<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1624">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A Weakly Supervised Learning Approach for Spoken Language Understanding</Title>
  <Section position="6" start_page="205" end_page="206" type="concl">
    <SectionTitle>
5 Conclusion and Future work
</SectionTitle>
    <Paragraph position="0"> We have presented a new SLU framework using two successive classifiers. The proposed framework exhibits the advantages as follows.</Paragraph>
    <Paragraph position="1"> null It has good robustness on processing spoken language: (1) The preprocessor provides the low level robustness. (2) It inherits the robustness of topic classification using statistical pattern recognition techniques. It can also make use of topic classification to guide slot filling. (3) The strategy of first finding the concepts or slot islands and then linking them is suited for processing spoken language.</Paragraph>
    <Paragraph position="2"> null It also keeps the understanding deepness: (1) The class of semantic classification is the slot name, which inherits the hierarchy from the domain model. (2) The semantic re-classification mechanism ensures the consistency among the identified slot-value pairs. null It is mainly data-driven and requires only minimally annotated corpus for training.</Paragraph>
    <Paragraph position="3"> Most importantly, our proposed SLU framework allows the employment of weakly supervised strategies for training the two classifiers, which can reduce the cost of annotating labeled sentences.</Paragraph>
    <Paragraph position="4"> The future work includes further evaluation of our approach in other application domains and languages. We also plan to integrate this understanding system into a whole dialog system. Then, high level knowledges, such as the dialog context, can also be included as the features of topic and semantic classifiers. Moreover, currently, the topics are manually defined through examination of the example sentences by human.</Paragraph>
    <Paragraph position="5"> Then, it is worthwhile to investigate how to appropriately define topics and the probability of  exploiting the sentence clustering techniques to facilitate the topic (frame) designment.</Paragraph>
  </Section>
class="xml-element"></Paper>