Large-scale evaluation of text features affecting perceived and actual text diffi

描述（由申请人提供）：随着越来越多的医学检查和治疗，越来越多的患者被诊断患有需要终身管理的慢性疾病，以及临床医生在有限时间内看到更多患者的压力越来越大，患者必须学习和了解如何最好地照顾他们的健康。不幸的是，估计有8900万人没有足够的健康知识来做到这一点，相关费用估计每年高达数十亿美元。虽然有许多令人兴奋的机会来教育消费者，从可视化到虚拟环境，文本仍然是社会中所有群体可用的最有效和最具成本效益的媒介。不幸的是，现有的写作指南，严重依赖于可读性公式，并没有被证明会影响文本难度或消费者的理解。因此，本项目的目标是解决和克服现有的障碍，在目前的可读性研究。我们解决障碍如下：我们将1）与代表性消费者一起工作，而不是代表消费者进行评估的专家，2）使用具有数千名参与者在他们自己的环境中工作的大样本，而不是在人工环境中只有少数参与者的实验室研究，以及3）评估文本的感知和实际难度，这两个变量经常在研究中混淆，4）发现可以在文本中自动发现的特征，使得可以开发难度检查器，其允许临床医生高效且有效地优化其文本，而不需要研究指南或语言学。设计本研究将利用现代技术和资源进行。在为期两年的时间内，硕士和博士水平的学生将与主要研究者一起设计和进行研究。这项研究汇集了计算语言学和信息科学应用于医学信息学的见解。 从语言学的角度出发，我们将系统地列出可能影响理解的文本特征的候选者。我们将研究文本的语法、语义和组成特征。通过使用一个现代化的市场，亚马逊的土耳其机器人，我们可以让成千上万的参与者参与这项研究。我们将对每个特征进行成对比较，例如，主题密度高低的句子。采用被试内设计，我们将测量多项选择问答任务中每个特征的感知和实际难度。

DESCRIPTION (provided by applicant): With increasingly more medical tests and treatments being available, more patients being diagnosed with chronic diseases that require life-long management, and increasing pressure on clinicians to see more patients in a limited amount of time, it is essential that patients learn and understand how best to take care of their health. Unfortunately, an estimated 89 million people have insufficient health literacy to do just that and the associated costs are estimated to be into the billions of dollars each year. Although there are many exciting opportunities to educate consumers, ranging from visualization to virtual environments, text is still the most efficient and cost-effective medium available to all groups in society. Unfortunately, existing writing guidelines, relying heavily on readability formulas, have not been shown to impact text difficulty or consumer understanding. Objectives The objectives of this project are therefore to address and overcome existing barriers in current readability research. We address barriers as follows: we will work 1) with representative consumers, not experts evaluating on behalf of consumers, 2) using a large sample with thousands of participants working in their own settings, not a laboratory study with few participants in an artificial environment, and 3) evaluate both perceived and actual difficulty of text, two variables often confounded in studies, 4) to discover features that can be automatically discovered in text so that difficulty checkers can be developed that allow clinicians to efficiently and effectively optimize their text without requiring study of guidelines or linguistics. Design The study will be conducted using modern technology and resources. Over a time period of two years, master and doctoral level students will design and conduct the studies together with the principle investigator. The study brings together insights from computational linguistics and information science applied to medical informatics. Starting from a linguistic perspective, we will we systematically list good candidates of text features that may influence understanding. We will look at features of grammar, semantics, and compositions of text. By using a modern market place, Amazon's Mechanical Turk, we can involve thousands of participants in the study. We will each pair-wise comparisons of the features, e.g., sentences with high versus low topic density. Using a within-subjects design, we will measure perceived and actual difficulty of each feature with multiple choice question-answering tasks.