<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2115">
  <Title>From Prosodic Trees to Syntactic Trees</Title>
  <Section position="2" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> This paper describes an ongoing effort to parse the Hebrew Bible. The parser consults the bracketing information extracted from the cantillation marks of the Masoetic text. We first constructed a cantillation treebank which encodes the prosodic structures of the text. It was found that many of the prosodic boundaries in the cantillation trees correspond, directly or indirectly, to the phrase boundaries of the syntactic trees we are trying to build. All the useful boundary information was then extracted to help the parser make syntactic decisions, either serving as hard constraints in rule application or used probabilistically in tree ranking.</Paragraph>
    <Paragraph position="1"> This has greatly improved the accuracy and efficiency of the parser and reduced the amount of manual work in building a Hebrew treebank.</Paragraph>
    <Paragraph position="2"> Introduction The text of the Hebrew Bible (HB) has been carefully studied throughout the centuries, with detailed lexical, phonological and morphological analysis available for every verse of HB.</Paragraph>
    <Paragraph position="3"> However, very few attempts have been made at a verse-by-verse syntactic analysis. The only known effort in this direction is the Hebrew parser built by George Yaeger (Yaeger 1998, 2002), but the analysis is still incomplete in the sense that not all syntactic units are recognized and the accuracy of the trees are yet to be checked.</Paragraph>
    <Paragraph position="4"> Since a detailed syntactic analysis of HB is of interest to both linguistic and biblical studies, we launched a project to build a treebank of the Hebrew Bible. In this project, the trees are automatically generated by a parser and then manually checked in a tree editor. Once a tree has been edited or approved, its phrase boundaries are recorded in a database. When the same verse is parsed again, the existing brackets will force the parser to produce trees whose brackets are exactly the same as those of the manually approved trees. Compared with traditional approaches to treebanking where the correct structure is preserved in a set of tree files, our approach has much more agility. In the event of design/format changes, we can automatically regenerate the trees according to the new specifications without manually touching the trees. The bracketing information will persist through the updates and the basic structure of the trees will remain correct regardless of the changes in the details of trees. We call this a &amp;quot;dynamic treebank&amp;quot; where, instead of maintaining a set of trees, we maintain a parser/grammar, a dictionary, a set of sentences, and a database of bracketing information. The trees can be generated at any time.</Paragraph>
    <Paragraph position="5"> Since our parser/grammar can consult known phrase boundaries to build trees, its performance can be greatly improved if large amounts of bracketing information are available. Human inspection and correction can provide those boundaries, but the amount of manual work can be reduced significantly if there is an existing source of bracketing information for us to use. Fortunately, a great deal of such information can be obtained from the cantillation marks of the Hebrew text.</Paragraph>
  </Section>
class="xml-element"></Paper>