<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0614">
  <Title>Intentional Context in Situated Natural Language Learning</Title>
  <Section position="9" start_page="110" end_page="110" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have introduced a model of language acquisition that explicitly incorporates intentional contexts in both learning and understanding. We have described pilot experiments on paired language and action data in order to demonstrate both the model's feasibility as well as the efficacy of using intentional context in understanding.</Paragraph>
    <Paragraph position="1"> Although we have demonstrated a first step toward an advanced model of language acquisition, there is a great deal that has not been addressed. First, what is perhaps most obviously missing is any mention of syntax in the language learning process and its role in bootstrapping for language acquisition. Future work will focus on moving beyond the IBM Model 1 assumptions, to develop more syntactically-structured models.</Paragraph>
    <Paragraph position="2"> Further, although the virtual environment used in this research bears similarity to situated applications that demand NL interfaces, it is not known exactly how well the model will perform &amp;quot;in the real world.&amp;quot; Future work will examine installing models in real world applications. In parallel investigations, we will explore our method as a cognitive model of human language learning.</Paragraph>
    <Paragraph position="3"> Finally, as was mentioned previously, the task model for this domain was hand annotated and, while the constrained nature of the domain simplified this process, further work is required to learn such models jointly with language.</Paragraph>
    <Paragraph position="4"> In summary, we have presented first steps toward tackling problems of ambiguity inherent in grounding the semantics of situated language. We believe this work will lead to practical applications for situated NLP, and provide new tools for modeling human cognitive structures and processes underlying situated language use (Fleischman and Roy, 2005).</Paragraph>
  </Section>
class="xml-element"></Paper>