B05: Modelling the Information Landscape (IL) for Assessing and Analyzing Domain-Specific and Generic Critical Online Reasoning (DOM-COR and GEN-COR)

语言线索对文本可读性或网络来源可信度的作用已被广泛研究。B05的主要研究人员使用离线文本的短语料库也表明，语言特征对于预测大学生在特定领域知识测试中的表现很重要。然而，这种相关性在多大程度上可以推广到在线信息景观（IL）仍有待研究。B05解决了在线IL中使用的语言线索建模的关键需求，学生在解决关键在线推理（COR）任务时进行导航。B05旨在开发一个基于语言特征的理论模型，该模型可以根据学生在COR任务解决过程中处理或产生的文本来预测学生的COR过程（嵌入在IL中）和表现。B05解决了以下研究问题：(i)在经济学、医学、社会科学和物理学四个领域中，学生在通用与特定领域的语言特征表现上的差异有多大？（ii）这些特征在在线信息获取、关键信息评估和循证推理、论证和综合这三个认知COR方面有何不同？（iii）这些特征适用于哪些层次：单个文本，多个文本，领域，体裁，整个IL或基础语言（如德语）作为一个整体？B05由定量和定性两个部分组成。首先，对提供证据性地位、信息来源和文本组织信息的语言特征进行定性选择。定量部分的任务是对这些特征进行操作，使用基于机器学习的模型对其进行扩展，并针对上述研究问题测试其预测能力和特异性。定性和定量分析的整合具有计算解释学循环的形式，其中定量部分生成统计评估和预测，这些评估和预测可作为定性部分语言分析的结果进行解释。B05提供了机器学习模型，使多个文本的语言特征作为信息景观的一部分可用于自动分析，基于COR在细粒度语言信息单元级别的语言分析。a项目从纵向样本中提供学生表现的文本和数据，并将从B05的语言分析中获得结果。由于语言特征是迄今为止在研究单元中分析的最详细的信息单位，因此它们与其他b项目在媒体和内容属性（B04）以及叙事和潜在意义结构（B06）方面的研究相关。C08的多模式学习数据科学系统对于整合来自B05的所有数据至关重要。

The role of linguistic cues for text readability or Web source credibility has been widely studied. Using a corpus of short offline texts, principal investigators of B05 have also shown that linguistic features are important for predicting university student performance in domain-specific knowledge tests. However, it remains under-researched to what extent such correlations can be generalized to the online information landscape (IL). B05 addresses a key desideratum regarding the modelling of linguistic cues used in the online IL, which students navigate when solving critical online reasoning (COR) tasks. B05 aims to develop a theoretically grounded model of linguistic features that allows predictions of student COR processes (that are embedded in the IL) and performances, depending on the texts students process or produce during COR task-solving. B05 addresses the following research questions: (i) to what extent do the linguistic features involved differ in students’ performance in generic vs. domain-specific COR – and within the four domains of economics, medicine, social sciences and physics? (ii) How do these features differ with respect to the three cognitive COR facets of online information acquisition, critical information evaluation, and reasoning with evidence, argumentation and synthesis? (iii) At which levels do these features apply: single texts, multiple texts, domains, genres, the entire IL or the underlying language(s) (e.g. German) as a whole? B05 consists of both quantitative and qualitative parts. It starts with the qualitative selection of linguistic features that provide information on the evidentiality status, the source of information and the organization of texts. The quantitative part performs the task of operationalizing these features, expanding them using a machine learning-based model, and testing their predictive power and specificity with regard to the above research questions. The integration of qualitative and quantitative analyses has the form of a computational hermeneutic circle in which the quantitative part generates statistical evaluations and predictions that are interpretable as the results of the qualitative part’s linguistic analyses. B05 provides machine learning models that make linguistic features of multiple texts as part of the information landscape accessible to automated analysis, based on a linguistic analysis of COR at the level of fine-grained linguistic information units. The A-projects provide texts and data on students’ performances from the longitudinal sample and will obtain results from B05’s linguistic analyses. Since linguistic features are by far the most detailed information units analyzed in the research unit, they are relevant for research in the other B-projects in terms of media and content properties (B04) and narrative and latent meaning structures (B06). The Multimodal Learning Data Science System of C08 is crucial for integrating all data from B05.