<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1112">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A Structural Similarity Measure</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Although the similarity of natural languages is in principal a very vague notion, the linguistic literature seems to be full of claims classifying two natural languages as being more or less similar. These claims are in some cases a result of a detailed comparative examination of lexical and/or syntactic properties of languages under question, in some cases they are based on a very subjective opinion of the author, in many other cases they reflect the application of some mathematical formula on textual data (a very nice example of such mathematical approach can be found at (Scannell, 2004)).</Paragraph>
    <Paragraph position="1"> Especially in the last case the notion of language similarity is very often confused with the notion of text similarity. Even the well known paper (Lebart and Rajman, 2000) deals more with the text similarity than language similarity. This general trend is quite understandable, the mathematical methods for measuring text similarity are of a prominent importance especially for information retrieval and similar fields. On the other hand, they concentrate too much on the surface similarity of word forms and thus may not reflect the similarity of languages properly. This paper tries to advocate different approach, based on the experience gained in MT experiments with closely related (and similar) languages, where it is possible to &amp;quot;measure&amp;quot; the similarity indirectly by a complexity of modules we have to use in order to achieve a reasonable translation quality. This experience led us to formulating an evaluation measure trying to capture not only textual, but also syntactic similarities between natural languages.</Paragraph>
  </Section>
class="xml-element"></Paper>