<?xml version="1.0" standalone="yes"?>
<Paper uid="C00-2110">
  <Title>Japanese Dependency Analysis using a Deterministic Finite State Transducer</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Syntactic analysis or parsing based on traditional methods, like Chm't parsing or the GLR parsing algorithm, takes cubic or greater tin:e in the sentence length to analyze natural language sentences. For Japmmse, Sekine et al.</Paragraph>
    <Paragraph position="1"> (Sekine et al., 2000) proposed a Japanese dependency analyzer which mmlyzes sentences in time quadratic in the sentence length using a backward search algorithm. Recently, a mnnber of research efforts using Finite State Transducers (FST) have been reported. Roche built an English syntactic mmlyzer by finding a fixed point in a non-deterministic FST (Roche, 1994). But it still can't anMyze a sentence in time linear in the sentence length.</Paragraph>
    <Paragraph position="2"> In this paper, we will propose a Japanese dependency analyzer using a Deterministic Finile State Transducer (DFST). The Japmtese dependency structure is usually represented by relationships between phrasal units called %unsets::'. A lmnsetsu usually contains one or more content words, like a noun, verb or adjective, and zero or more time:ion words, lil:e a post-position (case marker) or verb/noun suffix. A dependency between two bunsetsu has a direction fl'om a dependent to its head. Figure 1 shows examples of lmnsetsu and dependencies.</Paragraph>
    <Paragraph position="3"> Each lmnsetsu is separated by &amp;quot;1&amp;quot;. The frst seglnent &amp;quot;KANOJO-HA&amp;quot; consists of two words, KANOJO (She) and HA (subject case, ,narl:er).</Paragraph>
    <Paragraph position="4"> The nmnbers in the &amp;quot;head&amp;quot; line show the head ID of corresponding bunsctsus. Note that the last segment does not have a head, and it is the head lmnsetsu of the sentence. The task of the dependency analysis is to find the head ID for each lnmsetsu.</Paragraph>
  </Section>
class="xml-element"></Paper>