IRList Digest Monday, 1 December 1986 Volume 2 : Issue 63 Today's Topics: Abstracts - NSF IST Awards for Fiscal Year 1986 - Part 1 of 5 News addresses are ARPANET: fox%vt@csnet-relay.arpa BITNET: foxea@vtvax3.bitnet CSNET: fox@vt UUCPNET: seismo!vtisr1!irlistrq ---------------------------------------------------------------------- Date: Fri, 21 Nov 86 18:38:14 est From: vtopus!fox (Ed Fox) Subject: Information on NSF awards, sent by J. Deken at NSF Fiscal Year 1986 Research Projects Funded by the Information Science Program (now Knowledge and Database Systems Program) Part 1 of 5 IST-8504726 $42,196 - 12 mos. James F. Allen University of Rochester Plan-Based Approaches to Extended Dialogues - - - This research project investigates several basic problems that arise in constructing a computer system that can carry on a natural language dialogue with a user. The work is based on previously developed models of discourse based upon speech acts and plan recognition. Extensions to the dialogue model include facilities to handle both topic changes and clarification and correction of misconceptions. Extensions to the knowledge representation system include improvements to the theory of actions and plans and improved methods for reasoning about the beliefs of different agents. More theoretical work includes development of a theory of plan reasoning that integrates both plan construction and plan recognition. _____ BNS-8518675 $40,000 - 12 mos. James A. Anderson Brown University Cognitive Applications of Matrix Memory Models - - - There has been a good deal of recent interest in what have been called "neural models" or "connectionist models" as ways of explaining the organization and computational abilities of the nervous system. This research will continue study of the properties of a simple connectionist model that has been developed over the past decade, most recently under BNS- 14728. The particular version of a connectionist model that is used is called the "BSB" model, and technically, is a non- linear dynamical system with strengths of interconnections specified by a conjunctional learning rule (a "Hebbian" rule). It is sufficiently simple so it can be analyzed mathematically in considerable detail and is easy to simulate on a digital computer. The model will be used to understand how humans form concepts. The BSB model has powerful "concept" forming abilities, and demonstrates these properties both theoretically and in computer simulation. The BSB model is also studied in its role as a functioning memory, for example, to see if it can be used to store a large database of related information. _____ IST-8511531 $19,097 - 12 mos. Robert Berwick Massachusetts Institute of Technology Learnability and Parsability - - - In this collaborative project, Learnability and Parsability, the two principal investigators, Drs. Berwick and Wexler, are studying the relationship between language learning and language understanding. The issue is motivated by the existence of already known properties shared by grammars which are learnable and grammars which are parsable. (Grammars which allow a language to be analyzed into a unique underlying representation are said to be parsable.) A common framework is being developed for representing grammars. The common framework allows a comparison of learnability and parsability through the use of formal proofs. Additionally, child and adult parsers are being compared formally. The research furthers understanding of formal theories of parsability and learnability and the relation between the two. These are fundamental issues for fields concerned with syntactic issues in language. The work ultimately is aimed at the development of automated language understanding systems which can both learn and understand natural language. _____ DCR-8603231 $27,000 - 12 mos. Alan W. Biermann Duke University Dialog Processing for Voice Interactive Problem Solving - - - For several years now, Professor Alan W. Biermann at Duke University and Professor Robert D. Rodman at North Carolina State have collaborated on research efforts directed at voice interactive computing systems. They are studying the dynamics of human-machine voice interactive problem solving with the goal of constructing a fast and efficient voice interactive computer system. The voice interactive dialogue machine will allow the user to display data on a computer terminal and manipulate it using voice commands and touch sensitive imputs to the screen. Each command will result in immediate action visible on the screen allowing the user to confirm that a proper response has occured. A natural language computer (NLC) was used and considerable experience has been gained with subjects using the voice equipment. Among the accomplishments has been the integrationof a voice and touch facility into the NLC system and the modification of the system for the idiosyncracies of voice and touch input. The initial implementation of the voice interactive dialogue machine will address the equipment repair problem where the human provides the eyes and mechanical skills to work on a failing device and the machine has the appropriate function- al diagrams and debugging algorithms. The major issues to be addressed concern finding characterizations for typical human-machine