The cognitive and interactional causes of regularity in language

为什么人类是特别的？我们之所以特别，是因为我们的生理特征，还是我们相互学习的能力，还是两者兼而有之？我们通过研究语言共性、语言学习和语言传播之间的关系来解决这个基本问题。从表面上看，世界上的语言是相当不同的：它们使用不同的声音，以不同的方式构建复杂的话语。然而，所有语言似乎都共享一组共同的结构属性或设计特性。这些语言共性从何而来？一个有影响力的假设认为，它们直接反映了儿童在语言学习过程中强加给语言的高度限制性蓝图。另一种可能性是，这些特性可能是由于学习者的偏见较弱，这对特定儿童学习特定语言的影响很小，但由于语言在数千次这样的语言学习事件中的传播，影响很大。语言差异已经成为这类问题的一个测试案例。语言表现出多样性：例如，在英语中，我们对复数标记-s的准确发音是不同的（例如，有时“s”，如复数“cats”，有时“z”，如“dogs”）。然而，虽然从逻辑上讲，这种变化可能完全随机发生（例如，说话者可以随机选择使用“s”或“z”-所谓的自由变化），但这种行为很少（可能永远不会）发生在人类语言中。相反，语言变化是可预测的（在-s的情况下，从名词的最后一个音就可以预测发音），即可预测的变化是语言的设计特征。但是为什么语言是这样的呢？一些实验似乎支持这样的假设，即语言缺乏自由变化是由于儿童强加给语言的强烈偏见：儿童不期望不可预测的变化，因此在学习过程中忽略或消除它。这些实验使用一种被称为人工语言学习的方法，在这种方法中，儿童和成人学习微型语言（由实验者创造），然后测试他们在自己使用语言时是否会改变语言的某些方面。我们自己的工作使用相同的方法表明，儿童和成人学习者之间的差异可能不像以前的工作所表明的那样明显。当前项目的目标之一是进行进一步的学习实验，以便更好地理解成人和儿童学习者之间的差异，以及任何年龄的单个学习者可以在多大程度上重新塑造输入语言。因此，第一个研究方向是研究个体学习者如何对不可预测的语言做出反应。然而，如上所述，自然语言的可预测性可能是个体语言学习和使用中非常弱的偏好的结果，这种偏好在几代人的积累中积累起来。因此，我们将进行第二组实验，只使用成人参与者，在实验中，人们学习一种不可预测的语言，用它与同伴交流，然后通过传播链将他们的语言传递给其他学习者（一个相当于室内游戏“中国耳语”的实验，语言在人与人之间传递，并可能随着它的发展而变化）。这些实验，加上相应的计算模型，将使我们能够以一种非常简单的方式模拟语言在人群中的传播和使用。通过结合这两个方面，我们可以开始解决我们上面概述的基本问题：语言的普遍特征是否必然是由于儿童学习者对语言设计施加的强烈限制，还是在学习、传播和使用过程中固有的远较弱的偏见对人类语言结构产生强烈影响？

Why are humans special? Are we special because of our biology, or our capacity to learn from one another, or both? We address this fundamental question by studying the relationship between language universals, language learning, and language transmission. The languages of the world are superficially rather different: they employ different sounds and build complex utterances in different ways. However, all languages seem to share a common set of structural properties or design features. Where do these language universals come from? One influential hypothesis has suggested that they are a direct reflection of a highly-constraining blueprint for language, which children impose on language during language learning. An alternative possibility is that these properties might be due to weaker biases in learners, which have a very small impact for a particular child learning a particular language, but have a strong effect as a result of the transmission of language over thousands of such episodes of language learning. Linguistic variation has become a test-case for such questions. Languages exhibit variation: for instance, in English the precise way in which we pronounce the plural marker -s varies (e.g. sometimes "s", as in the plural "cats", sometimes "z", as in "dogs"). However, while it seems logically possible that this kind of variation could occur completely at random (e.g. speakers could randomly chose to use either "s" or "z" - so-called free variation), this type of behaviour very rarely (possibly never) occurs in human languages. Instead, linguistic variation is predictable (in the case of -s, the pronunciation is predictable from the last sound in the noun) i.e. predictable variation is a design feature of language. But why does language work like this? Some experiments seem to support the hypothesis that the lack of free variation in language is due to strong biases which children impose on language: children don't expect unpredictable variation, and therefore ignore or eliminate it during learning. These experiments use a methodology known as artificial language learning in which children and adults learn miniature languages (created by the experimenter) and are then tested to see whether they change aspects of the language when they use it themselves. Our own work using this same methodology suggests that the differences between children and adult learners may not be as clear cut as earlier work suggested. One of the objectives of the current project is to conduct further learning experiments in order to better understand this difference between adults and child learners, and the extent to which a single learner of any age can re-shape an input language. This first strand of research therefore studies how *individual* learners respond to unpredictable languages. However, as discussed above, the predictability of natural languages could be a consequence of very weak preferences in individual language learning and use, which accumulate over multiple generations. We will therefore run a second strand of experiments, using only adult participants, where people learn an unpredictable language, use it to communicate with a partner, and then transmit their language to other learners in a diffusion chain (an experimental equivalent to the parlour game "Chinese whispers", where language passes from person to person and potentially changes as it goes). These experiments, together with accompanying computational modelling, will allow us to simulate, in a very simple way, the transmission and use of language in a population. By combining both strands, we can begin to address the fundamental questions we outlined above: are the universal features of language necessarily due to strong restrictions on language design imposed by child learners, or can far weaker biases inherent in the processes of learning, transmission and use have strong effects on the structure of human language?