Christopher K. Riesbeck

You Can't Miss It!: Judging the Clarity of Directions

Recently there has been some very interesting AI work done on the representation of knowledge of large scale maps and their use in the construction of routes through some space (Kuipers, 1978; McDermott, 1980). The stress in this work has been on how spatial information is best represented and how reasoning is performed with the assumed representation. In this paper I would like to address the natural language processing of texts giving directions and make a claim that during a casual first reading, there actually does not need to be much spatial reasoning going on at all. When I read a written set of directions, I am primarily interested whether the directions seem clear and sensible+not in constructing a map or program specifying all the turns, distances, and locales that will be involved. Certainly there are times when I have to make some kind of route or map structure, such as when I am the subject in a spatial reasoning experiment, or when I am lost and ask someone on the street for directions. But I do not need to work so hard when I am given a piece of paper with a set of directions, and I know that I will have this piece of paper with me when I actually make the trip. I know that I can always read the directions again to figure out what to do, when I actually set out on the trip.

Inference and Paraphrase by Computer

~SrRACT. The MARGIE system is a set of three programs that attempt to understand natural language. They are based on the Conceptual Dependency system for meaning representation. The analysis program maps sentences into conceptual structures. The memory program makes inferences from input conceptual structures. The generator codes conceptual structures back into natural language. Together the programs function as a paraphrase and inference system.

Knowledge reorganization and reasoning style

Abstract To study the learning of expertise, two closely related stages of expertise in economics reasoning are analyzed and modelled, and a mechanism for going from the first to the second is proposed. Both stages share the same basic concepts and generate plausible economie scenarios, but reasoning in the first stage oversimplifies by focussing on how the goals of a few actors are affected. Reasoning in the second stage produces better arguments by taking into account how all the relevant parts of the economy might be affected. The first stage is modelled by highly interconnected goal forests and very selective, story understanding search heuristics. The second stage is modelled with more explicit links between economie quantities and a more appropriate set of search heuristics. The learning mechanism is a failure-driven process that not only records better arguments as they are seen, but also records the failure of existing inference rules to find these arguments on their own. The collected failures are used to determine which search heuristics work best in which situations.

An expectation-driven production system for natural language understanding

ELI, the English language interpreter for the SAM story understanding system at Yale, is a model of language understanding using productions. Productions are useful because they are flexible, but this flexibility means more work has to be spent controlling and manipulating them. ELI limits the number of productions that have to be manipulated with expectations. Expectations are constraints generated by frame structures that have been built by previously executed productions. Only productions that satisfy existing expectations are used.

Tools for incremental development of educational software interfaces

In this paper we describe the evolution of an educational software tool designed to let non-programmers build content-rich learning environments. Version 1 was a wholly model-driven authoring environment, but was unpopular with authors as they were forced them to build up-front domain representations before prototyping their interfaces. Version 2 uses a GUI method of interface development while the model is developed incrementally and as needed. In this version, authors built less of a model overall, but were more satisfied with the results. This paper discusses the natures of the two approaches to model-building and how they are authored.

An incremental model for developing computer-based learning environments for problem-based learning

Problem-based learning (PBL) is a pedagogical strategy that centers learning activities around the investigation and development of solutions to complex and ill-structured authentic problems. It requires additional support and resources for students and instructors to use it in schools. Computer-based interactive learning environments have been used to provide students authentic and supportive settings for PBL. These systems, however, require significant up-front development effort before they can be put into use. In this paper, we describe an incremental model that allows instructors to author the learning environment during real use. In our model, an instructor is part of the feedback loop, complementing system feedback and collecting materials to be incorporated into the system. By working within the system, the instructor can observe detailed traces of student learning activities in the system and provide individualized coaching and critiquing. We describe our experience in developing Corrosion Investigator, a Web-based learning environment, as an exemplar of our model. We focus on how the system is designed to facilitate instructor involvement and support incremental authoring. We present empirical results showing materials collected through use and benefits to the students and teachers.

Making critiquing practical: incremental development of educational critiquing systems

Expert critiquing systems in education can support teachers in providing high quality individualized feedback to students. These systems, however, require significant development effort before they can be put into use. In this paper, we describe an incremental approach that facilitates the development of educational critiquing systems by integrating manual critiquing with critique authoring. As a result of the integration, the development of critiquing systems becomes an evolutionary process. We describe a system that we built, the Java Critiquer, as an exemplar of our model. Results from real-life usage of the system suggest benefits for supporting teachers in critiquing student code.

Designing Web-based interactive learning environments for problem-based learning

Problem-based learning is a pedagogical strategy that centers learning activities around the investigation and development of solutions to complex and ill-structured authentic problems. A number of difficulties occur when implementing such approaches in schools. To address these difficulties, a software tool called INDIE was built to create Web-based interactive learning environments where students can run simulated experiments, analyze test results, form rationales, and construct arguments to support or refute possible hypotheses. This paper focuses on how INDIE supports problem-based learning by creating an authentic environment that incorporates important aspects in real life, providing tools to help students perform problem-solving and receive coaching and critiquing, providing support for instructors to assess student understanding and provide feedback, and using an interface that allows open-ended inquiry and exploration. We present preliminary results showing Corrosion Investigator, a learning environment delivered by INDIE, to be beneficial for both the students and teacher in a class.

Authorable critiquing for intelligent educational systems

An important issue in intelligent interfaces is making them as authorable as non-intelligent interfaces. In this paper, we descriie Iudie, an authoring tool for intelligent interactive education and training environments, with particular emphasis on how authors create knowledge bases for critiquing student arguments. A central problem was providing authors with tools that supported the entire development process from mock-up to final product. Two key ideas are: (1) MVC-based event-action triggers to support a gradual migration from interface-based to modeldriven interactions (2) rule-based evidence assessment events.

Knowledge acquisition from simplified text

The problem of entering and integrating new knowledge into a logic-based knowledge base is substantial. Our solution is to provide a natural language interface, which reads simplified English, enabled by a knowledge-based memory-retrieval driven natural language understander. This paper presents a set of tools and interfaces for interacting with such a system, and a discussion of the underlying Reader system, the reading component of the Learning Reader project. The interfaces presented provide direct feedback about what portions of the text are understood, and what interpretations are being produced from it. In addition tools are presented to, among other things, provide example sentences, to facilitate users producing simplified English text suitable for the Reader.