<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2155">
  <Title>Constituent-based Accent Prediction</Title>
  <Section position="6" start_page="942" end_page="943" type="evalu">
    <SectionTitle>
4 Results
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="942" end_page="942" type="sub_section">
      <SectionTitle>
4.1 Individual features
</SectionTitle>
      <Paragraph position="0"> Experimental results on individual features are reported in Table 4.1 in terms of the average percent correct classification and standard deviation. 4 A trend emerges that lexical features (i.e. word  and standard deviations for individual feature experiments. null lemma and broad class sequences, and form of expression) enable the largest improvements in classification, e.g. 2.7% and 2.3% for H1 using broad class sequence and form of expression information respectively. These results suggest that the abstract level of lexical description supplied by form of expression does the equivalent work of the lower-level lexical features. Thus, for CTS, accentuation class might be predicted when the more abstract form of expression information is known, and need not be 4Ripper experiments are conducted with 10-fold crossvalidation. Statistically significant differences in the performance of two systems are determined by using the Student's curve approximation to compute confidence intervals, following Litman (1996). Significant results at p &lt;.05 or stronger appear in italics.</Paragraph>
      <Paragraph position="1"> delayed until the tactical generation of the expression is completed. Conversely, for TTS, simple corpus analysis of lemma and POS sequences may perform as well as higher-level lexical analysis.</Paragraph>
    </Section>
    <Section position="2" start_page="942" end_page="942" type="sub_section">
      <SectionTitle>
4.2 Combinations of classes of features
</SectionTitle>
      <Paragraph position="0"> Experiments on combinations of feature classes are reported in Table 7.</Paragraph>
      <Paragraph position="1">  tion and standard deviations for combination experiments. null of 63.17% for H3 on the local focus and lexical feature class model, is the best obtained for all H3 experiments, increasing prediction accuracy by nearly 3%. The highest classification rate for H1 is 79.22% for the model including local and global focus, and lexical and syntactic feature classes, showing an improvement of 3.4%. These results, however, do not attain significance.</Paragraph>
    </Section>
    <Section position="3" start_page="942" end_page="943" type="sub_section">
      <SectionTitle>
4.3 Experiments on simple-baseNPs
</SectionTitle>
      <Paragraph position="0"> Three sets of experiments that showed strong performance gains are reported for the non-recursive simple-baseNPs. These are: (1) word lemma sequence alone, (2) lemma and broad class sequences together, and (3) local focus and lexical features combined. Table 8 shows the accent class distribution for simple-baseNPs.</Paragraph>
      <Paragraph position="1">  baseNPs.</Paragraph>
      <Paragraph position="2"> Results appear in Table 9. For H3, the lemma sequence model delivers the best performance, 65.71%, for a 4.3% improvement over the baseline. The best classification rate of 80.93% for H1 on the local focus and lexical feature model represents a 6.23% gain over the baseline. These figures represent an 11% reduction in error rate for H3, and a  25% reduction in error rate for HI, and are statistically significant improvements over the baseline.  and standard deviations for simple-baseNP experiments. null In the rule sets learned by Ripper for the H1 local focus/lexical model, interactions of the different features in specific rules can be observed. Two rule sets that performed with error rates of 13.6% and 13.7% on different cross-validation runs are presented in Figure 1.5 Inspection of the rule sets H1 local focus/lexical model rule set 1 reduced :- form of expr=proper name, broad class seq --- det, lemma seq ,-~ Harvard.</Paragraph>
      <Paragraph position="3"> supra :- broad class seq --~ adverbial.</Paragraph>
      <Paragraph position="4">  H1, local focus/lexical model.</Paragraph>
      <Paragraph position="5"> reveals that there are few non-lexical rules learned. The exception seems to be the rule that adverbial noun phrases belong to the supra accent class. However, new interactions of local focusing features (grammatical function and form of expression) with lexical information are discovered by Ripper. It also appears that as suggested by earlier experiments, 5In the rules themselves, written in Prolog-style notation, the tilde character is a two-place operator, X -,~ Y, signifying that Y is a member of the set-value for feature X. lexical features trade-off for one other as well as with form of expression information. In comparing the first rules in each set, for example, the clauses broad class seq ,,~ det and lemma seq ,~ the substitute for one another. However, in the first rule set the less specific broad class constraint must be combined with another abstract constraint, form of expr=proper name, to achieve a similar description of a rule for reduced accentuation on common place names, such as the Harvard Square T stop.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>