<?xml version="1.0" standalone="yes"?>
<Paper uid="N03-4003">
  <Title>DOGHED: A Template-Based Generator for Multimodal Dialog Systems Targeting Heterogeneous Devices[?]</Title>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 DOGHED
</SectionTitle>
    <Paragraph position="0"> DOGHED is a template-based multimodal output generator for dialog systems that need to support heterogeneous devices. It enables dialog systems to create multimodal presentations for different devices in realtime. DOGHED extends YAG (Yet Another Generator), [?]We acknowledge the financial support of the National Science Foundation (under grants IRI-9701617 and DUE9952703), Wright State University, and Intel Corporation.</Paragraph>
    <Paragraph position="1"> a template-based text realization system (Channarukul et al., 2000; Channarukul et al., 2001; McRoy et al., 2003) by providing a pre-defined set of multimodal templates.</Paragraph>
    <Paragraph position="2"> It employs JYAG (Channarukul et al., 2002), the Java implementation of YAG, to realize those templates for further display by appropriate browsers. It provides output in the Synchronized Multimedia Integration Language (SMIL) which can be presented on any SMIL player such as RealOne Player and X-SMILES (Pihkala et al., 2001); or any capable web browser.</Paragraph>
    <Paragraph position="3"> DOGHED is a generic and domain-independent component for generating multimodal presentations in that it accepts a feature structure as input. These feature structures can also embed other feature structures to represent a more complicated input specification. Moreover, an application can specify content, user preferences, and device constraints (e.g., multimedia capabilities, screen size and resolution, and network bandwidth) using this uniform formalism. Natural language can also be inserted by embedding a feature structure that calls for realization using one or more YAG's English syntactic templates.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Multimodal Templates
</SectionTitle>
    <Paragraph position="0"> There are two types of multimodal templates. The first type is for generating individual SMIL tags and structures. The other type is more abstract and captures the semantics of multimodal presentation. A semantic template can be authored so that an application only needs to specify its selected content, intended presentation style, and other constraints. Output will then be generated to best suit the given style and constraints. For example, an application might want to compare two diagrams or items that are related. On large display devices, both items can be displayed side-by-side. However, on smaller display devices, like PDAs, such presentation is not possible; it would be better to display one item at a time and switch between the two items being compared.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4 Template Authoring Tool
</SectionTitle>
    <Paragraph position="0"> In addition to the pre-defined multimodal template set, application developers can write their own templates to suit their needs. These new templates can be written from scratch or built on top of existing ones. We facilitate the task of template authoring by providing a graphical development environment. IDEY (Integrated Development Environment for YAG) provides support for authoring, testing, and managing templates (Channarukul et al., 2002). Its graphical interface reduces the amount of time needed for syntax familiarization through direct manipulation and template visualization. It also allows a developer to test newly constructed templates easily. Multimodal output can be immediately displayed and verified on an integrated SMIL player (Figure 1). Moreover, the interface also helps prevent errors by constraining the way in which templates may be constructed or modified.</Paragraph>
    <Paragraph position="1"> For example, values of slots in templates are constrained by context-sensitive pop-up menu choices.</Paragraph>
  </Section>
class="xml-element"></Paper>