Elaine Rich

Users are individuals

It has long been recognized that in order to build a good system in which a person and a machine cooperate to perform a task it is important to take into account some significant characteristics of people. These characteristics are used to build some kind of a “user model”. Traditionally, the model that is built is a model of a canonical (or typical) user. But often individual users vary so much that a model of a canonical user is insufficient. Instead, models of individual users are necessary. This article presents some examples of situations in which individual user models are important. It also presents some techniques that make the construction and use of such models possible. These techniques all reflect a desire to place most of the burden of constructing the models on the system, rather than on the user. This leads to the development of models that are collections of good guesses about the user. Thus some kind of probabilistic reasoning is necessary. And as the models are being used to guide the underlying system, they must also be monitored and updated as suggested by the interactions between the user and the system. The performance of one system that uses some of these techniques is discussed.

Stereotypes and User Modeling

A stereotype represents a collection of attributes that often co-occur in people. Stereotypes can play an important role in a user modeling system because they enable the system to make a large number of plausible inferences on the basis of a substantially smaller number of observations. These inferences must, however, be treated as defaults, which can be overridden by specific observations. Thus any stereotype-based user-modeling system must include techniques for nonmonotonic reasoning. This chapter will discuss the role that stereotypes can play in a user modeling system and it will outline specific techniques that can be used to implement stereotype-based reasoning in such systems.

Knowledge and natural language processing

KBNL is a knowledge-based natural language processing system that is novel in several ways, including the clean separation it enforces between linguistic knowledge and world knowledge, and its use of knowledge to aid in lexical acquisition. Applications of KBNL include intelligent interfaces, text retrieval, and machine translation.

An Architecture for Anaphora Resolution

In this paper, we describe the pronominal anaphora resolution module of Lucy, a portable English understanding system. The design of this module was motivated by the observation that, although there exist many theories of anaphora resolution, no one of these theories is complete. Thus we have implemented a blackboard-like architecture in which individual partial theories can be encoded as separate modules that can interact to propose candidate antecedents and to evaluate each others proposals.

Sex-Related Differences in Colour Vocabulary

This paper describes an experiment designed to test the hypothesis that women have larger colour vocabularies than men. The results indicate that they do. The results also indicate that, in at least one social class, younger men have larger colour vocabularies than do older men. No such difference exists for women. However, a group of Catholic nuns did score lower than the rest of the women but still higher than the men.

An introduction to HITS: Human Interface Tool Suite

Computers are the most plastic medium yet invented for the representation and propagation of information. They can mimic the behaviors of other information media and manifest behaviors not possible in any other medium. They provide us with interactive representational media of potentially revolutionary consequence. To realize the communicative potential of these new computationally-based media we must understand how to employ them to construct collaborative multimedia interfaces to high-functionality systems. In this report, we introduce the Human Interface Tool Suite, an integrated set of tools for the construction of collaborative multimedia interfaces. Rather than documenting HITS as a user or reference manual would, we explain the ideas that motivate the HITS research programme. Microelectronics and Computer Technology Corporation Advanced Computer Architecture Human Interface Laboratory 3500 West Balcones Center Drive Austin, Texas 78720 (512) 343-0978 Copyright c 1988 Microelectronics and Computer Technology Corporation All rights reserved. Shareholders and associates of MCC may reproduce and distribute these materials for internal purposes by retaining MCCs copyright notice, proprietary legends, and markings on all complete and partial copies. NT E R FACE B y M C C AN INTRODUCTION TO HITS: HUMAN INTERFACE TOOL SUITE James Hollan, Elaine Rich, William Hill, David Wroblewski, Wayne Wilner, Kent Wittenburg, Jonathan Grudln, and Members of the Human Interface Laboratory

Programs as data for their help systems

The goal of this research is to develop ways of representing the knowledge available to a help system in such a way that the system can actually reason with the knowledge rather than being restricted to simply retrieving and presenting stored answers to a restricted and anticipated class of questions. One kind of information that is useful to such an intelligent help system is knowledge of how the underlying system operates. This knowledge is contained in the code for the system. By exploiting system code as part of the help database, many problems of inconsistency between programs and their documentation can be avoided. In our initial investigations of this problem, we are representing the system code as a set of productions that are easier to manipulate than is code in most standard languages. As we develop techniques for answering questions by reasoning with knowledge about the system, we become increasingly able to answer the growing variety of questions that will occur as the language interface to a help system becomes more flexible.

Reversible Machine Translation: What to Do When the Languages Don’t Match Up

In this paper we deal with issues that face an interlingua-based, reversible machine translation system when the literal meaning of the source text is not identical to the literal meaning of the natural target translation. We first outline the problem and show, using examples, some common situations in which the problem arises. We then describe a theoretical framework that lets us begin to define what it means to solve the problem. Finally, we present an algorithm for lexical choice that handles many of these cases and that relies exclusively on reversible, monolingual linguistic descriptions and a language-independent domain knowledge base.

Knowledge Bases and Tools for Building Integrated Multimedia Intelligent Interfaces

A multimedia interface is one that supports more than one medium through which users and application programs can communicate. For example, such an interface might support graphics, gestures, natural language, menus, sketching, and video. An intelligent interface is one that exploits knowledge about the task, the application, the interface, and the user in ways that help the user accomplish the task effectively. This includes interpreting ambiguous inputs correctly in the current context, phrasing outputs appropriately for the current user, and providing advice on efficient ways to use the system to accomplish the users goals. An integrated interface is one in which events and objects in one part of the interface are accessible in the other parts so that a single task can be split across interface components as appropriate. In this paper, we present an architecture for an integrated set of tools now under development, called Human Interface Tools, or HITS. These tools are intended to support both the construction and the run-time execution of integrated multimedia intelligent interfaces. This paper describes our design of HITS, with some examples from parts of the system that have already been implemented. We will not attempt to articulate clearly the line between the current working toolset and our vision of a more complete system but will summarize the current state of the system at the close of the paper.

LUKE: AN EXPERIMENT IN THE EARLY INTEGRATION OF NATURAL LANGUAGE PROCESSING

Luke is a knowledge editor designed to support two tasks; the first is editing the classes and relations in a knowledge base. The second is editing and maintaining the semantic mapping knowledge neccesary to allow a natural language interface to understand sentences with respect to that knowledge base. In order to emphasize design decisions shared between the two tasks, Luke provides facilities to concurrently debug the application and the natural language interface. Luke also makes natural language available in its own user interface. This makes it possible for a knowledge base builder to exploit natural language both as a way of locating desired concepts within the knowledge base and as a a way of doing consistency checking on the knowledge base as it is being built.