<?xml version="1.0" standalone="yes"?>
<Paper uid="H05-1036">
  <Title>Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 281-290, Vancouver, October 2005. c(c)2005 Association for Computational Linguistics Compiling Comp Ling: Practical Weighted Dynamic Programming and the Dyna Language[?]</Title>
  <Section position="8" start_page="289" end_page="289" type="concl">
    <SectionTitle>
8 Conclusions
</SectionTitle>
    <Paragraph position="0"> Weighted deduction is a powerful theoretical formalism that encompasses many NLP algorithms (Goodman, 1999). We have given a bottom-up &amp;quot;inside&amp;quot; algorithm for general semiring-weighted deduction, based on a prioritized agenda, and a general &amp;quot;outside&amp;quot; algorithm that correctly computes weight gradients even when the inside algorithm is pruned.</Paragraph>
    <Paragraph position="1"> We have also proposed a declarative language, Dyna, that replaces Prolog's Horn clauses with &amp;quot;Horn equations&amp;quot; over terms with values. Dyna can express more than the semiring-weighted dynamic programs treated in this paper. Our ongoing work concerns the full Dyna language, program transformations, and feedback-directed optimization.</Paragraph>
    <Paragraph position="2"> Finally, we evaluated our first implementation of a Dyna-to-C++ compiler (download and documentation at http://dyna.org). We hope it will facilitate EMNLP research, just as FS toolkits have done for the FS case. It produces code that is slower than hand-crafted code but acceptably fast for our NLP research, where it has been extremely helpful.</Paragraph>
  </Section>
class="xml-element"></Paper>