Constraints on language acquisition and how they change (or don't) with age

语言是人类的一种基本能力，它几乎在人类的每一项活动中都发挥着作用。然而，尽管它很重要，我们对语言实际上是如何习得的却知之甚少。这就是本研究的目的，以更好地了解所涉及的过程。使用微型人工语言，我研究学习者从不同类型的输入中获得什么，以发现什么是可学习的，为什么，以及这对儿童与成人学习者可能有什么不同。研究的重点是学习者从噪声输入数据中提取什么。在真实的生活中，语言学习者接受他们听到或看到的输入，并从中提取有关语言的信息。然而，输入往往只提供概率支持所获得的模式。例如，在一个示例中，在英语中，要表示过去式，你要在动词后面加上-艾德，也就是说，除非动词是不规则的。尽管存在不规则的情况，但当要求说话者在过去时态中使用一个新颖的动词时，他们通常会加上-艾德，表明他们已经提取了整体模式。言语错误也会引入噪音，但孩子们不会习得这样的规则：“如果事件发生在现在时，你有98%的时间在动词末尾使用单数主语-s”。更极端的例子是，孩子们从非母语者那里学习，他们通常比母语者产生更多的错误。然而，即使这些孩子最终也会得到一种规则的、确定性的语言，这与他们的输入非常不同。学习者似乎非常擅长从不完全符合模式的数据中提取模式。然而，有时候，要学习的模式实际上是可变的，这意味着孩子必须学习变化，而不是过滤掉它。我问学习者如何做到这一点，他们如何学习的模式，尽管变化，以及他们如何学习的模式是变化，目的是更好地理解人类语言学习的限制。这项研究将有助于我们对人类语言学习的理解，并可能为语言治疗提供信息，改善成人语言学习者的学习效果。

Language is a quintessential human ability - it plays a role in almost every human activity. Yet despite its importance, we know relatively little about how language is actually acquired. That is the goal of this research, to better understand the processes involved. Using miniature artificial languages, I examine what learners acquire from different kinds of input to discover what is learnable and why, and how this may differ for child vs. adult learners. The research focuses on what learners extract from noisy input data. In real life, language learners take the input they hear or see and from it extract regularities about the language. However, the input often provides only probabilistic support for the patterns acquired. E.g., to indicate past tense in English you add -ed to the verb, that is, unless the verb is irregular. Despite the presence of irregulars, when a speaker is asked to use a novel verb in the past tense, they will usually add -ed, indicating that they have extracted the overall pattern. Speech errors also introduce noise, yet children do not acquire a rule like '98% of the time you have a singular subject use -s on the end of the verb if the event is in the present'. Even more extreme examples are seen in children learning from non-native speakers, who usually produce many more errors than native speakers. However, even these children end up with a language that is regular and deterministic, very much unlike their input. Learners seem to be very good at extracting patterns from data that does not perfectly fit the pattern. Sometimes, however, the pattern to be learned is actually variable, meaning that children have to learn the variation, rather than filter it out. I ask how learners do this, how they learn the patterns despite the variation, as well as how they learn that the pattern is the variation, with the goal of better understanding the constraints on human language learning. This research will contribute to our understanding of human language learning, and may inform language therapy and improve learning outcomes for adult language learners.