task oriented dialogues, developing formalisms for understanding them and for implementing them in a computer, and measuring the performance of human-machine cooperative efforts when utilizing such a grammar-driven voice conversational machine. _____ IST-8612177 $9,747 - 12 mos. Jeffrey Bonar University of Pittsburgh Partial Support for Third International Conference on Artificial Intelligence and Education, Pittsburgh, PA, May 1987 - - - This project is a workshop for researchers from areas supported by Information Science. The workshop explores issues in artificial intelligence, cognitive psychology, and education. At the meeting, artifical intelligence methodology is applied to issues of student performance, learning, diagnosis of student errors and instruction. This goal is achieved by considering strategies for teaching, research on learning, methods for inferring cognitive states from behavior, architectures for tutoring systems and interface design. The broad goal of the workshop is to facilitate interaction between intelligent tutoring systems researchers and educational researchers on currently unaddressed issues. The results of the workshop will be produced as conference proceedings. The workshop will benefit the study of Information Science by bringing together top researchers and allowing them to look at newly emerging issues and currently unanswered questions. ____ IST-8604923 $64,660 - 12 mos. Alan H. Borning University of Washington Automatic Generation of Interactive Displays - - - This project seeks a way to construct user-system interfaces with less effort than is currently required and at the same time to achieve a higher degree of uniformity among interfaces than is currently found. The approach taken is to relate a source object (such as some information in an application program) and view object (such as a graphical image on the computer's screen) by means of a "filter". By this term is meant a program which can satisfy some constraint on the two objects is relates. Filters are to be bidirectional: if the source changes, the view must be updated; if the view is edited, the system must find a corresponding change to the source to keep the constraint satisfied. An important property of filters is that they can be composed in a variety of ways, so that new filters can be defined using other filters as parts. One objective of the research is a formal theory which will allow transformations between equivalent systems of filters. The results are to be tested by implementing a prototype system for generating user interfaces. The research is carried out in collaboration with David Maier, of the Oregon Graduate Center, and Ralph London, of Tektronix, Inc. _____ IST-8643739 $121,074 - 12 mos. Bruce C. Buchanan Stanford University Information Structure and Use in Knowledge-Based Expert Systems - - - Expert systems and other knowledge-based computer models provide an important application of computer models as intelligent information systems. Among the problems confronting the design of such systems is the flow of knowledge between the expert and the builders of the models. In addition, a transfer of information problem exists between the users and the systems themselves. This research will test and analyze models developed by the research team for the improved transfer of information in the above situations. Specifically, the research proposes to explore; (1) a model for constructing knowledge-based programs by transferring expertise from a person to a program interactively and iteratively, and (2) a model for using such programs in a "critiquing" mode, as opposed to a "consulting" mode. At present one of the great bottlenecks in the construction of expert systems is the process of transferring information between the experts and the system. A great improve- ment in the expert system building process will be gained if the PIs are successful in the construction of a model to expediate and automate this procedure. _____ IST-8607303 $10,000 - 12 mos. Kathleen M. Carley Carnegie-Mellon University Knowledge Acquisition as a Social Phenomenon - - - The research develops a model of knowledge acquisition in an environment where information is incomplete and uncertain. The model explores the relationship between the individual's private cognitive world, in terms of knowledge acquisition, and the social world, in terms of social position, interactive patterns and shared knowledge. In particular, the model examines predictions about both the social processing of information and the related shifts in social structure, information flow and storage. The theoretical underpinnings for the model are based upon prior research in sociology, artificial intelligence, cognitive psychology and communication research. The research methods are mathematical modeling, simulation and model validation with field data. The research is relevant to an understanding of the diffusion of new ideas and information between and within groups, effective distribution of information in a multi-person, distributed decision making environment, and contributes to an understanding of the necessary and sufficient conditions for effective communication. _____ IST-8515005 $75,900 - 12 mos. Eugene Charniak Brown University A Single-Semantic-Process