<?xml version="1.0" standalone="yes"?>
<Paper uid="P95-1034">
  <Title>Two-Level, Many-Paths Generation</Title>
  <Section position="3" start_page="0" end_page="252" type="intro">
    <SectionTitle>
2 Knowledge Gaps
</SectionTitle>
    <Paragraph position="0"> In our machine translation experiences, we traced generation disfluencies to two sources: 1 (1) incomplete or inaccurate conceptual (interlingua) structures, caused by knowledge gaps in the source language analyzer, and (2) knowledge gaps in the generator itself. These two categories of gaps include:  * The generation lexicon does not contain much collocational knowledge (e.g., on the field vs.</Paragraph>
    <Paragraph position="1"> *on the end zone).</Paragraph>
    <Paragraph position="2"> * Lexico-syntactic constraints (e.g., tell her hi vs. *say her hi), syntax-semantics mappings (e.g., the vase broke vs. *the food ate), and selectional restrictions are not always available or accurate. The generation system we use, PENMAN (Penman, 1989), is robust because it supplies appropriate defaults when knowledge is missing. But the default choices frequently are not the optimM ones; the hybrid model we describe provides more satisfactory solutions.</Paragraph>
  </Section>
class="xml-element"></Paper>