<?xml version="1.0" standalone="yes"?>
<Paper uid="W94-0206">
  <Title>PARSING USING LINEARLY ORDERED PHONOLOGICAL RULES</Title>
  <Section position="12" start_page="66" end_page="67" type="concl">
    <SectionTitle>
5. CONCLUSION
</SectionTitle>
    <Paragraph position="0"> The algorithm as described and implemented models segmental phonology. An extension to multiple strata of rules, as in lexical phonology, is trivial, and has also been implemented. Allowing cyclic application of rules is also simple, although it has not been implemented yet (because most phonologists since Pesetsky 1979 have interpreted cycli c phonology as the interleaving of phonological rules and morphological rules, and morphological rules have not yet been implemented).</Paragraph>
    <Paragraph position="1"> The algorithm could be extended to autosegmentai models of phonology by reinterpreting e.g. feature spreading rules as feature assignment rules with alpha variables during the analysis phase, and reverting to the standard interpretation of autosegmental rules during synthesis. For instance, an autosegrnental rule spreading the place of articulation features of an obstruent onto a preceding nasal consonant can be modeled during analysis by the following rule: . _, r-continuent-F+cons \] \[~high 1, \]c~ high poacK / k+nas J /rcoron J--/pba k - L?,corona 1 The modeling of autosegmental phonology has not been implemented, although the use of alpha variables has.</Paragraph>
    <Paragraph position="2"> In summary, and contrary to many earlier claims, it need not be eomputationally expensive to parse surface forms into their underlying forms using linearly ordered rules. Furthermore, unlike rule compilers (Kaplan and Kaye in press), the use of a rule interpreter simplifies grammar debugging, as the input and output of each rule can be studied (a sample trace is shown in the appendix).</Paragraph>
  </Section>
class="xml-element"></Paper>