<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1510">
  <Title>Resource sharing among HPSG and LTAG communities by a method of grammar conversion from FB-LTAG to HPSG</Title>
  <Section position="6" start_page="1" end_page="6" type="evalu">
    <SectionTitle>
4 Experiments
</SectionTitle>
    <Paragraph position="0"> The RenTAL system is implemented in LiLFeS (Makino et al., 1998)  . LiLFeS is one of the fastest inference engines for processing feature structure logic, and efficient HPSG parsers have already been built on this system (Nishida et al., 1999; Torisawa et al., 2000). We applied our system to the XTAG English gram- null We used the grammar attached to the latest distribution of an LTAG parser which we used for the parsing experiment. The parser is available at: ftp://ftp.cis.upenn.edu/pub/xtag/lem/lem-0.13.0.i686.tgz  elementary tree templates and around 45,000 lexical items  . We successfully converted all the elementary tree templates in the XTAG English grammar to HPSG lexical entry templates. Table 1 shows the classifications of elementary tree templates of the XTAG English grammar, according to the conditions we introduced in Section 3, and also shows the number of corresponding HPSG lexical entry templates. Conversion took about 25 minutes CPU time on a 700 Mhz Pentium III Xeon with four gigabytes main memory. null The original and the obtained grammar generated exactly the same number of derivation trees in the parsing experiment with 457 sentences from the ATIS corpus (Marcus et al., 1994)  (the average length is 6.32 words). This result empirically attested the strong equivalence of our algorithm. null Table 2 shows the average parsing time with the LTAG and HPSG parsers. In Table 2, lem refers to the LTAG parser (Sarkar et al., 2000), ANSI C implementation of the two-phase parsing algorithm that performs the head corner parsing (van Noord, 1994) without features (phase 1), and then executes feature unification (phase 2). TNT refers to the HPSG parser (Torisawa et al., 2000), C++ implementation of the two-phase parsing algorithm that performs filtering with a compiled CFG (phase 1) and then executes feature unification (phase 2). Table 2 clearly shows that the HPSG parser is significantly faster than the LTAG parser. This result implies that parsing techniques for HPSG are also beneficial for LTAG  We eliminated 32 elementary trees because the LTAG parser cannot produce correct derivation trees with them.  These lexical items are a subset of the original XTAG English grammar distribution.</Paragraph>
    <Paragraph position="1">  We eliminated 59 sentences because of a time-out of the parsers, and 61 sentences because the LTAG parser does not produce correct derivation trees because of bugs in its preprocessor.</Paragraph>
    <Paragraph position="2"> parsing. We can say that the grammar conversion enables us to share HPSG parsing techniques in LTAG parsing. Another paper (Yoshinaga et al., 2001) describes the detailed analysis on the factor of the difference of parsing performance.</Paragraph>
  </Section>
class="xml-element"></Paper>