<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1106">
  <Title>Translating Lexical Semantic Relations: The First Step Towards Multilingual Wordnets*</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> A semantic network is critical to knowledge processing, including all NLP and Semantic Web applications. The construction of semantic networks, however, is notoriously difficult for 'small' (or 'low-density') languages. For these languages, the poverty of language resources, and the lack of prospect of material gains for infrastructure work conspire to create a vicious circle. This means that the construction of a semantic network for any small language must start from scratch and faces inhibitive financial and linguistic challenges.</Paragraph>
    <Paragraph position="1"> In addition, semantic networks serve as reliable ontolog(ies) for knowledge processing only if their conceptual bases are valid and logically inferable across different languages.</Paragraph>
    <Paragraph position="2"> Take wordnets (Fellbaum 1998), the de facto standard for linguistic ontology, for example.</Paragraph>
    <Paragraph position="3"> Wordnets express ontology via a network of words linked by lexical semantic relations. Since these words are by definition the lexicon of each language, the wordnet design feature ensures versatility in faithfully and comprehensively representing the semantic content of each language. Hence, on one hand, these conceptual atoms reflect linguistic idiosyncrasies; on the other hand, the lexical semantic relations (LSR's) receive universal interpretation across different languages. For example, the definition of relations such as synonymy or hypernymy is universal. The universality of the LSR's is the foundation that allows wordnet to serve as a potential common semantic network representation for all languages. The premise is tacit in Princeton WordNet (WN), EuroWordNet (EWN, Vossen 1998), and MultiWordNet (MWN, Pianta et al. 2002). It is also spelled out explicitly in the adaptation of LSR tests for Chinese (Huang et al. 2001).</Paragraph>
    <Paragraph position="4"> Given that LSR's are semantic primitives applicable to all language wordnets, and that the solution to the low-density problem in building language wordnets must involve bootstrapping from another language, LSR's seem to be the natural units for such bootstrapping operations.</Paragraph>
    <Paragraph position="5"> The rich and structured semantic information described in WN and EWN can be transported through accurate translation if the conceptual relations defined by LSRs remain constant in both languages. In practice, such an application would also serve the dual purpose of creating a bilingual wordnet in the process.</Paragraph>
    <Paragraph position="6"> In this paper, we will examine the validity of cross-lingual LSR inferences by bootstrapping a Chinese Wordnet with WN. In practice, this small-scale experiment shows how a wordnet for a low-density language can be built through bootstrapping from an available wordnet. In theoretical terms, we explore the logical conditions for the cross-lingual inference of LSR's.</Paragraph>
  </Section>
class="xml-element"></Paper>