Prediction in Language Across the Lifespan

认知功能在生命周期中波动，跨越无数领域。尽管语言理解的许多方面随着年龄的增长而保持和/或继续提高(例如，词汇)，但老年人(OA)似乎并不像年轻人(YA)那样预测即将到来的语言输入。鉴于目前的观点认为预测是人类语言理解的一个基本方面，这一发现令人惊讶。 根据基于资源的说法，这种所谓的语言预测下降是由于与年龄相关的执行功能普遍下降。然而，这一说法的证据喜忧参半。目前的工作测试了一个基于经验的竞争性账户的关键预测，其中语言预测在OA中是完整的，但预测的性质与YA的预测不同。这一建议建立在最近对语言处理的信息论描述的基础上，根据该描述，理解者使用当地语境(例如，前面的单词)以及他们在其一生中积累的长期语言知识来推导语言预测。这项基于经验的提议的核心关键是，鉴于在不同的时间段(语言统计数字在这段时间内)有多年的额外语言接触，办公自动化可能会做出不同的预测。为了检验基于资源的账户和基于经验的账户，本研究提出了两种互补的方法： 目标1：揭示语言统计数据如何在人的一生中发生变化。大规模的历时语料库将被用来描述几十年来语义和句法共现的变化。然后，将使用语言模型(n-gram，RNN)来测试单词/句子的可预测性取决于语言经验来源(例如，从训练数据被采样的时间段)而改变的假设。 目标2：产生一个大规模的刺激集，用于研究整个寿命的可预测性。将在网上进行一系列行为研究，以比较在相同语境下由OA和YA产生的预测，并为一大组刺激制定特定年龄的标准，用于设计未来关于语义和句法可预测性的实验。重要的是，这些标准将在一个大的年龄连续样本(18-80岁)中收集，并将提供整个生命周期内语境语言统计的直接测量。 本文提出的研究将为研究衰老对语言加工的影响提供重要的新线索。这些发现将有助于a)通过描绘健康老龄化和神经退化对语言理解的影响来早期发现痴呆症，以及b)支持终生认知健康的策略(例如，通过基于语言或执行功能的训练)。

Cognitive functioning fluctuates over the lifespan, across myriad domains. Though many aspects of language comprehension are preserved and/or continue to improve with age (e.g., vocabulary), older adults (OA) do not appear to predict upcoming linguistic input to the same extent as younger adults (YA). This finding is surprising given current views that prediction is a fundamental aspect of human language comprehension. On a Resource-based account, this alleged decline in linguistic prediction is due to a general age-related decline in executive functions. However, the evidence for this account is mixed. The present work tests key predictions of a competing Experience-based Account account whereby linguistic prediction is intact in OA, but the nature of the predictions differs from those of YA. This proposal builds on recent information-theoretic accounts of language processing, according to which comprehenders derive linguistic predictions using the local context (e.g., preceding words) along with their long-term knowledge of the language, accumulated across their lifespan. The critical observation at the core of the Experience-based proposal is that OA may make different predictions given the many additional years of language exposure over different time periods (over which language statistics fluctuate). In order to test the Resource-based and Experience-based accounts, the current research proposes two complementary approaches: Aim 1: Reveal how language statistics change over a person’s lifespan. Large-scale, diachronic language corpora will be used to characterize the changes in semantic and syntactic co-occurrences over the course of several decades. Language models (n-gram, RNN) will then be used to test the hypothesis that the predictability of a word/sentence changes depending on the source of language experience (e.g., the time period from which training data are sampled). Aim 2: Generate a large-scale stimulus set for the investigation of predictability across the lifespan. A series of behavioral studies will be conducted online, in order to compare predictions generated by OA and YA given the same contexts and develop age-specific norms for a large set of stimuli which can be used for designing future experiments on semantic and syntactic predictability. Critically, these norms will be collected in a large age-continuous sample (age 18-80) and will provide a direct measurement of contextual language statistics across the lifespan. The research proposed here will shed critical new light on the effects of aging on language processing. The findings will inform a) early detection of dementia by delineating the effects of healthy aging and neurodegeneration on language comprehension and b) strategies for supporting cognitive health across the lifespan (e.g., through linguistic or executive function-based training).