<?xml version="1.0" standalone="yes"?>
<Paper uid="P95-1002">
  <Title>Automatic Induction of Finite State Transducers for Simple Phonological Rules</Title>
  <Section position="8" start_page="13" end_page="14" type="relat">
    <SectionTitle>
7 Related Work
</SectionTitle>
    <Paragraph position="0"> Johnson (1984) gives one of the first computational algorithms for phonological rule induction. His algorithm works for rules of the form (5) a ---+ b/C where C is the feature matrix of the segments around a. Johnson's algorithm sets up a system of constraint equations which C must satisfy, by considering both the positive contexts, i.e., all the contexts Ci in which a b occurs on the surface, as well as all the negative contexts Cj in which an a occurs on the surface. The set of all positive and negative contexts will not generally determine a unique rule, but will determine a set of possible rules.</Paragraph>
    <Paragraph position="1"> Touretzky et al. (1990) extended Johnson's insight by using the version spaces algorithm of Mitchell (1981) to induce phonological rules in their Many Maps architecture. Like Johnson's, their system looks at the underlying and surface realizations of single segments. For each segment, the system uses the version space algorithm to search for the proper statement of the context.</Paragraph>
    <Paragraph position="2"> Riley (1991) and Withgott &amp; Chen (1993) first proposed a decision-tree approach to segmental mapping. A decision tree is induced for each phoneme, classifying possible realizations of the phoneme in terms of contextual factors such as stress and the surrounding phonemes. However, since the decision tree for each phoneme is learned separately, the the technique misses generalizations about the behavior of similar phonemes. In addition, no generalizations are made about similar context phonemes. In a transducer based formalism, generalizations about similar context phonemes naturally follow from generalizations about individual phonemes' behavior, as the context is represented by the current state of the machine, which in turn depends on the behavior of the machine on the previous phonemes.</Paragraph>
    <Paragraph position="3"> We hope that our hybrid model will be more successful at learning long distance dependencies than the simple decision tree approach. To model long distance rules such as vowel harmony in a simple decision tree approach, one must add more distant phonemes to the features used to learn the decision tree. In a transducer, this information is represented in the current state of the transducer.</Paragraph>
  </Section>
class="xml-element"></Paper>