<?xml version="1.0" standalone="yes"?>
<Paper uid="J89-4002">
  <Title>NATURAL LANGUAGE GENERATION FROM PLANS</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Planning is a central concept in Artificial Intelligence, and the state of the art in planning systems allows quite complex plans to be produced with very little human guidance. If these plans are to be for human consumption, they must be explained in a way that is comprehensible to a human being. There is thus a practical reason for considering ways of generating natural language from plans. There are also theoretical reasons why plans are a good domain for studying natural language generation. Although there may be a great deal of material in a given plan, there is a kind of consensus among planning researchers on what sort of information a plan is likely to contain. Thus it is possible that interesting general principles about producing explanations of plans can be formulated, independently of the domains in which the plans are produced. This property, of providing a relatively formally defined and yet domain-independent input, makes plans very attractive from a natural language generation point of view.</Paragraph>
    <Paragraph position="1"> This paper discusses a system that accepts a plan structure of the sort generated by AI planning programs and produces natural language text explaining how to execute the plan. Our objective in building this system has been to develop a clear model of a possible architecture for a language generation system that makes use of simple, well-understood, and restricted computational techniques. We feel that too much of the work in this area has been characterized by the use of arbitrary procedures, which often do not provide a clear basis for future work. We believe that by providing a simple yet nontrivial account of language generation, we can contribute at least by providing a &amp;quot;straw man&amp;quot; with known limitations, with respect to which other work can be compared.</Paragraph>
    <Paragraph position="2"> Describing plans represents in many ways an obvious Copyright 1989 by the Association for Computational Linguistics. Permission to copy without fee all or part of this material is granted provided that the copies are not made for direct commercial advantage and the CL reference and this copyright notice are included on the first page. To copy otherwise, or to republish, requires a fee and/or specific permission. 0362-613 X/89/010233-249-$03. O0  application of natural language generation, and our approach has been to tackle this problem in a fairly straightforward way, informed by the state of the art as we perceive it. The results from our system are promising, but our texts lack much of the smoothness of human-generated explanations. An analysis of the reasons behind some of the system's failures points to a number of deep problems concerning the connection between AI plans and natural language explanations.</Paragraph>
    <Paragraph position="3"> In the next section we briefly introduce the inputs and structure of the language generation system. We then run through the parts of the system by showing a worked example. The core of this paper concerns the mapping from plans to messages, which can be thought of as abstract descriptions of natural language discourses. We describe our repertoire of messages, how plan structures are mapped onto messages, and how messages are simplified. Finally we look at further examples of the system's output and analyze some of its failures and successes.</Paragraph>
  </Section>
class="xml-element"></Paper>