<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0818">
  <Title>LIHLA: Shared task system description</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Alignment of words and multiword units plays an important role in many natural language processing (NLP) applications, such as example-based machine translation (EBMT) (Somers, 1999) and statistical machine translation (SMT) (Ayan et al., 2004; Och and Ney, 2000), transfer rule learning (Carl, 2001; Menezes and Richardson, 2001), bilingual lexicography (G'omez Guinovart and Sacau Fontenla, 2004), and word sense disambiguation (Gale et al., 1992), among others.</Paragraph>
    <Paragraph position="1"> Aligning two (or more) texts means finding correspondences (translation equivalences) between segments (paragraphs, sentences, words, etc.) of the source text and segments of its translation (the target text). Following the same idea of many recently proposed approaches on lexical alignment (e.g., Wu and Wang (2004) and Ayan et al. (2004)), the method described in this paper, LIHLA (Language-Independent Heuristics Lexical Aligner) starts from statistical alignments between single words (defined in bilingual lexicons) and applies language-independent heuristics to them, aiming at finding the best alignments between words or multiword units.</Paragraph>
    <Paragraph position="2"> Although the most frequent alignment category is 1 : 1 (in which one source word is translated exactly as one target word), other categories such as omissions (1 : 0 or 0 : 1) or those involving multiword units (n : m, with n and/or m [?] 1) are also possible. This paper is organized as follows: section 2 explains how LIHLA works; section 3 describes some experiments carried out with LIHLA together with their results and, in section 4, some concluding remarks are presented.</Paragraph>
  </Section>
class="xml-element"></Paper>