<?xml version="1.0" standalone="yes"?>
<Paper uid="C00-2175">
  <Title>Comparing two trainable grammatical relations finders</Title>
  <Section position="2" start_page="0" end_page="1146" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Grmnnmtical relationships (GRs), whidl include arguments (e.g., subject and object) and modifiers, form an important level of natural language processing. Glls in the sentence Yesterday, my cat ate th, e food in the bowl.</Paragraph>
    <Paragraph position="1"> include ate having tile subject my cat, the object the food and the time modifier Ycstcr'day, ~md t, hc .food having the location modifier in (the bowl).</Paragraph>
    <Paragraph position="2"> However, different sets of GRs are useful for dii%rent purposes. For exmnple, Ferro et al.</Paragraph>
    <Paragraph position="3"> (1999) is interested in semantic interpretation, and needs to differentiate between time, location and other modifiers. The SPARKLE project (Carroll et al., 1997), on the other lmnd, * This paper reports on work performed at, the MITRE Corporation under the support of the MITRE Sponsored Research Program. Marc Vilain, Lynette Hirsehman and Warren Greiff have helped make this work happen. Christine l)oran and John Henderson provided helpflfl editing. Copyright @2000 The MITRE Corporation. All rights reserved.</Paragraph>
    <Paragraph position="4"> does not differentiate between these types of modifiers. As has been mentioned by John Carroll (personal communication), this is fine for infbrmation retrieval. Also, having less differentiation of tile modifiers can make it, easier to find them (Ferro et al., 1999).</Paragraph>
    <Paragraph position="5"> Unless the desired set of GRs matches the set already annotated in some large training col pus (e.g., the Buchholz el; al. (1999) GR finder used the GRs annotated in the Penn 3~'eelmnk (Marcus el; al., 1993)), one will have to either manually write rules to find tile GI{s or mmotate a training corpus tbr the desired set. Manually writing rules is expensive, as is annotating a large corpus. We have performed experiments on learning to find ORs with just a small annotated training set. Our starting point is the work described in l?erro et al. (1999), which used a faMy smM1 training set.</Paragraph>
    <Paragraph position="6"> This paper reports on a comparison between the transforination-based error-driven learner described in Ferro et al. (1999) and the lnemory-based learner for GRs described in Buchholz et M. (1.999) on finding GIls to verbs 1 by retraining the memory-based learner with tile data used in Ferro et al. (1999). We find that the transformation versus memory-based difference only seems to cause a small difference in the results. Most of the result differences seem to instead be caused by differences in tile representations and information used by tile learners. An example is that different GR length measures are used. In English, one measure seems better fbr recovering simple argument ORs, while another measure seems better ibr modifier GIl.s. We also find that partitioning  ing.</Paragraph>
  </Section>
class="xml-element"></Paper>