<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0410">
  <Title>Using Summarization for Automatic Briefing Generation</Title>
  <Section position="3" start_page="0" end_page="101" type="intro">
    <SectionTitle>
2 Approach
</SectionTitle>
    <Paragraph position="0"> Our work forms part of a larger DARPA-funded project aimed at improving analysis and decision-making in crisis situations by providing tools that allow analysts to collaborate to develop structured arguments in support of particular conclusions and to help predict likely future scenarios. These arguments, along with background evidence, are packaged together as briefing s to high-level decision-makers. In leveraging automatic methods along the lines suggested above to generate briefings, our approach needs to allow the analyst to take on as much of the briefing authoring as she wants to (e.g., it may take time for her to adapt to or trust the machine, or she may want the machine to present just part of the briefing). The analyst's organisation usually will instantiate one of several templates dictating the high-level structure of a briefing; for example, a briefing may always have to begin with an executive summary. The summarization methods also need to be relatively domain-independent, given that the subject matter of crises are somewhat unpredictable; an analyst in a crisis situation is likely to be inundated with large numbers of crisis-related news and intelligence reports from many different sources. This means that we cannot require that a domain knowledge base be available to help the briefing generation process. Given these task requirements, we have adopted an approach that is flexible about accommodating different degrees of author involvement, that is relatively neutral about the rhetorical theory underlying the briefing structure (since a template may be provided by others), and that is domain-independent. In our approach, the author creates the briefing outline, which is then fleshed out further by the system based on information in the outline. The system fills out some content by invoking specified summarizers; it also makes decisions, when needed, about output media type; it introduces narrative elements to improve the coherence of the briefing; and finally, it assembles the final presentation, making decisions about spatial layout in the process.</Paragraph>
    <Paragraph position="1"> A briefing is represented as a tree. The structure of the tree represents the rhetorical structure of the briefing. Each node has a label, which offers a brief textual description of the node. Each leaf node has an associated goal, which, when realized, provides content for that node. There are two kinds of goals: content-level goals and narrative-level goals. Content-level goals are also of two kinds: retrieve goals, which retrieve existing media objects of a particular type (text, audio, image, audio, video) satisfying some description, and create goals, which create new media objects of these types using programs (called summarization filters). Narrative-level goals introduce descriptions of content at other nodes: they include captions and running text for media objects, and segues, which are rhetorical moves describing a transition to a node.</Paragraph>
    <Paragraph position="2"> Ordering relations reflecting temporal and spatial layout are defined on nodes in the tree.</Paragraph>
    <Paragraph position="3"> Two coarse-grained relations, seq for precedence, and par for simultaneity, are used to specify a temporal ordering on the nodes in the tree. As an example, temporal constraints for a (tiny) tree of 9 nodes may be expressed as:  The tree representation, along with the temporal constraints, can be rendered in text as XML; we refer to the XML representation as a script.</Paragraph>
    <Paragraph position="4">  The overall architecture of our system is shown in Figure 1, The user creates the briefing outline in the form of a script, by using a GUI. The briefing generator takes the script as input. The Script Validator applies an XML parser to the script, to check for syntactic correctness. It then builds a tree representation for the script, which represents the briefing outline, with temporal constraints attached to the leaves of the tree.</Paragraph>
    <Paragraph position="5"> Next, a Content Creator takes the input tree and expands it by introducing narrative-level goals including segues to content nodes, and rtmning text and captions describing media objects at content nodes. Running text and short captions are generated from meta-information associated with media objects, by using shallow text generation methods (canned text). The end result of content selection (which has an XML representation callod a ground script) is that the complete tree has been fully specified, with all the create and retrieve goals fully specified , with all the output media types decided. The Content Creator is thus responsible for both content selection and creation, in terms of tree structure and node content.</Paragraph>
    <Paragraph position="6"> Then, a Content Executor executes all the create and retrieve goals. This is a very simple step, resulting in the generation of all the media objects in the presentation, except for the audio files for speech to be synthesized. Thus, this step results in realization of the content at the leaves of the tree.</Paragraph>
    <Paragraph position="7"> Finally, the Presentation Generator takes the tree which is output from Content Execution, along with its temporal ordering constraints, and generates the spatial layout of the presentation. If no spatial layout constraints are specified (the default is to not specify these), the system allocates space using a simple method based on the temporal layout for nodes which have spatial manifestations. Speech synthesis is also carried out here. Once the tree is augmented with spatial layout constraints, it is translated by the  Language) (SMIL 99), a W3C-developod extension of HTML that can be played by standard multimedia players (such as Real 3 and Grins 4. This step thus presents the realized content, synthesizing it into a multimedia presentation laid out spatially and temporally.</Paragraph>
    <Paragraph position="8"> This particular architecture, driven by the above project requirements, does not use planning as an overall problem-solving strategy, as planning requires domain knowledge. It therefore differs from traditional intelligent multimedia presentation planners, e.g., (Wahlster etal. 93). Nevertheless, the system does make a number of intelligent decisions in organizing and coordinating presentation decisions. These are discussed next, after which we turn to the main point of the paper, namely the leveraging of summarization in automatic briefing generation.</Paragraph>
  </Section>
class="xml-element"></Paper>