Constraints on Learning from Inconsistent Input

从不一致的输入中学习的限制

• 批准号: 7390362
• 负责人: CARLA L HUDSON KAM
• 金额: $ 18.03万
• 依托单位: UNIVERSITY OF CALIFORNIA BERKELEY
• 依托单位国家: 美国
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-04-01 至 2010-01-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7390362
• 关键词: Address; Adult; Affect; Age; Animals; Child; Childhood; Cognitive; Complex; Computer information processing; Data; Development; Disease; Elements; Goals; Human; Knowledge; Language; Language Development; Lead; Learning; Machine Learning; Modality; Nature; Numbers; Pattern; Positioning Attribute; Principal Investigator; Process; Property; Research; Research Design; Research Personnel; Series; Short-Term Memory; Source; Specificity; Time; Variant; Work; age related; design; programs; research study; stem

All human learners with normal cognitive capacities who are exposed to language input in childhood manage to acquire a language that looks very much like their input. However, at present we have little idea how, exactly, learners accomplish this feat. We have little understanding of the nature of the learning mechanisms that underlie and support language learning. This is the focus of the research in this proposal. Three general questions guide the research.(1) What are the constraints on language learning mechanisms? (2) Are the learning mechanisms involved in language acquisition specific to language or are they of a more general nature? (3) Do these mechanisms change over time as learners age, and if 'so, how? The proposed research examines these questions by investigating the learning of probabilistic and inconsistent patterns in the input using miniature artificial languages. This provides a rather unique view of the constraints on learning mechanisms, examining the limits of what can and cannot be learned. Previous work has shown that learners can acquire probabilistic patterns, however they sometimes impose consistency on variation. Moreover, children are more likely to change such patterns than are adults. The present research expands on the earlier results, asking about the nature of the interaction between learners and input. Series 1 asks about the nature of the input that makes some inconstancy learnable and some not. The studies examine the limits of veridical learning of inconsistent patterns in languages, asking questions about the specificity of computations learners can perform, the representations over which such computations can be performed, and the effect of prior domain-specific knowledge. Series 2 asks about the nature of the learner, asking why learners regularize over inconsistency at all.In particular, the studies examine whether working memory constraints, rather than constraints stemming directly from the learning mechanisms themselves, are an important factor influencing whether learners acquire the variation or instead impose regularity. Both series will be conducted with adults and children to examine how learning changes over development. Although the input in the proposed studies is somewhat atypical, the results from these studies will contribute to our understanding of the learning mechanisms involved in language acquisition more generally, and ultimately, this increases our understanding of both normal and disordered acquisition.

所有具有正常认知能力的人类学习者在童年时期接受语言输入 设法获取一种看起来非常类似于其输入的语言。但是，目前我们几乎没有 了解学习者如何完成这一壮举。我们对学习的本质几乎没有理解 基于语言学习的机制。这是研究的重点 提议。三个一般问题指导研究。（1）语言学习的限制是什么 机制？ （2）是专门针对语言的语言获取涉及的学习机制或 他们的性质更一般吗？ （3）随着学习者的年龄，这些机制会随着时间的流逝而改变，如果 ``那怎么样？拟议的研究通过研究概率的学习来研究这些问题 使用微型人工语言在输入中的模式不一致。这提供了一个相当独特的 对学习机制的限制的看法，检查了可以和无法学习的限制。 以前的工作表明，学习者可以获取概率模式，但是有时他们有时 对变异施加一致性。此外，孩子更有可能改变这种模式 成年人。本研究扩大了较早的结果，询问相互作用的性质 在学习者和输入之间。系列1询问使一些不稳定的输入的性质 可以学习，有些不可学习。研究研究了对不一致模式的垂直学习的局限性 语言，询问有关计算的特异性可以执行的问题， 可以执行此类计算的表示形式以及先前域特异性的效果 知识。系列2询问学习者的性质，询问为什么学习者正规化 尤其是不一致。研究检查工作记忆约束是否而不是 直接源于学习机制本身的约束是一个重要因素 影响学习者是获得变化还是施加规律性。两个系列将是 与成人和儿童一起进行研究，以研究学习如何改变发展。虽然 拟议的研究中的输入有些非典型，这些研究的结果将有助于我们 了解更普遍地了解语言涉及的学习机制，以及 最终，这增加了我们对正常和无序习得的理解。