一座桥梁。我们的做法是务实的。我们推测，语言理论和实验数据之间的明显差异可以通过语用学来解决，我们可以通过对量词使用的研究来理解一般的语用原则。相关原则还包括其他相关的认知因素，如数量感知和加工中的不精确和错误，对话者词典的不确定性，所需精度的语境相关水平的不确定性，以及语言使用者能够或愿意参与心理推理理论以收敛于最佳语言使用的程度的限制。由于这样一个心理丰富的图景可能很快就会变得笨拙，我们将使用数据驱动的概率建模，遵循最近的博弈论和贝叶斯方法。实际上，这个项目开始从一个统一的任务系列中系统地收集数据，以衡量量化表达的产生和解释。我们特别感兴趣的是说话人生产和听话人理解之间的潜在不对称，以及重复的说话人-听话人互动中的动态调整过程。此外，我们不仅会招募神经典型的成年人，也会招募神经典型的儿童和患有自闭症谱系障碍的成年人。众所周知，这两个群体都经历了语用缺陷。通过在静态或动态环境中使用统一的刺激和任务类型从这些不同的人群中收集数据，我们将能够使用显式概率模型来检验对患有自闭症谱系障碍的儿童和成人的语用能力的现有理论描述。

The study of words that express quantification (e.g., some, most) has occupied philosophers, linguists, logicians, and psychologists since the work of Aristotle. The current consensus in formal semantics is that many or most quantifying expressions denote generalised quantifiers, i.e., relations between sets, with clearly defined but usually logically weak meanings. Evidence from psychological experiments on quantifier use and interpretation, however, suggests a rather different picture. For example, participants in experiments would infer from Some A are B that about a quarter of the A are B, rather than at least one A, as the usual semantic denotation of `some' implies. We call these narrow ranges observed in experimental data `focal ranges'. Moreover, a quantifier's focal range, as attested by experimental data, is usually not precisely characterised but may have a vague and elusive circumference. Still, regularities in quantifier use and interpretation persist, and so the main goal of this project is to build a bridge between formal semantics and psychological data of quantifier use and interpretation. Our approach is pragmatic. We conjecture that apparent discrepancies between linguistic theory and experimental data can be resolved by pragmatics and that we can use the study of quantifier use to understand general pragmatic principles. The relevant principles crucially include further relevant cognitive factors, such as imprecision and error in the perception and processing of quantity, uncertainty about an interlocutor's lexicon, uncertainty about contextually relevant levels of required precision, and limits on the extent to which language users are able or willing to engage in theory of mind reasoning to converge on optimal language use. Since such a psychologically rich picture can grow unwieldy very soon, we will make use of data-driven probabilistic modeling, following recent game-theoretic and Bayesian approaches.Empirically, this project sets out to systematically collect data from a uniform family of tasks gauging the production and interpretation of quantifying expressions. We are specifically interested in potential asymmetries between speaker production and listener comprehension, as well as in dynamic alignment processes in repeated speaker-listener interaction. Moreover, we will recruit not only neurotypical adults, but also neurotypical children and adults with autism spectrum disorder. Both of these latter groups are known to experience pragmatic deficits. By collecting data using uniform stimuli and task types in static or dynamic settings from these different populations, we will be able to test, using explicit probabilistic modeling, extant theoretical accounts of the pragmatic abilities of children and adults with autism spectrum disorder.