文本信息加工中的知识结构的测量

Measures of Knowledge Structure in Reading Comprehension

知识结构是信息加工中各类信息在记忆系统中的组织结构的反映。为更好地表征和分析知识结构,研究者从教育科学技术和计算机技术的角度发展出一系列的研究方法和测量工具,包括潜在语义分析、心理模型分析和语义聚合分析。研究者可以结合具体情况选用合适的测量工具。而如何实现眼动、功能性成像数据与当前的测量相结合,则需要研究者在未来的工作中展开进一步探讨。

Reading comprehension is a complex process,in which text content information interrelates with reader’s prior structural knowledge to form an integrated understanding of the text.Knowledge structure(KS) is the apposite organization of text information.Using available measures,mainly computer-based technologies developed in the last two decades,researchers have shown that knowledge structure distinctly affects several cognitive processing outcomes,including reading comprehension scores of new texts,second language learning,and performance in problem solving.In the present report,we introduced some key methods which were widely used in the recent years.In the literature,summary essays,concept maps,the scores of pair-wise task,list-wise task,and sorting task,combined with eye-tracking and fMRI data,were considered as indexes of KS,and researchers have formalized how to transform all these various kinds of data into readable and comparable measures.Latent semantic analysis(LSA) developed in 1990s is one related measure that is based on the underlying structure of content that is distributed across many documents.LSA establishes a high-dimensional semantic space consisting of associations of words using the computationally demanding singular value decomposition of a large document by terms matrix(based on thousands of documents).In this way,similarity among different texts and between terms in texts is established as the cosine between their vectors in that semantic space.However,the matrix needed in LSA is so large that it takes extensive computing power to create it,so some researchers have turned to more efficient local semantic spaces such as SMD(Surface,Matching,Deep Structure) and T-MITOCAR(Text-Model Inspection Trace of Concepts and Relations).SMD is used to analyze concept maps,whereas T-MITOCAR is an adaptation of SMD that can be used to analyze essays.Researchers who prefer using SMD and T-MITOCAR hold that knowledge is usually organized and stored as hierarchical structures or networks in human memory.Under SMD or T-MITOCAR,mental models could be indirectly measured.Analysis of lexical aggregates(ALA) is another alternative method which can be used to investigate knowledge structure from essays.Different from T-MOTICAR,ALA emphasize proximity/adjacency of different nodes rather than termterm distance.After obtaining a matrix of data based on preselected keywords from ALA-Reader,the matrix is then analyzed using Pathfinder network analysis to obtain a PFNet(i.e.,a kind of graphical representation of KS) that can be used to compare the similarity of each network to a benchmark referent.Generally,results received via ALA are more intuitively displayed than those received via other methods.Generally speaking,both mental model analysis and analysis of lexical aggregate are easy to use and can provide various information.Such network approaches provide new ways to measure and describe learning and behavioral performance,and so researchers are committed to improve the functions of these technologies.For example,AKOVIA(Automated Knowledge Visualization and Assessment;developed version of T-MITOCAR)and GIKS(Graphical Interface of Knowledge Structure;developed version of ALA-Reader) are available now on an internet platform,which can provide immediate visual feedback(as concept map) of student’s essay writing.However,improvements are still required,and suggestions were proposed at the end of the current report.