Walter Kintsch, Ph.D. - US grants

This project is concerned with the investigation, elaboration, and experimental testing of a general theory of text comprehension. The strategies are investigated that people use in constructing a coherent representation of a text, both at the local and global level (macrostructure). Knowledge use plays an important role in these processes. We are not only concerned with text representations per se (the textbase), but also with representing what the text tells a reader about certain aspects of the world (situation models). A series of experiments are proposed to explore these interdependent levels of representation: that of the actual verbal text, of the semantic content and structure of the text, and of the integrated textworld model. Of special interest is the role of inferences. Other experiments concern the dynamics of the comprehension process using priming procedures. Finally, problems of retrieval are considered, both knowledge retrieval and retrieval from episodic text memory. Explicit submodels of the general theory, describing specific experimental tasks and performances, will be developed. This work can provide a basis for studying language processes in normal as well as clinical populations.

Modern high-function computer systems are difficult to master and use. This problem is attacked in an interdisciplinary project that combines the development of innovative systems with a theor- etical investigation of the cognitive processes involved in com- puter use. The goal is to develop design principles for systems that are radically easier to understand, and hence to use, than any now in existence. Rooted in a comprehension-centered theory of human-computer interaction, the principles are to be embodied in concrete applications, which will serve as testbeds for both principles and the theory they reflect. This project promises a fundamental contribution to the art and science of augmenting human intellectual productivity. The degree to which it integrates exploratory system design with fundamental research on user-system interaction is unique. If successful, it could yield theoretical results of unprecedented explanatory power and design principles of wide applicabilty to interactive systems.

9314467 Kintsch This three-year award supports U.S.-France cooperative research in information sciences between Walter Kintsch of the University of Colorado and Guy Denhiere of the University of Paris, VIII. The investigators specialize in systems that mediate between computers and humans, specifically, in natural language understanding. The objective of their research is to develop a new computer architecture for cognition suited to the study of comprehension processes. Their proposed model is a hybrid that combines a production system based on weak rules with a constraint process that stimulates the formation of mental representations. Weak rules, for example, are word identification rules that lead to more than one possible meaning of a word. U.S. and French complementary expertise in theory and experiment will advance this effort. A joint experimental program to test the new model through computer simulations is proposed. This project will advance understanding in natural language processing. ***

Science WRITE, a computer-supported learning environment, will offer an innovative solution to the challenging problem of helping undergraduate science students to write well, by directly facilitating the complex cognitive processes that are critical for success. The system's educational foundation is a technique called procedural facilitation, which fosters independence in high-level thinking as students develop from novice to expert writers. We will deliver Science WRITE through World Wide Web technology because it is uniquely suited for accomplishing these learning goals. We will design, write, program, test, refine, and disseminate 10 interactive modules containing an array of learning activities that will help students develop their own independent processes for planning and revising scientific documents. We will integrate these modules with each other, with a document annotation system that we recently developed, with collaborative and knowledge-construction tools, and with links to other Web resources for disciplinary and discourse knowledge. The system will be designed for use by the nation's undergraduate students who have scientific writing projects in kinesiology. Science WRITE will be the first comprehensive resource (in print or on line) that is grounded in cognitive research and that supports undergraduate science students through all the processes involved in planning and revising papers. In addition, the system will be designed as a research laboratory in which we can continue to investigate and refine strategies for helping students learn to write in scientific disciplines. Science WRITE will demonstrate the powerful potential of this educational approach to improve the quality of undergraduate writing and thinking.

In this project a program of individualized, computer-aided reading instruction will be implemented and assessed. The computer-aided reading instruction is based on two classes of tools - one based on speech and animation technology and the other on language comprehension technology. An example of the former tool is Vocabulary Tutor which focuses on decoding practices. It includes a 3D animated talking head synchronized with recorder or synthesized speech, paired with illustrations and printed words that can be entered by the teacher or the child. Children look at the images, listen to the words, see their spelling, pronounce them, and receive feedback. An example of the second kind is Summary Street. Summary Street is based on a statistical theory of meaning, latent semantic analysis. Children write summaries, Summary Street automatically compares their summaries with the text they are summarizing and provides feedback about the content and the adequacy of their summaries. These tools are classroom tested. This project extends the foundation of technology and application development to a comprehensive reading program called Colorado Literacy Tutor that will be tested throughout the state of Colorado after testing in the Boulder School district.

EIA-0121201
Kintsch, Walter
University of Colorado Boulder

ITR/PE: Latent Semantic Analysis: Theory and Technology


This research program is designed to motivate, create and evaluate a new theory and methodology of learning, in which computer analysis of writing and speech is used to teach students to process and comprehend information more effectively. This work will inform educators on instructional and assessment strategies for improved reading comprehension to a degree not yet present in schools today. The methodology will teach students to read more effectively and to comprehend and learn more from what they read. Since the methodology can be applied to any text, it can be incorporated into reading programs to improve reading achievement, and be used as an effective tool to improve achievement in science and mathematics.

The work is intended to achieve both theoretical and research breakthroughs in four areas: (1) extend LSA to incorporate syntactic as well as semantic information; (2) extend the power and scope of LSA by advancing its conceptual and mathematical framework to include other psychological process models that handle phenomena such as metaphor and causal inferences; and (3) extend LSA to process transcriptions of natural continuous speech, enabling LSA to be used to teach comprehension of both read and spoken text by children who cannot write or type well enough to produce textual responses; and (4) demonstrate that these theoretical advances improve comprehension of speech and text by students in different grades, ethic backgrounds, and in different subjects.