<?xml version="1.0" standalone="yes"?>
<Paper uid="J88-3002">
  <Title>MODELING THE USER IN NATURAL LANGUAGE SYSTEMS</Title>
  <Section position="32" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 CONCLUSION
</SectionTitle>
    <Paragraph position="0"> The ability to interact with people in an easy and natural manner is the promise natural language interfaces hold for computer systems. To realize this promise, systems need to acquire and use various kinds of information about the people with whom they are interacting. That is, they need models of their users.</Paragraph>
    <Paragraph position="1"> Sophisticated user models can serve many important functions in natural language systems: they can be used to tailor the interaction to an individual user, to increase the system's cooperativeness, and to correct or even prevent misconceptions by the user. This paper has made several general points about the role of user models in question answering systems.</Paragraph>
    <Paragraph position="2"> * What constitutes a user model is a matter of some debate. The view taken in this paper is that a user model is an explicit source of knowledge containing the beliefs and assumptions the system holds about the user.</Paragraph>
    <Paragraph position="3"> * User models must hold many diverse types of information. Natural language systems need to know about the user's goals and plans, capabilities, attitudes, and beliefs.</Paragraph>
    <Paragraph position="4"> * User models can be classified along various dimensions. In general terms, these dimensions characterize the agents being modeled, how the model changes with time, and how it is used.</Paragraph>
    <Paragraph position="5"> * The acquisition of information about the user is a central problem that must be faced. The process can be explicit, implicit, or a mixture of the two. The techniques used for acquisition depend on the kind of information.</Paragraph>
    <Paragraph position="6"> * Environmental issues, such as how the model will be used, place added constraints on the type of user model that may be employed in a particular implementation. null To date, most of the work involving the kind of user models discussed in this paper is in an early research stage. This research typically focuses on just one aspect of the overall user modeling problem, such as plan recognition or modeling multiple agents. There is still a great deal of research to be done in these individual areas. Goal recognition and modeling is central to many AI problems and has not yet been adequately handled in any real systems. Many of the ways that a user model can improve natural language interaction have not yet been explored. In the context of generation systems, for example, no existing systems use their knowledge of the user as a factor in the lexical choice problem.</Paragraph>
    <Paragraph position="7"> Addressing individual problems in user modeling and looking at particular applications where a user model can help have been appropriate research strategies in early investigations. Ultimately, however, user modeling must be addressed from a more global point of view. A rich, interactive system will need to model many things about many human agents. This information can form a central knowledge base for reasoning about agents in many contexts.</Paragraph>
    <Paragraph position="8"> The notion of a central user modeling facility has motivated work on a general user modeling system or general user modeling module (Finin and Drager 1986, Kass 1987a, Kass and Finin 1987c). A general user modeling system would provide an environment for building systems that used a user model, including various facilities for maintaining and updating user models. A general user modeling module is an independent component of a larger system that provides information about the user to other modules, much like a data base or knowledge base. The interface to the general user modeling module is well-defined, enabling it to be used in a variety of systems with little or no customization.</Paragraph>
    <Paragraph position="9"> Future work in user modeling for natural language systems should focus in two directions: establishing how user models should be used in systems that communicate in natural language, and determining how user models can be built more effectively. Many authors have emphasized the need for user models in certain contexts, or have demonstrated that the availability of user model information can improve the behavior of a system. This work needs to be extended to identify what information applications will expect a user model to have, how that information should be provided to the application, and when the information needs to be available. Answers to these questions will help define the services that a user modeling component must provide.</Paragraph>
    <Paragraph position="10"> The second focus of research should be on building user models. This work could progress in two ways.</Paragraph>
    <Paragraph position="11"> First, the task of explicitly building user models (such as building stereotypes) could be made easier. Research in this area seems to parallel efforts to find better ways to acquire knowledge for knowledge bases from experts.</Paragraph>
    <Paragraph position="12"> However, if general user modeling modules that can Computational Linguistics, Volume 14, Number 3, September 1988 19 Kass and Finin Modeling the User in Natural Language Systems function in diverse systems are to be built, the focus must be placed on the second approach: implicit user model acquisition. In this regard, a user modeling module could be general either with respect to the underlying domain or to the type of interaction. At this time, domain generality seems both a useful and practical goal. The work described in Kass (1987a) and Kass and Finin (1987c) is a beginning in this area, presenting a set of domain general user model acquisition rules for cooperative consultation situations.</Paragraph>
    <Paragraph position="13"> User modeling is not an easy task. Effective user modeling requires sophisticated knowledge representation, acquisition, and reasoning abilities--no wonder user modeling is such a new field. On the other hand, advances in any of these areas should provide immediate benefits to user modeling. Thus progress in some of the fundamental areas of AI can result in progress in user modeling as well.</Paragraph>
  </Section>
class="xml-element"></Paper>