<?xml version="1.0" standalone="yes"?>
<Paper uid="P96-1040">
  <Title>The Rhythm of Lexical Stress in Prose</Title>
  <Section position="6" start_page="306" end_page="306" type="concl">
    <SectionTitle>
5 Conclusions and future work
</SectionTitle>
    <Paragraph position="0"> We have quantified lexical stress regularity, measured it in a large sample of written English prose, and shown there to be a significant contribution from word order that increases with lexical perplexity.</Paragraph>
    <Paragraph position="1"> This contribution was measured by comparing the entropy rate of lexical stress in natural sentences with randomly permuted versions of the same. Randomizing the word order in this way yields a fairly crude baseline, as it produces asyntactic sequences in which, for example, single-syllable function words can unnaturally clash. To correct for this we modified the randomization algorithm to permute only open-class words and to fix in place determiners, particles, pronouns, and other closed-class words.</Paragraph>
    <Paragraph position="2"> We found the entropy rates to be consistently midway between the fully randomized and unrandomized values. But even this constrained randomization is weaker than what we'd like. Ideally we should factor out semantics as well as word choice, comparing each sentence in the corpus with its grammatical variations. While this is a difficult experiment to do automatically, we're hoping to approximate it using a natural language generation system based on link grammar under development by the author.</Paragraph>
    <Paragraph position="3"> Also, we're currently testing other data sources such as the Switchboard corpus of telephone speech (Godfrey, Holliman, and McDaniel, 1992) to measure the effects of rhythm in more spontaneous and grammatically relaxed texts.</Paragraph>
  </Section>
class="xml-element"></Paper>