<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0811">
  <Title>Models for Inuktitut-English Word Alignment</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Conventional word-alignment methods have been successful at treating many language pairs, but may be limited in their ability to generalize beyond the Western European language pairs for which they were originally developed, to pairs which exhibit more complex divergences in word order, morphology and lexical granularity. Our approach to Inuktitut-English alignment was to carefully consider the data in identifying dif culties particular to Inuktitut-English as well as possible simplifying assumptions. We used these observations to construct a novel weighted nite-state transducer alignment model as well as a specialized transliteration model. We combined these customized systems with 3 systems based on IBM Model 4 alignments under several methods of classi er combination. These combination strategies allowed us to produce multiple submissions targeted at the distinct evaluation measures via a precision/recall tradeoff. null</Paragraph>
  </Section>
class="xml-element"></Paper>