<?xml version="1.0" standalone="yes"?>
<Paper uid="P85-1028">
  <Title>Explana~..: 3tructures in XSEL</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> Expert systems provide a rich testbed from which to develop and test techniques for natural language processing. These systems capture the knowledge needed to solve real-world problems in their respective domains, and that knowledge can and should be exploited for testing computational procedures for natural language processing. Parsing. semantic ,nterpretation, dialog monitoring, discourse organization, and text gef,eration are just a few of the language processinq problems that might takeadvantage of the pre.structured semantic knowledge of an expert system. In particular, the need for explanation generation facilities for expert systems provides an opportunity to explore the relationships between the underlying knowleqge structures needed for automated reasoning and those needed for natural language processing. One such exploration was the development of an explanation generator for XSEL, which is an expert system that hellos a salesperson in producing a purchase order for a computer system\[10\]. This pager describes a technique called &amp;quot;link-dependent message generation&amp;quot; that forms the basis for explanation generation in XSEL.</Paragraph>
    <Paragraph position="1"> 1.1. Overview of XSEL Briefly, the function of the XSEL system is to assist a salesperson in configuring a custom-tailored purchase order for a Digital Equipment Corporation VAX computer system. XSEL works with the salesperson tO elicit the functional computing requirements of the individual customer, and then goes on to select the components that best fit those requirements. The output of an XSEL session is a purchase order consisting of a list of line-items that specify hardware and software components.</Paragraph>
    <Paragraph position="2"> There ~re two main phases to XSEL's processincj, a fact gathering phase and a component select=on phase. During the fact gathering phase XSEL carries on an interactive dialog with the salesperson to elicit values for facts that determine the customer's functional computing requirements. These might include requirements for total disk space, percent of removable disk storage, number of terminals, lines-per.minute of printing, etc. Natural language processing during the fact gathering dialog is minimal: XSEL displays menues and pre-formutated queries and accepts one- or two-word answers from the user.</Paragraph>
    <Paragraph position="3"> Once enough facts have been collected XSEL begins a silent phase of processing. During this phase a set of candidate components that satisfy the customer's basic requirements is retrieved from the DEC parts database. Within each class of component, i.e., processor, disk, terminal, etc., candidates are ranked according to their score on a~q evaluation function that measures the degree to which a candidate satisfies the customer's weighted functional requirements. The candidate with the highest score is selected and placed on the purchase order.</Paragraph>
    <Paragraph position="4"> The most important knowledge structure used by XSEL during the fact gathering I~ase is a fact. A fact is simply a list of attribute-value pairs that represent knowledge about one of the customer's functional computing requirements. Figure 1-1 depicts a sample facL</Paragraph>
  </Section>
class="xml-element"></Paper>