<?xml version="1.0" standalone="yes"?>
<Paper uid="J93-4004">
  <Title>Planning Text for Advisory Dialogues: Capturing Intentional and Rhetorical Information</Title>
  <Section position="4" start_page="655" end_page="656" type="intro">
    <SectionTitle>
SYSTEM What characteristics of the program would you like to enhance? \[1\]
USER Readability and maintainability. \[2\]
</SectionTitle>
    <Paragraph position="0"> SYSTEM You should replace (8ETO X 1) with (SETF X 1). SETQ can only be used \[3\] to assign a value to a simple-variable. In contrast, SETF can be used to assign a value to any generalized-variable. A generalized-variable is a storage location that can be named by any access function.</Paragraph>
    <Paragraph position="1"> Figure 3 Partial dialogue.</Paragraph>
    <Paragraph position="2"> jects; and Hovy (1988) uses pragmatic and stylistic information to produce different accounts of the same incident.</Paragraph>
    <Paragraph position="3"> 3.2.1 Inadequacies of Schemata for Advisory Dialogue. Like others, we found that schemata were not sufficient to handle the issues we wished to investigate. When we attempted to use schemata for our purposes, two main problems arose. First, schemata lack an explicit representation of the intentional structure of the text being produced, and therefore are missing the information needed to recover from explanatory failures. Second, we found that schemata are too rigid to handle certain of the opportunistic phenomena we observed in naturally occurring dialogues. We discuss these two problems in more detail.</Paragraph>
    <Paragraph position="4"> Lack of Intentional Structure. As we have seen, schemata encode standard patterns of discourse structure. However, they do not include an explicit representation of the effects that individual components of a schema are intended to have on the hearer, or of how these intentions relate to one another or to the rhetorical structure of the text. This presents a serious problem for a system that must participate in a dialogue where users can ask follow-up questions like the ones we saw in the sample dialogue of Figure 1. If a system does not keep a record of the intentions behind its utterances, it cannot determine what went wrong when the user indicates that an explanation was not completely understood, nor provide an alternative explanation to correct the problem.</Paragraph>
    <Paragraph position="5"> To allow a system to handle follow-up questions that may arise if the user does not fully understand an explanation, a generation facility must be able to determine what portion of the text failed to achieve its intended purpose. If the generation system only knows the top-level communicative goal that was being achieved by the text (e.g., to make the hearer know a concept, or to make the hearer want to perform an action), and not what effect the individual parts of the text were intended to have on the hearer or how they fit together to achieve this top-level goal, its only recourse is to use a different strategy to achieve the top-level goal. It is not able to re-explain or clarify any part of the explanation.</Paragraph>
    <Paragraph position="6"> We illustrate this important point by working through an example taken from an actual dialogue with a system called the Program Enhancement Advisor (PEA) (Neches, Swartout, and Moore 1985). (Note that, for precisely the reasons we describe in this paper, PEA does not employ schemata to generate its utterances. We describe PEA's text planner in Section 5.) As shown in Figure 3, PEA begins its interaction with the user by asking what characteristics of the user's program are to be enhanced and then suggests changes that will improve these aspects of the program. Now consider what a schema that could produce the system's utterance in turn 3 of the sample dialogue in Figure 3 would look like. One schema that would suffice, which we have called the Recommend-Replacement Schema, is shown instantiated in Figure 4.  Johanna D. Moore and C6cile L. Paris Planning Text for Advisory Dialogs System's Utterance (1) You should replace (SETQ X 1) with (SETF X 1). (2) SETQ can only be used to assign a value to a simple-variable. (3) SETF can be used to assign a value to any generalized-variable. (4) A generalized-variable is a storage location that can be named by any access function.</Paragraph>
    <Paragraph position="7">  Hypothetical schema representation of system's utterance.</Paragraph>
    <Paragraph position="8"> This schema begins with a RECOMMENDATION of a replacement act, followed by a COMPARE &amp; CONTRAST predicate that highlights the important difference(s) between the replacee and the replacer. 2 Instead of using a simple predicate, we instantiate the COMPARE &amp; CONTRAST predicate using a schema that expands into two ATTRIBUTIVE predicates and an IDENTIFICATION predicate that defines a term introduced in the second ATTRIBUTIVE.</Paragraph>
    <Paragraph position="9"> Note that this schema indicates what to do when, i.e., recommend the action and contrast the replacee with replacer, but it does not say why this information is being presented. For example, the schema does not indicate that, by contrasting SETQ with SETF, the speaker is trying to persuade the hearer to do the replace act. Nor does it indicate that the text produced by the IDENTIFICATION predicate appears because the speaker is trying to make the hearer know about the concept generalized-variable. In addition, the relationships between these intentions are not represented. To make clear what is missing, we have represented in Figure 5 the intentional structure of this text using Grosz and Sidner's (1986) notions of dominance and satisfaction-precedence. In Grosz and Sidner's theory (1986, p. 179), if an action that satisfies one intention, h, is intended to provide part of the satisfaction of another intention,/2, then/2 dominates h. h satisfaction-precedes 12 whenever h must be satisfied before/2. The representation shown in Figure 5 makes it clear that the expert system's (E) top-level intention (I0) is to get the user (U) to intend to replace (SETQ X 1) with (SETF X 1), and this intention dominates E's intentions to recommend this act (/1) and to persuade U to perform it (/2). In addition, for this schema, the recommendation (h) must be satisfied before the persuade (/2) is attempted.</Paragraph>
    <Paragraph position="10"> A schema can be viewed as the result of a &amp;quot;compilation&amp;quot; process where the rationale for all of the steps in the process has been compiled out. What remains is the top-level communicative goal that invoked the schema (in this case something like Get the user to adopt the goal of replacing SETQ with SETF), and the sequence of actions (i.e. instantiated</Paragraph>
  </Section>
class="xml-element"></Paper>