<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2188">
  <Title>Dialogue Act Tagging with Transformation-Based Learning</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> For the task of recognizing dialogue acts, we are applying the Transformation-Based Learning (TBL) machine learning algorithm. To circumvent a sparse data problem, we extract values of well-motivated features of utterances, such as speaker direction, punctuation marks, and a new feature, called dialogue act cues, which we find to be more effective than cue phrases and word n-grams in practice. We present strategies for constructing a set of dialogue act cues automatically by minimizing the entropy of the distribution of dialogue acts in a training corpus, filtering out irrelevant dialogue act cues, and clustering semantically-related words. In addition, to address limitations of TBL, we introduce a Monte Carlo strategy for training efficiently and a committee method for computing confidence measures. These ideas are combined in our working implementation, which labels held-out data as accurately as any other reported system for the dialogue act tagging task.</Paragraph>
    <Paragraph position="1"> Introduction Although machine learning approaches have achieved success in many areas of Natural Language Processing, researchers have only recently begun to investigate applying machine learning methods to discourse-level problems (Reithinger and Klesen, 1997; Di Eugenio et al., 1997; Wiebe et al., 1997; Andernach, 1996; Litman, 1994). An important task in discourse understanding is to interpret an utterance's dialogue act, which is a concise abstraction of a speaker's intention, such as SUGGEST and AC-CEPT. Recognizing dialogue acts is critical for discourse-level understanding and can also be useful for other applications, such as resolving ambiguity in speech recognition. However, computing dialogue acts is a challenging task, because often a dialogue act cannot be directly inferred from a literal interpretation of an utterance. null We have investigated applying Transformation-Based Learning (TBL) to the task of computing dialogue acts. This method, which has not been used previously in discourse, has a number of attractive characteristics for our task. However, it also has some limitations, which we address with a Monte Carlo strategy that significantly improves the training time efficiency without compromising accuracy and a committee method that enables TBL to compute confidence measures for the dialogue acts assigned to utterances.</Paragraph>
    <Paragraph position="2"> Our machine learning algorithm makes use of abstract features extracted from utterances.</Paragraph>
    <Paragraph position="3"> In addition, we utilize an entropy-minimization approach to automatically identify dialogue act cues, which are words and short phrases that serve as signals for dialogue acts. Our experiments demonstrate that dialogue act cues tend to be more effective than cue phrases and word n-grams, and this strategy can be further improved by adding a filtering mechanism and a semantic-clustering method. Although we still plan to implement more modifications, our system has already achieved success rates comparable to the best reported results for computing dialogue acts.</Paragraph>
  </Section>
class="xml-element"></Paper>