<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2040">
  <Title>FINITE-STATE PARSING AND DISAMBIGUATION</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> The present approach is surface oriented and shallow, and it does not aim to uncover semantically oriented distinctions. An important source of inspiration has been Fred Karlsson's syntactic parser for Finnish, FPARSE (1985). The present approach tries to formalize the underlying ideas of that parser in a finite-state framework (cf.</Paragraph>
    <Paragraph position="1"> Karlsson 1989a,b). The finite-state formalism attacks the very basic things in syntax such as: what are the correct readings of ambiguous words, what are the clauses in a complex sentence, how the words form constituents, and what are the syntactic roles of the constituents.</Paragraph>
    <Paragraph position="2"> Let us consider the full framework of automatic syntactic parsing. One possible partition of the whole process is given in the following figure 1.</Paragraph>
    <Paragraph position="3"> The morphological analysis is done eg. by using the two-level model (Koskenniemi 1983).</Paragraph>
    <Paragraph position="4"> Comprehensive systems exist now for Finnish, Swedish, English and Russian (about 30-40,000 root entries in each), and some twenty smaller ones.</Paragraph>
    <Paragraph position="6"> i disambiguated sentence (clause boundaries, : , correct senses and syntactic functions selected) :</Paragraph>
    <Paragraph position="8"> The local disambiguation is an essential step eg. in Swedish, because many longer word-forms have several possible interpretations. In part, the local disambiguation supplements the two-level description by imposing more sophisticated restrictions on eg. compounds, and by reducing redundant or duplicate analyses (eg, in case a derived word both exists as a given lexicalized entry and is productively generated from its root), The remaining logic concerns weighing various alternatives and excluding readings which are significantly less probable than the best ones.</Paragraph>
    <Paragraph position="10"/>
  </Section>
class="xml-element"></Paper>