<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3408">
  <Title>ChAT: A Time-Linked System for Conversational Analysis</Title>
  <Section position="2" start_page="0" end_page="50" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The ability to extract and summarize content from data is a fundamental goal of computational linguistics. As such, a number of tools exist to automatically categorize, cluster, and extract information from documents. However, these tools do not transfer well to data sources that are more conversational in nature, such as multi-party meetings, telephone conversations, email, chat rooms, etc. Given the plethora of these data sources, there is a need to be able to quickly and accurately extract and process pertinent information from these sources without having to cull them manually.</Paragraph>
    <Paragraph position="1"> Much of the work on computational analysis of dialogue has focused on automatic topic segmentation of conversational data, and in particular, using features of the discourse to aid in segmentation (Galley et al, 2003; Stolcke et al., 1999; Hirschberg &amp; Hakatani, 1996.). Detailed discourse and conversational analytics have been the focus of much linguistic research and have been used by the computational community for creating models of dialogue to aid in natural language understanding and generation (Allen &amp; Core, 1997; Carletta et al., 1997; van Deemter et al., 2005; Walker et al., 1996). However, there has been much less focus on computational tools that can aid in either the analysis of conversations themselves, or in rendering conversational data in ways such that it can be used with traditional data mining techniques that have been successful for document understanding.</Paragraph>
    <Paragraph position="2"> This current work is most similar to the NITE XML Toolkit (Carletta &amp; Kilgour, 2005) which was designed for annotating conversational data.</Paragraph>
    <Paragraph position="3"> NITE XML is system in which transcripts of conversations are viewable and time aligned with their audio transcripts. It is especially useful for adding annotations to multi-modal data formats. NITE XML is not analysis tool, however. Annotations are generally manually added. In this paper, we present a Conversational Analysis Tool (ChAT) which integrates several language processing tools (topic segmentation, affect scoring, named entity extraction) that can be used to automatically annotate conversational data. The processing components have been specially adapted to deal with conversational data.</Paragraph>
    <Paragraph position="4"> ChAT is not an annotation tool, however, it is analysis tool. It includes a UI that combines a variety of data sources onto one screen that enables users to progressively explore conversational data.</Paragraph>
    <Paragraph position="5"> For example, one can explore who was present in a  given conversation, what they talked about, and the emotional content of the data. The data can be viewed by time slice or in a semantic graph. The language processing components in ChAT are versatile in that they were developed in modular, open designs so that they can be used independently or be integrated into other analytics tools. We present ChAT architecture and processing components in Section 2. In section 3 we present the UI , with a discussion following in section 4.</Paragraph>
  </Section>
class="xml-element"></Paper>