<?xml version="1.0" standalone="yes"?>
<Paper uid="A88-1025">
  <Title>AUTOMATICALLY GENERATING NATURAL LANGUAGE REPORTS IN AN OFFICE ENVIRONMENT</Title>
  <Section position="10" start_page="183" end_page="183" type="concl">
    <SectionTitle>
7. CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> Our goal here was to generate a natural language text report on a set of concurrent processes. A full-fledged activity report should comprise three distinct segments. The preamble of the report should contain a succinct description of events which had occurred prior to the point in time when the last report was generated (in cases where multiple reports are requested).</Paragraph>
    <Paragraph position="1"> This should be followed by the main body of the report stating the events which have occurred between the time of the last report and the current time. Finally, there should be a brief mention of events which are expected to occur in the immediate future.</Paragraph>
    <Paragraph position="2"> Although attempts have been made in the current project to provide information about the past, the approach taken is simplistic. Improving this section of the report will involve reasoning about saliency of events in order to select events from past history for reporting. In addition, investigation regarding the formulation of the actual text for summarization must also be carried out.</Paragraph>
    <Paragraph position="3"> We believe that representation of status should allow the system, in principle, to draw inferences about expected future events. Our implementation does not address the issue of expectation, but can be a starting point toward the goal of including inferred knowledge about the immediate future with the description of the system's history.</Paragraph>
    <Paragraph position="4"> Another direction in which we intend to pursue further research is regarding presentation of unexpected events. This will necessitate incorporation of the ability to make inferences about what is expected and what is not. Additionally, appropriate text to report such events has to generated.</Paragraph>
    <Paragraph position="5"> Currently, in our system, no explicit textual links are established between the contents of a sentence and that of its predecessor(s). This seems satisfactory due to the general characteristics of narrative text where temporal succession as well as simultaneity is indicated implicitly by simple sequencing of sentences. However, further improvement in the naturalness of the text can be achieved by inclusion of appropriate clue words which function as explicit inter-sentential links. The PROTEUS sentence planner in \[Ritchie 84\] may be able to provide helpful insights in this attempt.</Paragraph>
    <Paragraph position="6"> Furthermore, the quality of text produced can be improved substantially by incorporating a selection of commonly used temporal expressions in English. A rich compendium of such expressions is available in \[Smith 78\].</Paragraph>
    <Paragraph position="7"> Finally, we want to investigate the issues involved in generating reports from the points of view of various participants. This will involve selecting events relevant to the person for whom the report is being prepared, and generating text in appropriate English.</Paragraph>
    <Paragraph position="8"> Choice of events will be dictated by various factors such as direct or indirect involvement of the reportee, his/her goals and responsibilities, and limitations regarding what he/she is allowed to know. Appropriate textual generation, i.e., choices of voice, sentence and clause structures, subjects and objects for the sentences as well as those of words will be governed by the nature of the listener's involvement with the events being described among other factors to be investigated. This will involve making a distinction between real events as stored in the historic knowledge base and virtual events characterizing how real events are reported from various perspectives.</Paragraph>
  </Section>
class="xml-element"></Paper>