<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-1302">
  <Title>On Statistical Parameter Setting</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> We present a model and an experimental platform of a bootstrapping approach to statistical induction of natural language properties that is constraint based with voting components. The system is incremental and unsupervised. In the following discussion we focus on the components for morphological induction. We show that the much harder problem of incremental unsupervised morphological induction can outperform comparable all-at-once algorithms with respect to precision. We discuss how we use such systems to identify cues for induction in a cross-level architecture.</Paragraph>
  </Section>
class="xml-element"></Paper>