<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2916">
  <Title>Unsupervised Grammar Induction by Distribution and Attachment</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Distributional approaches to grammar induction exploit the principle of substitutability: constituents of the same type may be exchanged with one another without affecting the syntax of the surrounding context. Reversing this notion, if we can identify &amp;quot;surrounding context&amp;quot; by observation, we can hypothesize that word sequences occurring in that context will be constituents of the same type. Thus, distributional methods can be used to segment text into constituents and classify the results. This work focuses on distributional learning from raw text.</Paragraph>
    <Paragraph position="1"> Various models of distributional analysis have been used to induce syntactic structure, but most use probabilistic metrics to decide between candidate constituents. We show that the efficiency of these systems can be improved by exploiting some properties of probable constituents, but also that this reliance on probability is problematic for learning from text. As a consequence, we propose an extension to strict distributional learning that incorporates more information about constituent boundaries.</Paragraph>
    <Paragraph position="2"> The remainder of this paper describes our experiences with a heuristic system for grammar induction. We begin with a discussion of previous distributional approaches to grammar induction in Section 2 and describe their implications in Section 3.</Paragraph>
    <Paragraph position="3"> We then introduce a heuristic distributional system in Section 4, which we analyze empirically against a treebank. Poor system performance leads us to examine actual constituent-context distributions (Section 5), the implications of which motivate a more structured extension to our learning system, which we describe and analyze in Section 6.</Paragraph>
  </Section>
class="xml-element"></Paper>