<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-2022">
  <Title>Cross-linguistic phoneme correspondences</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Tiberius and Cahill (2000) presented the theory of cross-linguistic phoneme correspondences (metaphonemes) with an example pilot study of the vowels of Dutch, English and German. The aim of this work is to allow the type of generalisation that is permitted by the use of phonemes with allophonic variation to be taken one level higher, i.e. above the level of the single language. The idea behind it is to represent the near-identities that closely related languages such as Dutch, English and German so often share. For example, the Dutch word 'kat' /kAt/ has the English equivalent 'cat' /k a0 t/, and the German 'Katze' /kats@/2. While the consonants are largely identical (/k/-/k/-/k/ and /t/-/t/-/ts/), the vowels are subtly different. However, they are not distinctive - i.e. if the / a0 / in English were replaced with /a/ it would not sound like a different word, but rather it would sound like a different accent. Thus our aim is not to construct a universal phoneme set representing all phonemes occurring in a particular set  1995) and use the SAMPA phonetic alphabet (Wells, 1989). of languages, but we aim to capture phoneme correspondences between languages such as the / a0 / - /a/ /A/ correspondence mentioned above. Our work is, therefore, different from proposals put forward by Deng (1997) who defines a set of universal phonological features to be used for multilingual speech recognition.</Paragraph>
    <Paragraph position="1"> The three language-specific vowels discussed above can be grouped together into the metaphoneme a1 a0 Aa a1 , which will be realised as an / a0 / in English, an /A/ in Dutch, and an /a/ in German3. Tiberius and Cahill (2000) described the vowel metaphonemes for these three languages. In this paper we describe a similar experiment that looked at the consonants of the three languages. The consonants are interestingly different from the vowels for a number of reasons: a2 the consonant space is more discrete than the vowel space, so there is less scope for small and non-meaning-bearing distinctions within the consonants; a2 the phoneme inventories of the three languages show that, while they have significantly different vowel inventories, their consonant inventories overlap greatly; a2 while vowels were considered to occur one per syllable (i.e. long vowels and diphthongs were treated as single vowels), consonants can occur in clusters at either the beginning or end of syllables; a2 unlike vowels, consonants can be lost altogether, thus leading to synchronic alternations 3We use the notation a3 xyz a3 to denote the metaphoneme where x, y and z are normally the sounds for the three languages in the order English, Dutch, and German. However, this is intended only as a mnemonic and does not necessarily imply that these three sounds always occur. Metaphonemes may involve quite complex definitions that are dependent on phonological context as well as just the language in question.</Paragraph>
    <Paragraph position="2"> between zero and other segments.</Paragraph>
    <Paragraph position="3"> In this paper we present the results of our experiment, first discussing the exact nature of metaphonemes, comparing them with archiphonemes and morphophonemes as used in traditional approaches to morphology as well as the keysymbols used by Fitt (2001) to define cross-accent differences. We then go on to describe the methodology used and the results obtained. Finally we discuss the implications of our findings both on the intended application, i.e. multilingual lexicons, and for other fields such as historical linguistics.</Paragraph>
  </Section>
class="xml-element"></Paper>