<?xml version="1.0" standalone="yes"?>
<Paper uid="J90-3003">
  <Title>A COMPUTATIONAL GRAMMAR OF DISCOURSE-NEUTRAL PROSODIC PHRASING IN ENGLISH</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> In previous work (Bachenko et al. 1986), we described an experimental text-to-speech system that determined prosodic phrasing for the Olive-Liberman synthesizer (Olive and Liberman 1985). The system generated phrase boundaries using information derived from the syntactic structure of a sentence. While we saw significant improvements in the resulting synthesized speech, we also observed problems with the system. Often these stemmed from our assumptions that both clausal structure and predicate-argument relations were important in determining prosodic phrasing. This paper reconsiders those assumptions and describes an analysis of phrasing that we believe corrects many of the problems of the earlier version. Like the earlier version, it has been implemented in a text-to-speech system that uses a natural language parser and prosody rules to generate information about the location and relative strength of prosodic phrase boundaries.</Paragraph>
    <Paragraph position="1"> Our current analysis rests on two ideas. First, it is possible to describe a level of prosodic phrasing that is independent of discourse semantics. Second, this discourse-neutral phrasing depends on a mix of syntactic and nonsyntactic factors; chiefly, syntactic constituency, left-to-right word order, and constituent length. There is no necessary fit between syntactic structure and phrasing, since prosodic phrasing may ignore major syntactic boundaries in order to satisfy the constraints on phrase length. Our approach thus follows that of Grosjean et al. (1979), namely, that phrasing reflects &amp;quot;... two (sometimes conflicting) demands on the speaker: the need to respect the linguistic structure of the sentence and the need to balance the length of the constituents in the output&amp;quot; (p. 75).</Paragraph>
    <Paragraph position="2"> Section 2 will outline our analysis, focusing on the relationship between syntactic and prosodic structure. The analysis is developed within the framework of generative grammar, but we believe it is consistent with other approaches to syntactic description. 1 Our main point will be that the syntax plays a necessary but not sufficient role in determining phrasing, its effects being filtered by separate conditions on prosodic well-formedness (e.g. length). Section 3 describes the implementation of our analysis in an experimental text-to-speech system, and Section 4 summarizes our main conclusions. Unless otherwise noted, the corpus we used as a source of observations on phrasing in human speech consisted of a taped professional dramatization of the Sherlock Holmes story The Speckled Band and a documentary about Mount Everest that includes both professional &amp;quot;prepared&amp;quot; narration and the spontaneous speech of interviews. The Holmes story involved two male speakers and one female speaker; the Everest documentary involved a male narrator and several male interviewees.</Paragraph>
    <Paragraph position="3"> Both of us independently transcribed the tapes according to our perceptions of prosodic phrasing. Other examples come from transcriptions of speech that we recorded at Bell Laboratories. In the transcriptions, we both distinguished three types of prosodic event: a primary phrase boundary, a secondary phrase boundary, and the absence of a boundary.</Paragraph>
    <Paragraph position="4"> The most salient characteristic of the primary phrase boundary was a pause, while that of the secondary boundary involved a change in pitch. In comparing the two transcriptions, we discarded cases in which there was a discrepancy between the two markings, which left us with a corpus of approximately 500 sentences against which the prosodic phrasing rules were tested.</Paragraph>
    <Paragraph position="5"> Computational Linguistics Volume 16, Number 3, September 1990 155 J. Bachenko and E. Fitzpatrick Discourse-Neutral Prosodic Phrasing in English</Paragraph>
  </Section>
class="xml-element"></Paper>