<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3208">
  <Title>Morphology Induction from Limited Noisy Data Using Approximate String Matching</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In order to develop morphological analyzers for languages that have limited resources (either in terms of experienced linguists, or electronic data), we must move beyond data intensive methods developed for rich resource languages that rely on large amounts of data for statistical methods. New approaches that can deal with limited, and perhaps noisy, data are necessary for these languages.</Paragraph>
    <Paragraph position="1"> Printed dictionaries often exist for languages before large amounts of electronic text, and provide a variety of information in a structured format. In this paper, we propose Morphology Induction from Noisy Data (MIND), a natural language morphology induction framework that operates on from information in dictionaries, specifically headwords and examples of usage. We use string searching algorithms to morphologically segment words and identify prefixes, suffixes, circumfixes, and infixes in noisy and limited data. We present our preliminary results on two data sources (Cebuano and Turkish), give a detailed analysis of results, and compare them to a state-of-the-art morphology learner. We employ the automatically induced affixes in a simple word segmentation process, decreasing the error rate of incorrectly segmented words by 35.41%.</Paragraph>
    <Paragraph position="2"> The next section discusses prior work on morphology learning. In Section 3 and 4, we describe our approach and MIND framework in detail. Section 6 explains the experiments and presents results. We conclude with future work.</Paragraph>
  </Section>
class="xml-element"></Paper>