<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1072">
  <Title>Improving Summaries by Revising Them</Title>
  <Section position="2" start_page="0" end_page="558" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Writing improves with revision. Authors are familiar with the process of condensing a long paper into a shorter one: this is an iterative process, with the results improved over successive drafts. Professional abstractors carry out substantial revision and editing of abstracts (Cremrains 1996). We therefore expect revision to be useful in automatic text summarization. Prior research exploring the use of revision in summarization, e.g., (Gabriel 1988), (Robin 1994), (McKeown et al. 1995) has focused mainly on structured data as the input. Here, we examine the use of revision in summarization of text input.</Paragraph>
    <Paragraph position="1"> First, we review some summarization terminology. In revising draft summaries, these condensation operations, as well as stylistic rewording of sentences, play an important role. Summaries can be used to indicate what topics are addressed in the source text, and thus can be used to alert the user as to the source content (the indicative function). Summaries can also be used to cover the concepts in the source text to the extent possible given the compression requirements for the summary (the in formative function). Summaries can be tailored to a reader's interests and expertise, yielding topic-related summaries, or they can be aimed at a particular- usually broad - readership community, as in the cash of (so-called) generic summaries. Revision here applies to generic and topic-related informative summaries, intended for publishing and dissemination.</Paragraph>
    <Paragraph position="2"> Summarization can be viewed as a text-to-text reduction operation involving three main condensation operations: selection of salient portions of the text, aggregation of information from different portions of the text, and abstraction of specific information with more general information (Mani and Maybury 1999). Our approach to revision is to construct an initial draft summary of a source text and then to add to the draft additional background information.</Paragraph>
    <Paragraph position="3"> Rather than concatenate material in the draft (as surface-oriented, sentence extraction summarizers do), information in the draft is combined and excised based on revision rules involving aggregation (Dalianis and Hovy 1996) and elimination operations. Elimination can increase the amount of compression (summary length/source length) available, while aggregation can potentially gather and draw in relevant background information, in the form of descriptions of discourse entities from different parts of the source. We therefore hypothesize that these operations can result in packing in more information per unit compression than possible by concatenation. Rather than opportunistically adding as much background information that can fit in the available compression, as in (Robin 1994), our approach adds background information from the source text to the draft based on an information weighting function.</Paragraph>
    <Paragraph position="4"> Our revision approach assumes input sentences are represented as syntactic trees whose  nodes are annotated with coreference information. In order to provide open-domain coverage the approach does not assume a meaninglevel representation of each sentence, and so, unlike many generation systems, the system does not represent and reason about what is being said 1. Meaning-dependent revision operations are restricted to situations where it is clear from coreference that the same entity is being talked about.</Paragraph>
    <Paragraph position="5"> There are several criteria our revision model needs to satisfy. The final draft needs to be informative, coherent, and grammatically wellformed. Informativeness is explored in Section 4.2. We can also strive to guarantee, based on our revision rule set, that each revision will be syntactically well-formed. Regarding coherence, revision alters rhetorical structure in a way which can produce disfiuencies. As rhetorical structure is hard to extract from the source 2, our program instead uses coreference to guide the revision, and attempts to patch the coherence by adjusting references in revised drafts.</Paragraph>
  </Section>
class="xml-element"></Paper>