Theory of Parsing - - - Current automated text understanding systems have numerous components (modules with identifiably different functions). All of the components use semantic and pragmatic information in accomplishing tasks such as word-sense disambiguation. However, each makes only limited use of the semantic and pragmatic imformation available. This researcher implements an understanding system with only two components through a technique called "marker passing". In "marker passing" pieces of semantic information, i.e., markers, are passed throughout the system, making full pragmatic and semantic information available to all components. This advance in implementation stems from the underlying theoretical perspective of the researcher; all components of a system should take full, rather than partial, advantage of semantic and pragmatic information. The system, as a result, makes the larger number of components usually found in such systems unnecessary. The first component of this system generates possible meanings for a given sentence and the second component then chooses among them. The research offers a more parsimonious theory of natural language comprehension and implements it as a natural language understanding system. The ultimate aim of a system which implements this type of theory is the automated understanding of ambiguous statements. _____ IST-8644629 $60,449 - 12 mos. Eugene Charniak Brown University An Approach to Abductive Inference in Artificial Intelligence Systems - - - The ability to explain observed events is an important capability in artificial intelligence systems for many different kinds of applications. In order to accomplish this, a system must be able to postulate a possible cause and then determine whether or not it is reasonable. Looking for possible explanations involves matching observed actions with appropriate parts of the system's knowledge base. In this research the knowledge base is assumed to be structured into script or plans which are formally represented. A major goal is to determine appropriate and efficient algorithms, based on this formal representation for determining if possible explanations are consistent with the knowledge base of the system. The research focuses on the kinds of problems that arise in the analysis of short narrative texts. This research will contribute to the understanding of intelligent computer-based systems. _____ IST-8608362 $172,9212 - 12 mos. Richard Cullingford Georgia Institute of Technology Robust Interaction and Natural Problem-Solving in Advice-Giving Systems - - - The purpose of this research is to develop an intelligent advisory system. The system functions as an aid during problem solving, it both volunteers and takes advice and is able to accept ungrammatical input. Several research issues underlie the development of this problem solving aid: memory and reasoning are modeled and used in the advisory process; natural language capabilities are developed; and the reasoning and language elements of the system are integrated. Artificial intelligence advisory systems need to be able to interact with humans in a flexible fashion. In order to be flexible, these systems need to be able to understand ungrammatical input and to be able to take, as well as give, advice. This research deals with these issues and contributes to the develop- ment of more sophisticated computer systems. _____ IST-8506706 $60,942 - 12 mos. Donald Dearholt New Mexico State University Properties of Networks Derived from Proximities - - - The purpose of this project is to develop a way of using networks to represent the relationships among sets of related concepts. An algorithm for finding the salient relationships is developed. It is based on human proximity judgments of the conceptual closeness of the members of a set of concepts. The algorithm is implemented as a computer system. The system produces a network structure which makes the relationships between concepts salient. Experiments are con- ducted to test that the relationships identified by the algorithm are actually judged salient by humans. The project will allow large databases to be used more fully and more flexibly. For example, the system can supply users of a database with keywords which are either prototypical or atypical for a category in the database. _____ IST-8518706 $37,243 - 12 mos. Andre de Korvin Indiana University Foundation Modeling Goal Uncertainty and Goal Shaping in a Generalized Information System - - - This project is a study of the effectiveness of information in a generalized information system, which is a formal model of information flow and decision analysis. Mathematical analysis is used to determine the effectiveness of feedback processes based on utility functions. The results of previous work illuminate how information should be selected, when stopping points can be reached, and how long information must be gathered before convergence is achieved. This project builds upon and expands earlier findings to cover the case where the goals of the decision manager are not known and fixed. In the expanded and more realistic model, goals themselves are uncertain and subject to change by the information gathering process. ------------------------------ END OF IRList Digest ********************