<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3003">
  <Title>Interactive Machine Learning Techniques for Improving SLU Models</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> We presented an interactive speech data analysis system for creating and testing spoken language understanding systems. Spoken language understanding is a critical component of automated customer service applications.</Paragraph>
    <Paragraph position="1"> Creating effective SLU models is inherently a data driven process and requires considerable human intervention. The fact that this process relies heavily on human expertise prevents a total automation of the process. Our experience indicates that augmenting the human expertise with interactive data analysis techniques made possible by machine learning techniques can go a long way towards increasing the efficiency of the process and the quality of the final results. The automatic preprocessing of the utterance data prior to its use by the UE expert results in a considerable reduction in the number of utterances that needs to be manually examined. Clustering uncovers certain structures in the data that can then be refined by the UE expert. Supervised machine learning capabilities provided by interactive relevance feedback tend to capture the knowledge of the UE expert to create the guidelines for labeling the data. The ability to test the generated call types during the design process helps detect and remove problematic call types prior to their inclusion in the SLU model.</Paragraph>
    <Paragraph position="2"> This tool has been used to create the labeling guide for several applications by different UE experts. Aside from the increased efficiency and improved quality of the generated SLU systems, the tool has resulted in increased uniformity in the way different UE experts classify calls into call type labels.</Paragraph>
  </Section>
class="xml-element"></Paper>