<?xml version="1.0" standalone="yes"?>
<Paper uid="E85-1019">
  <Title>Finite State Parsing of Phrase Structure Languages and the Status of Readjustment Rules</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> This paper describes an implemented parser-interpreter which is intended as an abstract formal model of part of the process of sentence comprehension. It is illustrated here for Phrase Structure Grammars with a translation into a familiar type of logical form, although the general principles are intended to apply to any grammatical theory sharing certain basic assumptions, which are discussed in the paper. The procedure allows for incremental semantic interpretation as a sentence is parsed, and provides a principled explanation for some familiar observations concerning properties of deeply recursive constructions.</Paragraph>
    <Paragraph position="1"> Background The starting point for the present work is a set of familiar and, for the most part, uncontroversial c|~Lm~ s about the nature of grammatical description and of human parsing of natural language.</Paragraph>
    <Paragraph position="2"> These claims and assumptions can be briefly summarised as follows: A Hierarchical Structure Linguists assign constituent structures to sentences on the basis ~f distributional tests of various kinds. On the basis of these tests, the 'correct' structures are always hierarchical and often deeply nested. The tree representing a sentence may impose a great deal of structure on it, with string-adjacent items often appearing at very different levels in the tree. In general, shallow, 'flat' structures are not generated by grammars, nor warranted on distributional grounds. However, as we shall see, it is likely that these deeply nested structures may be somewhat remote from any that are actually computed during parsing.</Paragraph>
    <Paragraph position="3"> B Semantics is (1) compositional and (ll) syntax-drlven.</Paragraph>
    <Paragraph position="4"> Both of these claims can be made in a variety of versions of different strengths, from the trivially true to the fairly clearly false. What is intended here is the assumption sometimes called the 'rule to rule' hypothesis, shared by almost all current grammatical frameworks, that to each syntactic rule of a grammar (or for each subrree induced by such a rule) there is an associated semantic rule, either producing an interpretation directly, or translating into some formal language. Interpretations for whole sentences are built up from the constituent parts in ways specified by these rules, in a fashion which mimics and uses the syntactic structure of the sentence.</Paragraph>
  </Section>
class="xml-element"></Paper>