We are using Latent-Semantic Analysis in developing new ways of measuring comprehension. Latent-Semantic Analysis or LSA is a statically-based tool that estimates the semantic similarity between two units of text (words, paragraphs). We are using this tool to uncover the basic strategies of comprehension that people use when they read stories and textbooks. The new measure will use LSA to compare protocols (thoughts) that the user types in as she is reading a text with previously coded protocols, allowing the computer to hypothesize as to the content and strategy of the user's protocol.
Whether searching the WWW or writing a research paper, people frequently find themselves in situations that require skilled use and evaluation of information from text. We have identified several such skills including: Sourcing (i.e., noticing and evaluating a document’s source), Corroborating (i.e., checking new information against other independent sources), and Integrating (i.e., making connections between prior and new knowledge). Although these advanced literacy skills were derived from research on history learning, we posit that they are not specifically the purview of any one domain but are used by all disciplines to some extent. In an effort to train students in the use of these skills, we created the Sourcer's Apprentice (SA); a Java application that provides explicit tutoring and structured practice opportunities for students to engage in sourcing and corroboration.
Compared to concrete concepts, relatively little is know how abstract concepts such as "love" and "goal" are represented, organized, and constructed in the human mind. Abstract concepts are very elusive, and it is an interesting question how we compare them semantically. Some abstract concepts are highly similar in meaning (e.g., "happiness" & "gladness"), and we can use either of them in the same place. However, other abstract concepts are related in different ways. E.g., "action" & "consequence" are related, but they are not similar. This large-scale project aims to do several things: (1) compare how abstract and concrete concepts are organized and interrelated with different sorting and rating tasks, (2) test various hypotheses about how and whether abstract concept similarity is based on the number of contexts in which they can occur, and (3) examine abstractness ratings in four different languages: Chinese, English, German, and Spanish.
We are developing a computer program that detects presuppositions in questions. An example for a presupposition: when you are asked where you have parked your car, the presupposition is that you own a car, and have come with it to wherever you are now. We will further equip the system with a mechanism to validate the presuppositions, i.e., to find out from context if the presuppositions can be made (i.e., if one can tell from previous discourse whether indeed you have come there with your car).
Situation models refer to "deep understanding" of discourse, film, and artwork. We have several projects that examine how situation models are constructed during comprehension. In addition, we are comparing situation models which are constructed from text with models that are constructed from viewing film. Lastly, we are testing Kintsch's construction-integration model in the context of artwork and aesthetic experiences.