Statistical approaches to linguistic pattern learning

项目摘要 这项拟议中的研究的长期目标是为儿童如何学习语法提供一个解释。 从语言输入中的分布信息来了解他们的母语结构，以及这些学习如何 机制可能不同于成人学习者。分布信息是元素在一个 大量的句子我们假设学习者从语言结构中习得语言结构的各个方面， 从这种分布信息中产生的统计数据，例如哪些元素共同出现，它们在什么位置， 规则地占据一个词或句子，以及它们经常与什么相邻的元素出现。我们 到目前为止，一个研究项目集中在分词上（学习者如何确定哪个声音 序列形成单词）和单词类别（学习者如何确定哪些单词形成语法 如名词和动词）。这项工作记录了婴儿的力量和健壮性， 儿童和成人使用复杂的分布信息来发现语言的这些方面的能力。 我们现在建议将我们的研究扩展到新的方向，研究学习高等教育的两个关键方面- 语言层次结构。在第一部分中，我们研究了导致学习者概括一个新的屈折变化的因素。 词素（如名词复数的-s）到新单词。在第二部分中，我们研究了学习者如何习得短语， 简单的层次结构的句子，我们问是什么导致学习者更喜欢的类型的短语和 自然语言中最常见的层次结构。 在我们提出的研究中，我们使用微型人工语言范式来测试我们的假设， 控制输入中的分布线索，这几乎是不可能的，只使用来自 自然语言学习在每一个实验中，参与者听一种微型语言的话语， 然后产生他们自己的话语或对他们的可接受性做出判断。关键是，在学习过程中， 在这个阶段，他们只听到人工语言中法律的可能的话语的一个样本;有些是 保留用于稍后的后测试，以确定学习者是否将他们观察到的内容概括为 新的实例（如果是，到哪些类型的新实例）。我们已经开发出非常成功的范例 让幼儿参与微型语言研究，我们已经证明了重要的差异， 儿童和成人语言学习者之间的差异。我们还将为儿童和成人提供 视觉运动领域的可比学习范式，以评估学习过程的时间进程 以及使用听觉语言材料的结果的特异性或普遍性。综合来看， 这些研究将记录使分布式学习机制能够获得 词的结构（屈折形态）和句子（短语和层次结构），并将突出 这些机制的方式可能因年龄和刺激域而异。

PROJECT SUMMARY The long-term aim of the proposed research is to provide an account of how children learn the grammatical structure of their native language from distributional information in linguistic input, and also how these learning mechanisms may differ from those of adult learners. Distributional information is the patterning of elements in a large corpus of sentences. We hypothesize that learners acquire aspects of language structure from the statistics arising from this distributional information, such as which elements co-occur, what positions they regularly occupy in a word or sentence, and with what neighboring elements they frequently occur. Our program of research to date has focused on word segmentation (how learners determine which sound sequences form words) and on word categories (how learners determine which words form grammatical categories such as noun and verb). This work has documented the power and robustness of infants’, children’s, and adults’ ability to use complex distributional information to discover these aspects of language. We now propose to extend our research in new directions, to examine two crucial aspects of learning higher- level linguistic structure. In Part 1 we study the factors that lead learners to generalize a novel inflectional morpheme (like –s for noun plurals) to novel words. In Part 2 we examine how learners acquire phrases and simple hierarchical structure in sentences, and we ask what leads learners to prefer the types of phrase and hierarchical structures that are most common in natural languages. In our proposed studies we test our hypotheses using miniature artificial language paradigms that afford control over the distributional cues in the input, something that is virtually impossible using only data from natural language learning. In each experiment, participants listen to utterances in a miniature language and then produce their own utterances or make judgments about their acceptability. Crucially, during the learning phase they hear only a sample of the possible utterances that are legal in the artificial language; some are withheld for use in a later post-test, to determine whether learners generalize what they have observed to novel instances (and if so, to which types of novel instances). We have developed highly successful paradigms for engaging young children in miniature language studies, and we have demonstrated important differences between child and adult language learners in these studies. We will also present children and adults with comparable learning paradigms in the visual-motor domain, to assess the time-course of the learning process and the specificity or generality of the results using auditory linguistic materials. Taken together, the results of these studies will document the key variables that enable a distributional learning mechanism to acquire the structure of words (inflectional morphology) and sentences (phrase and hierarchical structure) and will highlight the ways these mechanisms may differ over age and stimulus domain.